AI Research Engineer
For engineers aiming at frontier AI research. Build math, ML and systems depth, then train, align, scale and evaluate models and publish reproducible results.
{"nodes":[{"id":"title","type":"title","position":{"x":0,"y":0},"data":{"label":"AI Research Engineer"},"width":1688,"height":61,"style":{"width":1688,"height":61},"measured":{"width":1688,"height":61},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":0,"y":0}},{"id":"summary","type":"paragraph","position":{"x":404,"y":77},"data":{"label":"For engineers aiming at frontier AI research. Build math, ML and systems depth, then train, align, scale and evaluate models and publish reproducible results."},"width":880,"height":56,"style":{"width":880,"height":56},"measured":{"width":880,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":404,"y":77}},{"id":"meta","type":"paragraph","position":{"x":524,"y":147},"data":{"label":"10 STAGES · 42 TOPICS · 148 SUBTOPICS","style":{"fontSize":13,"fontFamily":"jetbrains","fontWeight":500,"color":"var(--color-fg-subtle)"}},"width":640,"height":21,"style":{"width":640,"height":21},"measured":{"width":640,"height":21},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":524,"y":147}},{"id":"stage-1","type":"section","position":{"x":0,"y":224},"data":{"label":"Mathematical Foundations","description":"Build the linear algebra, calculus, probability and information theory fluency that ML papers assume.","number":1},"width":824,"height":982,"style":{"width":824,"height":982},"measured":{"width":824,"height":982},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":0,"y":224}},{"id":"topic-1-1","type":"topic","position":{"x":28,"y":345},"data":{"label":"Linear Algebra","number":"1.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":345}},{"id":"sub-1-1-1","type":"subtopic","position":{"x":28,"y":413},"data":{"label":"Vectors, Matrices and Tensors","description":"Vector Spaces · Norms · Rank · Tensor Operations"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":413}},{"id":"sub-1-1-2","type":"subtopic","position":{"x":288,"y":413},"data":{"label":"Matrix Decompositions","description":"Eigendecomposition · SVD · PCA · Cholesky"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":413}},{"id":"sub-1-1-3","type":"subtopic","position":{"x":548,"y":413},"data":{"label":"Matrix Calculus","description":"Jacobians · Hessians · Vector-Jacobian Products"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":413}},{"id":"topic-1-2","type":"topic","position":{"x":28,"y":547},"data":{"label":"Calculus and Optimization","number":"1.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":547}},{"id":"sub-1-2-1","type":"subtopic","position":{"x":28,"y":615},"data":{"label":"Multivariate Calculus","description":"Partial Derivatives · Gradients · Chain Rule · Taylor Expansion"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":615}},{"id":"sub-1-2-2","type":"subtopic","position":{"x":288,"y":615},"data":{"label":"Convex Optimization","description":"Convexity · Duality · Lagrange Multipliers · KKT Conditions"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":615}},{"id":"sub-1-2-3","type":"subtopic","position":{"x":548,"y":615},"data":{"label":"Gradient-Based Optimizers","description":"SGD · Momentum · Adam · AdamW · Muon"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":615}},{"id":"sub-1-2-4","type":"subtopic","position":{"x":28,"y":712},"data":{"label":"Learning Rate Schedules","description":"Warmup · Cosine Decay · WSD Schedules"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":712}},{"id":"topic-1-3","type":"topic","position":{"x":28,"y":825},"data":{"label":"Probability and Statistics","number":"1.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":825}},{"id":"sub-1-3-1","type":"subtopic","position":{"x":28,"y":893},"data":{"label":"Probability Theory","description":"Bayes' Theorem · Expectations · Moments · Common Distributions"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":893}},{"id":"sub-1-3-2","type":"subtopic","position":{"x":288,"y":893},"data":{"label":"Estimation","description":"MLE · MAP · Estimator Bias"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":893}},{"id":"sub-1-3-3","type":"subtopic","position":{"x":548,"y":893},"data":{"label":"Hypothesis Testing","description":"p-values · Confidence Intervals · Bootstrapping · Multiple Comparisons"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":893}},{"id":"topic-1-4","type":"topic","position":{"x":28,"y":1025},"data":{"label":"Information Theory","number":"1.4"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1025}},{"id":"sub-1-4-1","type":"subtopic","position":{"x":28,"y":1093},"data":{"label":"Entropy and Cross-Entropy"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1093}},{"id":"sub-1-4-2","type":"subtopic","position":{"x":288,"y":1093},"data":{"label":"KL Divergence and Mutual Information"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":1093}},{"id":"sub-1-4-3","type":"subtopic","position":{"x":548,"y":1093},"data":{"label":"Compression and Perplexity","description":"Bits per Token · Minimum Description Length"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":1093}},{"id":"stage-2","type":"section","position":{"x":864,"y":224},"data":{"label":"Programming and Computer Science","description":"Write efficient, well-tested Python and understand the algorithms and systems your experiments run on.","number":2},"width":824,"height":982,"style":{"width":824,"height":982},"measured":{"width":824,"height":982},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":864,"y":224}},{"id":"topic-2-1","type":"topic","position":{"x":892,"y":345},"data":{"label":"Scientific Python","number":"2.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":345}},{"id":"sub-2-1-1","type":"subtopic","position":{"x":892,"y":413},"data":{"label":"Python Proficiency","description":"Typing · Generators · Decorators · Packaging"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":413}},{"id":"sub-2-1-2","type":"subtopic","position":{"x":1152,"y":413},"data":{"label":"NumPy and Vectorization","description":"Broadcasting · Advanced Indexing · einsum"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":413}},{"id":"sub-2-1-3","type":"subtopic","position":{"x":1412,"y":413},"data":{"label":"Data Handling","description":"pandas · Polars · Apache Arrow · Parquet"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":413}},{"id":"sub-2-1-4","type":"subtopic","position":{"x":892,"y":510},"data":{"label":"Plotting","description":"Matplotlib · Seaborn"},"width":248,"height":67,"style":{"width":248,"height":67},"measured":{"width":248,"height":67},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":510}},{"id":"topic-2-2","type":"topic","position":{"x":892,"y":605},"data":{"label":"Algorithms and Complexity","number":"2.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":605}},{"id":"sub-2-2-1","type":"subtopic","position":{"x":892,"y":673},"data":{"label":"Data Structures","description":"Trees · Graphs · Heaps · Hash Maps"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":673}},{"id":"sub-2-2-2","type":"subtopic","position":{"x":1152,"y":673},"data":{"label":"Algorithm Design","description":"Dynamic Programming · Graph Search · Divide and Conquer"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":673}},{"id":"sub-2-2-3","type":"subtopic","position":{"x":1412,"y":673},"data":{"label":"Complexity Analysis","description":"Big O · Time Complexity · Memory Complexity"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":673}},{"id":"topic-2-3","type":"topic","position":{"x":892,"y":786},"data":{"label":"Systems and Parallel Computing","number":"2.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":786}},{"id":"sub-2-3-1","type":"subtopic","position":{"x":892,"y":854},"data":{"label":"Computer Architecture","description":"Memory Hierarchy · Caches · Bandwidth"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":854}},{"id":"sub-2-3-2","type":"subtopic","position":{"x":1152,"y":854},"data":{"label":"Concurrency and Parallelism","description":"Threads · Processes · Async · SIMD"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":854}},{"id":"sub-2-3-3","type":"subtopic","position":{"x":1412,"y":854},"data":{"label":"Distributed Systems Basics","description":"Networking · Collective Communication · Fault Tolerance"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":854}},{"id":"topic-2-4","type":"topic","position":{"x":892,"y":988},"data":{"label":"Research Software Engineering","number":"2.4"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":988}},{"id":"sub-2-4-1","type":"subtopic","position":{"x":892,"y":1056},"data":{"label":"Version Control","description":"Git · Branching · Code Review"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":1056}},{"id":"sub-2-4-2","type":"subtopic","position":{"x":1152,"y":1056},"data":{"label":"Testing and Debugging","description":"pytest · Debuggers · Profilers"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":1056}},{"id":"sub-2-4-3","type":"subtopic","position":{"x":1412,"y":1056},"data":{"label":"Reproducible Environments","description":"uv · Conda · Docker · Random Seeds"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":1056}},{"id":"stage-3","type":"section","position":{"x":0,"y":1246},"data":{"label":"Core Machine Learning","description":"Master classical learning algorithms and the theory that explains when and why models generalize.","number":3},"width":824,"height":1242,"style":{"width":824,"height":1242},"measured":{"width":824,"height":1242},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":0,"y":1246}},{"id":"topic-3-1","type":"topic","position":{"x":28,"y":1367},"data":{"label":"Supervised Learning","number":"3.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1367}},{"id":"sub-3-1-1","type":"subtopic","position":{"x":28,"y":1435},"data":{"label":"Linear Models","description":"Linear Regression · Logistic Regression · Regularization"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1435}},{"id":"sub-3-1-2","type":"subtopic","position":{"x":288,"y":1435},"data":{"label":"Kernel Methods","description":"Kernel Trick · SVMs · Gaussian Processes"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":1435}},{"id":"sub-3-1-3","type":"subtopic","position":{"x":548,"y":1435},"data":{"label":"Tree Ensembles","description":"Random Forests · XGBoost · LightGBM"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":1435}},{"id":"topic-3-2","type":"topic","position":{"x":28,"y":1548},"data":{"label":"Unsupervised Learning","number":"3.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1548}},{"id":"sub-3-2-1","type":"subtopic","position":{"x":28,"y":1616},"data":{"label":"Clustering","description":"K-Means · DBSCAN · Gaussian Mixtures · EM Algorithm"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1616}},{"id":"sub-3-2-2","type":"subtopic","position":{"x":288,"y":1616},"data":{"label":"Dimensionality Reduction","description":"PCA · t-SNE · UMAP"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":1616}},{"id":"topic-3-3","type":"topic","position":{"x":28,"y":1729},"data":{"label":"Model Evaluation","number":"3.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1729}},{"id":"sub-3-3-1","type":"subtopic","position":{"x":28,"y":1797},"data":{"label":"Data Splits and Cross-Validation","description":"Train-Val-Test · K-Fold · Data Leakage"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1797}},{"id":"sub-3-3-2","type":"subtopic","position":{"x":288,"y":1797},"data":{"label":"Metrics","description":"Accuracy · F1 · ROC-AUC · Calibration"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":1797}},{"id":"sub-3-3-3","type":"subtopic","position":{"x":548,"y":1797},"data":{"label":"Bias-Variance Tradeoff","description":"Underfitting · Overfitting · Model Capacity"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":1797}},{"id":"topic-3-4","type":"topic","position":{"x":28,"y":1931},"data":{"label":"Statistical Learning Theory","number":"3.4"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1931}},{"id":"sub-3-4-1","type":"subtopic","position":{"x":28,"y":1999},"data":{"label":"PAC Learning and VC Dimension"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":1999}},{"id":"sub-3-4-2","type":"subtopic","position":{"x":288,"y":1999},"data":{"label":"Generalization Bounds","description":"Rademacher Complexity · PAC-Bayes"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":1999}},{"id":"sub-3-4-3","type":"subtopic","position":{"x":548,"y":1999},"data":{"label":"Empirical Risk Minimization","description":"ERM · Structural Risk · Regularization"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":1999}},{"id":"stage-4","type":"section","position":{"x":864,"y":1246},"data":{"label":"Deep Learning Foundations","description":"Understand how neural networks are built, trained and implemented, from backpropagation to the Transformer.","number":4},"width":824,"height":1242,"style":{"width":824,"height":1242},"measured":{"width":824,"height":1242},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":864,"y":1246}},{"id":"topic-4-1","type":"topic","position":{"x":892,"y":1389},"data":{"label":"Neural Network Basics","number":"4.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":1389}},{"id":"sub-4-1-1","type":"subtopic","position":{"x":892,"y":1457},"data":{"label":"Perceptrons and MLPs","description":"Activation Functions · GELU · SiLU · Universal Approximation"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":1457}},{"id":"sub-4-1-2","type":"subtopic","position":{"x":1152,"y":1457},"data":{"label":"Backpropagation","description":"Computational Graphs · Autodiff · Chain Rule"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":1457}},{"id":"sub-4-1-3","type":"subtopic","position":{"x":1412,"y":1457},"data":{"label":"Loss Functions","description":"Cross-Entropy · MSE · Contrastive Losses"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":1457}},{"id":"topic-4-2","type":"topic","position":{"x":892,"y":1570},"data":{"label":"Training Deep Networks","number":"4.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":1570}},{"id":"sub-4-2-1","type":"subtopic","position":{"x":892,"y":1638},"data":{"label":"Initialization","description":"Xavier · Kaiming · muP"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":1638}},{"id":"sub-4-2-2","type":"subtopic","position":{"x":1152,"y":1638},"data":{"label":"Normalization Layers","description":"BatchNorm · LayerNorm · RMSNorm · GroupNorm"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":1638}},{"id":"sub-4-2-3","type":"subtopic","position":{"x":1412,"y":1638},"data":{"label":"Regularization","description":"Dropout · Weight Decay · Data Augmentation · Label Smoothing"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":1638}},{"id":"sub-4-2-4","type":"subtopic","position":{"x":892,"y":1735},"data":{"label":"Training Stability","description":"Gradient Clipping · Residual Connections · Loss Spikes"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":1735}},{"id":"topic-4-3","type":"topic","position":{"x":892,"y":1848},"data":{"label":"Deep Learning Frameworks","number":"4.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":1848}},{"id":"sub-4-3-1","type":"subtopic","position":{"x":892,"y":1916},"data":{"label":"PyTorch Fundamentals","description":"Tensors · Autograd · nn.Module · DataLoader"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":1916}},{"id":"sub-4-3-2","type":"subtopic","position":{"x":1152,"y":1916},"data":{"label":"Advanced PyTorch","description":"torch.compile · Hooks · Custom Autograd · Profiler"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":1916}},{"id":"sub-4-3-3","type":"subtopic","position":{"x":1412,"y":1916},"data":{"label":"JAX Ecosystem","description":"jit · vmap · shard_map · Flax NNX · Optax"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":1916}},{"id":"topic-4-4","type":"topic","position":{"x":892,"y":2029},"data":{"label":"Core Architectures","number":"4.4"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2029}},{"id":"sub-4-4-1","type":"subtopic","position":{"x":892,"y":2097},"data":{"label":"Convolutional Networks","description":"Convolutions · Pooling · Receptive Fields"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2097}},{"id":"sub-4-4-2","type":"subtopic","position":{"x":1152,"y":2097},"data":{"label":"Recurrent Networks","description":"RNN · LSTM · GRU"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":2097}},{"id":"sub-4-4-3","type":"subtopic","position":{"x":1412,"y":2097},"data":{"label":"Attention Mechanisms","description":"Self-Attention · Cross-Attention · Multi-Head Attention"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":2097}},{"id":"sub-4-4-4","type":"subtopic","position":{"x":892,"y":2194},"data":{"label":"The Transformer","description":"Positional Encodings · RoPE · KV Cache · Decoder-Only"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2194}},{"id":"topic-4-5","type":"topic","position":{"x":892,"y":2307},"data":{"label":"Deep Learning Theory","number":"4.5"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2307}},{"id":"sub-4-5-1","type":"subtopic","position":{"x":892,"y":2375},"data":{"label":"Neural Tangent Kernel","description":"Infinite Width · Lazy Training · Feature Learning"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2375}},{"id":"sub-4-5-2","type":"subtopic","position":{"x":1152,"y":2375},"data":{"label":"Generalization Phenomena","description":"Double Descent · Grokking · Lottery Tickets"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":2375}},{"id":"sub-4-5-3","type":"subtopic","position":{"x":1412,"y":2375},"data":{"label":"Loss Landscape Geometry","description":"Sharpness · Mode Connectivity · Edge of Stability"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":2375}},{"id":"stage-5","type":"section","position":{"x":0,"y":2528},"data":{"label":"Deep Learning Domains","description":"Apply deep learning across vision, language, multimodal and graph data, the main arenas of modern research.","number":5},"width":824,"height":1067,"style":{"width":824,"height":1067},"measured":{"width":824,"height":1067},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":0,"y":2528}},{"id":"topic-5-1","type":"topic","position":{"x":28,"y":2671},"data":{"label":"Computer Vision","number":"5.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":2671}},{"id":"sub-5-1-1","type":"subtopic","position":{"x":28,"y":2739},"data":{"label":"Vision Backbones","description":"ResNet · ConvNeXt · ViT · Swin"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":2739}},{"id":"sub-5-1-2","type":"subtopic","position":{"x":288,"y":2739},"data":{"label":"Detection and Segmentation","description":"DETR · Mask R-CNN · SAM 2"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":2739}},{"id":"sub-5-1-3","type":"subtopic","position":{"x":548,"y":2739},"data":{"label":"3D Vision","description":"NeRF · 3D Gaussian Splatting · Depth Estimation"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":2739}},{"id":"topic-5-2","type":"topic","position":{"x":28,"y":2852},"data":{"label":"Natural Language Processing","number":"5.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":2852}},{"id":"sub-5-2-1","type":"subtopic","position":{"x":28,"y":2920},"data":{"label":"Tokenization","description":"BPE · WordPiece · SentencePiece · Vocabulary Design"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":2920}},{"id":"sub-5-2-2","type":"subtopic","position":{"x":288,"y":2920},"data":{"label":"Word Embeddings","description":"Word2Vec · GloVe · Contextual Embeddings"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":2920}},{"id":"sub-5-2-3","type":"subtopic","position":{"x":548,"y":2920},"data":{"label":"Language Modeling","description":"Perplexity · Masked LM · Causal LM"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":2920}},{"id":"topic-5-3","type":"topic","position":{"x":28,"y":3052},"data":{"label":"Self-Supervised and Multimodal Learning","number":"5.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3052}},{"id":"sub-5-3-1","type":"subtopic","position":{"x":28,"y":3120},"data":{"label":"Contrastive Learning","description":"SimCLR · CLIP · SigLIP"},"width":248,"height":87,"style":{"width":248,"height":87},"measured":{"width":248,"height":87},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3120}},{"id":"sub-5-3-2","type":"subtopic","position":{"x":288,"y":3120},"data":{"label":"Masked and Joint-Embedding Methods","description":"MAE · DINOv2 · I-JEPA"},"width":248,"height":87,"style":{"width":248,"height":87},"measured":{"width":248,"height":87},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":3120}},{"id":"sub-5-3-3","type":"subtopic","position":{"x":548,"y":3120},"data":{"label":"Vision-Language Models","description":"Cross-Modal Attention · LLaVA · Qwen-VL"},"width":248,"height":87,"style":{"width":248,"height":87},"measured":{"width":248,"height":87},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":3120}},{"id":"sub-5-3-4","type":"subtopic","position":{"x":28,"y":3219},"data":{"label":"Any-to-Any Multimodality","description":"Audio · Video · Image · Text"},"width":248,"height":67,"style":{"width":248,"height":67},"measured":{"width":248,"height":67},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3219}},{"id":"topic-5-4","type":"topic","position":{"x":28,"y":3314},"data":{"label":"Graph and Geometric Deep Learning","number":"5.4"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3314}},{"id":"sub-5-4-1","type":"subtopic","position":{"x":28,"y":3382},"data":{"label":"Message Passing Networks","description":"GCN · GraphSAGE · GIN"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3382}},{"id":"sub-5-4-2","type":"subtopic","position":{"x":288,"y":3382},"data":{"label":"Graph Attention and Transformers","description":"GAT · Graph Transformers · Positional Encodings"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":3382}},{"id":"sub-5-4-3","type":"subtopic","position":{"x":548,"y":3382},"data":{"label":"Equivariant Networks","description":"Symmetry Groups · E(3) Equivariance · Geometric Priors"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":3382}},{"id":"sub-5-4-4","type":"subtopic","position":{"x":28,"y":3500},"data":{"label":"Physics-Informed ML","description":"PINNs · Neural Operators"},"width":248,"height":67,"style":{"width":248,"height":67},"measured":{"width":248,"height":67},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3500}},{"id":"stage-6","type":"section","position":{"x":864,"y":2528},"data":{"label":"Generative Models and Reinforcement Learning","description":"Learn the generative and decision-making frameworks behind modern foundation models and post-training.","number":6},"width":824,"height":1067,"style":{"width":824,"height":1067},"measured":{"width":824,"height":1067},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":864,"y":2528}},{"id":"topic-6-1","type":"topic","position":{"x":892,"y":2649},"data":{"label":"Generative Models","number":"6.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2649}},{"id":"sub-6-1-1","type":"subtopic","position":{"x":892,"y":2717},"data":{"label":"VAEs and GANs","description":"ELBO · Reparameterization · Adversarial Training"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2717}},{"id":"sub-6-1-2","type":"subtopic","position":{"x":1152,"y":2717},"data":{"label":"Diffusion Models","description":"DDPM · Score Matching · SDEs · Latent Diffusion"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":2717}},{"id":"sub-6-1-3","type":"subtopic","position":{"x":1412,"y":2717},"data":{"label":"Flow Matching","description":"Rectified Flow · Normalizing Flows · Optimal Transport"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":2717}},{"id":"sub-6-1-4","type":"subtopic","position":{"x":892,"y":2814},"data":{"label":"Autoregressive Generation","description":"Next-Token Prediction · VQ-VAE · Discrete Tokenizers"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2814}},{"id":"topic-6-2","type":"topic","position":{"x":892,"y":2927},"data":{"label":"Reinforcement Learning Fundamentals","number":"6.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2927}},{"id":"sub-6-2-1","type":"subtopic","position":{"x":892,"y":2995},"data":{"label":"Markov Decision Processes","description":"Bellman Equations · Value Functions · Dynamic Programming"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":2995}},{"id":"sub-6-2-2","type":"subtopic","position":{"x":1152,"y":2995},"data":{"label":"Value-Based Methods","description":"TD Learning · Q-Learning · DQN"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":2995}},{"id":"sub-6-2-3","type":"subtopic","position":{"x":1412,"y":2995},"data":{"label":"Policy Gradient Methods","description":"REINFORCE · Actor-Critic · PPO · SAC"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":2995}},{"id":"topic-6-3","type":"topic","position":{"x":892,"y":3127},"data":{"label":"Advanced Reinforcement Learning","number":"6.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":3127}},{"id":"sub-6-3-1","type":"subtopic","position":{"x":892,"y":3195},"data":{"label":"Offline RL","description":"CQL · IQL · Decision Transformer"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":3195}},{"id":"sub-6-3-2","type":"subtopic","position":{"x":1152,"y":3195},"data":{"label":"Model-Based RL","description":"World Models · MuZero · Dreamer"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":3195}},{"id":"sub-6-3-3","type":"subtopic","position":{"x":1412,"y":3195},"data":{"label":"Multi-Agent and Hierarchical RL","description":"Self-Play · Options · Cooperative Agents"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":3195}},{"id":"stage-7","type":"section","position":{"x":0,"y":3635},"data":{"label":"Large Language Models","description":"Understand how frontier LLMs are pre-trained, aligned, extended with reasoning and tools, and made safe.","number":7},"width":824,"height":1675,"style":{"width":824,"height":1675},"measured":{"width":824,"height":1675},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":0,"y":3635}},{"id":"topic-7-1","type":"topic","position":{"x":28,"y":3756},"data":{"label":"Pre-training Data","number":"7.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3756}},{"id":"sub-7-1-1","type":"subtopic","position":{"x":28,"y":3824},"data":{"label":"Web-Scale Data Pipelines","description":"Crawling · Deduplication · Quality Filtering · PII Removal"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3824}},{"id":"sub-7-1-2","type":"subtopic","position":{"x":288,"y":3824},"data":{"label":"Data Mixtures and Curricula","description":"Domain Weighting · Annealing · Code and Math"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":3824}},{"id":"sub-7-1-3","type":"subtopic","position":{"x":548,"y":3824},"data":{"label":"Synthetic Data","description":"Self-Instruct · Rejection Sampling · Distillation"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":3824}},{"id":"sub-7-1-4","type":"subtopic","position":{"x":28,"y":3921},"data":{"label":"Benchmark Contamination","description":"Leakage Detection · N-gram Overlap · Decontamination"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":3921}},{"id":"topic-7-2","type":"topic","position":{"x":28,"y":4034},"data":{"label":"Pre-training and Scaling","number":"7.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4034}},{"id":"sub-7-2-1","type":"subtopic","position":{"x":28,"y":4102},"data":{"label":"Training Objectives","description":"Causal LM · Masked LM · Multi-Token Prediction"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4102}},{"id":"sub-7-2-2","type":"subtopic","position":{"x":288,"y":4102},"data":{"label":"Scaling Laws","description":"Compute-Optimal Training · Chinchilla · Data-Constrained Scaling"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":4102}},{"id":"sub-7-2-3","type":"subtopic","position":{"x":548,"y":4102},"data":{"label":"Long Context","description":"RoPE Scaling · Context Extension · Sparse Attention"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":4102}},{"id":"sub-7-2-4","type":"subtopic","position":{"x":28,"y":4218},"data":{"label":"In-Context Learning","description":"Few-Shot Prompting · Emergent Abilities"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4218}},{"id":"topic-7-3","type":"topic","position":{"x":28,"y":4331},"data":{"label":"Emerging Architectures","number":"7.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4331}},{"id":"sub-7-3-1","type":"subtopic","position":{"x":28,"y":4399},"data":{"label":"Mixture of Experts","description":"Routing · Load Balancing · Expert Parallelism"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4399}},{"id":"sub-7-3-2","type":"subtopic","position":{"x":288,"y":4399},"data":{"label":"State Space Models","description":"Mamba · Linear Attention · Hybrid Models"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":4399}},{"id":"sub-7-3-3","type":"subtopic","position":{"x":548,"y":4399},"data":{"label":"Attention Variants","description":"MQA · GQA · MLA · Sliding Window"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":4399}},{"id":"topic-7-4","type":"topic","position":{"x":28,"y":4512},"data":{"label":"Post-Training and Alignment","number":"7.4"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4512}},{"id":"sub-7-4-1","type":"subtopic","position":{"x":28,"y":4580},"data":{"label":"Supervised Fine-Tuning","description":"Instruction Tuning · Chat Templates · Data Quality"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4580}},{"id":"sub-7-4-2","type":"subtopic","position":{"x":288,"y":4580},"data":{"label":"RLHF and RLAIF","description":"Reward Modeling · PPO · Constitutional AI"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":4580}},{"id":"sub-7-4-3","type":"subtopic","position":{"x":548,"y":4580},"data":{"label":"Preference Optimization","description":"DPO · KTO · ORPO · SimPO"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":4580}},{"id":"sub-7-4-4","type":"subtopic","position":{"x":28,"y":4677},"data":{"label":"RL with Verifiable Rewards","description":"GRPO · RLVR · Process Reward Models"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4677}},{"id":"topic-7-5","type":"topic","position":{"x":28,"y":4790},"data":{"label":"Reasoning, Agents and Tool Use","number":"7.5"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4790}},{"id":"sub-7-5-1","type":"subtopic","position":{"x":28,"y":4858},"data":{"label":"Chain-of-Thought Reasoning","description":"Self-Consistency · Tree of Thoughts · Reasoning Models"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4858}},{"id":"sub-7-5-2","type":"subtopic","position":{"x":288,"y":4858},"data":{"label":"Test-Time Compute","description":"Best-of-N · Search · Verifiers · Budget Forcing"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":4858}},{"id":"sub-7-5-3","type":"subtopic","position":{"x":548,"y":4858},"data":{"label":"Retrieval-Augmented Generation","description":"Embeddings · Vector Search · Chunking · Re-ranking"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":4858}},{"id":"sub-7-5-4","type":"subtopic","position":{"x":28,"y":4976},"data":{"label":"Tool Use and Function Calling","description":"Structured Outputs · MCP · Code Execution"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":4976}},{"id":"sub-7-5-5","type":"subtopic","position":{"x":288,"y":4976},"data":{"label":"Agentic Systems","description":"ReAct · Planning · Reflection · Multi-Agent"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":4976}},{"id":"topic-7-6","type":"topic","position":{"x":28,"y":5110},"data":{"label":"Interpretability and Safety","number":"7.6"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":5110}},{"id":"sub-7-6-1","type":"subtopic","position":{"x":28,"y":5178},"data":{"label":"Mechanistic Interpretability","description":"Circuits · Probing · Sparse Autoencoders · Activation Patching"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":5178}},{"id":"sub-7-6-2","type":"subtopic","position":{"x":288,"y":5178},"data":{"label":"Red Teaming","description":"Jailbreaks · Prompt Injection · Adversarial Attacks"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":5178}},{"id":"sub-7-6-3","type":"subtopic","position":{"x":548,"y":5178},"data":{"label":"Alignment Research","description":"Reward Hacking · Scalable Oversight · Honesty"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":5178}},{"id":"stage-8","type":"section","position":{"x":864,"y":3635},"data":{"label":"Infrastructure and Efficiency","description":"Train and serve models at scale by mastering GPUs, distributed training and efficient inference.","number":8},"width":824,"height":1675,"style":{"width":824,"height":1675},"measured":{"width":824,"height":1675},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":864,"y":3635}},{"id":"topic-8-1","type":"topic","position":{"x":892,"y":3756},"data":{"label":"GPU Programming","number":"8.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":3756}},{"id":"sub-8-1-1","type":"subtopic","position":{"x":892,"y":3824},"data":{"label":"GPU and TPU Architecture","description":"SMs · HBM · Tensor Cores · Memory Bandwidth"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":3824}},{"id":"sub-8-1-2","type":"subtopic","position":{"x":1152,"y":3824},"data":{"label":"CUDA Basics","description":"Threads · Blocks · Shared Memory · Kernels"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":3824}},{"id":"sub-8-1-3","type":"subtopic","position":{"x":1412,"y":3824},"data":{"label":"Custom Kernels","description":"Triton · Kernel Fusion · CUTLASS"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":3824}},{"id":"sub-8-1-4","type":"subtopic","position":{"x":892,"y":3921},"data":{"label":"Performance Analysis","description":"Roofline Model · Nsight · PyTorch Profiler"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":3921}},{"id":"topic-8-2","type":"topic","position":{"x":892,"y":4034},"data":{"label":"Distributed Training","number":"8.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4034}},{"id":"sub-8-2-1","type":"subtopic","position":{"x":892,"y":4102},"data":{"label":"Data Parallelism","description":"DDP · FSDP2 · ZeRO"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4102}},{"id":"sub-8-2-2","type":"subtopic","position":{"x":1152,"y":4102},"data":{"label":"Model Parallelism","description":"Tensor · Pipeline · Sequence · Expert"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":4102}},{"id":"sub-8-2-3","type":"subtopic","position":{"x":1412,"y":4102},"data":{"label":"Training Frameworks","description":"Megatron-LM · DeepSpeed · TorchTitan"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":4102}},{"id":"sub-8-2-4","type":"subtopic","position":{"x":892,"y":4199},"data":{"label":"Cluster Computing","description":"Slurm · Kubernetes · NCCL · Fault-Tolerant Checkpointing"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4199}},{"id":"topic-8-3","type":"topic","position":{"x":892,"y":4312},"data":{"label":"Memory and Compute Efficiency","number":"8.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4312}},{"id":"sub-8-3-1","type":"subtopic","position":{"x":892,"y":4380},"data":{"label":"Mixed Precision","description":"BF16 · FP8 · Loss Scaling"},"width":248,"height":67,"style":{"width":248,"height":67},"measured":{"width":248,"height":67},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4380}},{"id":"sub-8-3-2","type":"subtopic","position":{"x":1152,"y":4380},"data":{"label":"Activation Checkpointing and Offloading"},"width":248,"height":67,"style":{"width":248,"height":67},"measured":{"width":248,"height":67},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":4380}},{"id":"sub-8-3-3","type":"subtopic","position":{"x":1412,"y":4380},"data":{"label":"Efficient Attention","description":"FlashAttention · Ring Attention"},"width":248,"height":67,"style":{"width":248,"height":67},"measured":{"width":248,"height":67},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":4380}},{"id":"sub-8-3-4","type":"subtopic","position":{"x":892,"y":4459},"data":{"label":"Parameter-Efficient Fine-Tuning","description":"LoRA · QLoRA · Adapters"},"width":248,"height":87,"style":{"width":248,"height":87},"measured":{"width":248,"height":87},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4459}},{"id":"topic-8-4","type":"topic","position":{"x":892,"y":4574},"data":{"label":"Efficient Inference","number":"8.4"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4574}},{"id":"sub-8-4-1","type":"subtopic","position":{"x":892,"y":4642},"data":{"label":"Inference Engines","description":"vLLM · SGLang · TensorRT-LLM"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4642}},{"id":"sub-8-4-2","type":"subtopic","position":{"x":1152,"y":4642},"data":{"label":"Serving Optimizations","description":"PagedAttention · Continuous Batching · Speculative Decoding"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":4642}},{"id":"sub-8-4-3","type":"subtopic","position":{"x":1412,"y":4642},"data":{"label":"Quantization","description":"GPTQ · AWQ · FP8 · INT4"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":4642}},{"id":"sub-8-4-4","type":"subtopic","position":{"x":892,"y":4739},"data":{"label":"Pruning and Distillation","description":"Structured Pruning · Knowledge Distillation"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4739}},{"id":"topic-8-5","type":"topic","position":{"x":892,"y":4852},"data":{"label":"Experiment Management","number":"8.5"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4852}},{"id":"sub-8-5-1","type":"subtopic","position":{"x":892,"y":4920},"data":{"label":"Experiment Tracking","description":"Weights & Biases · MLflow · TensorBoard"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":4920}},{"id":"sub-8-5-2","type":"subtopic","position":{"x":1152,"y":4920},"data":{"label":"Checkpoints and Artifacts","description":"Model Versioning · Hugging Face Hub · Resumable Training"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":4920}},{"id":"sub-8-5-3","type":"subtopic","position":{"x":1412,"y":4920},"data":{"label":"Configuration and Sweeps","description":"Hydra · Hyperparameter Search · Sweep Tools"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":4920}},{"id":"stage-9","type":"section","position":{"x":0,"y":5349},"data":{"label":"Research Methodology","description":"Run the full research loop: read critically, design rigorous experiments, evaluate honestly and communicate.","number":9},"width":824,"height":1273,"style":{"width":824,"height":1273},"measured":{"width":824,"height":1273},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":0,"y":5349}},{"id":"topic-9-1","type":"topic","position":{"x":28,"y":5470},"data":{"label":"Reading and Synthesizing Papers","number":"9.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":5470}},{"id":"sub-9-1-1","type":"subtopic","position":{"x":28,"y":5538},"data":{"label":"Tracking the Literature","description":"arXiv · OpenReview · Conference Proceedings · Semantic Scholar"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":5538}},{"id":"sub-9-1-2","type":"subtopic","position":{"x":288,"y":5538},"data":{"label":"Reading Papers Critically","description":"Claims · Assumptions · Baselines · Limitations"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":5538}},{"id":"sub-9-1-3","type":"subtopic","position":{"x":548,"y":5538},"data":{"label":"Reproducing Papers","description":"Reimplementation · Matching Baselines · Reporting Discrepancies"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":5538}},{"id":"sub-9-1-4","type":"subtopic","position":{"x":28,"y":5654},"data":{"label":"Formulating Hypotheses","description":"Identifying Gaps · Research Questions"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":5654}},{"id":"topic-9-2","type":"topic","position":{"x":28,"y":5767},"data":{"label":"Experiment Design","number":"9.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":5767}},{"id":"sub-9-2-1","type":"subtopic","position":{"x":28,"y":5835},"data":{"label":"Compute Budgeting","description":"Small-Scale Proxies · Scaling Experiments"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":5835}},{"id":"sub-9-2-2","type":"subtopic","position":{"x":288,"y":5835},"data":{"label":"Baselines and Ablations","description":"Strong Baselines · Controlled Comparisons"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":5835}},{"id":"sub-9-2-3","type":"subtopic","position":{"x":548,"y":5835},"data":{"label":"Statistical Rigor","description":"Multiple Seeds · Confidence Intervals · Significance Tests"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":5835}},{"id":"sub-9-2-4","type":"subtopic","position":{"x":28,"y":5932},"data":{"label":"Rapid Prototyping","description":"Sanity Checks · Overfitting Tiny Batches · Debugging Training"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":5932}},{"id":"topic-9-3","type":"topic","position":{"x":28,"y":6045},"data":{"label":"Evaluation and Benchmarking","number":"9.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":6045}},{"id":"sub-9-3-1","type":"subtopic","position":{"x":28,"y":6113},"data":{"label":"Static Benchmarks","description":"MMLU-Pro · GPQA · SWE-bench Verified · Benchmark Saturation"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":6113}},{"id":"sub-9-3-2","type":"subtopic","position":{"x":288,"y":6113},"data":{"label":"LLM-as-a-Judge","description":"Pairwise Comparison · Rubrics · Judge Bias"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":6113}},{"id":"sub-9-3-3","type":"subtopic","position":{"x":548,"y":6113},"data":{"label":"Human Preference Evaluation","description":"LMArena · Elo Ratings · Crowdsourcing"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":6113}},{"id":"sub-9-3-4","type":"subtopic","position":{"x":28,"y":6231},"data":{"label":"Open-Ended Evaluation","description":"Long Context · Tool Use · Agentic Tasks"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":6231}},{"id":"sub-9-3-5","type":"subtopic","position":{"x":288,"y":6231},"data":{"label":"Evaluation Harnesses","description":"lm-evaluation-harness · Inspect · HELM"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":6231}},{"id":"topic-9-4","type":"topic","position":{"x":28,"y":6344},"data":{"label":"Scientific Communication","number":"9.4"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":6344}},{"id":"sub-9-4-1","type":"subtopic","position":{"x":28,"y":6412},"data":{"label":"Writing Papers","description":"Structure · Related Work · NeurIPS · ICLR · ICML"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":6412}},{"id":"sub-9-4-2","type":"subtopic","position":{"x":288,"y":6412},"data":{"label":"Figures and Visualization","description":"Plots · Diagrams · Tables"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":288,"y":6412}},{"id":"sub-9-4-3","type":"subtopic","position":{"x":548,"y":6412},"data":{"label":"Peer Review","description":"Reviewing · Rebuttals · Workshops"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":548,"y":6412}},{"id":"sub-9-4-4","type":"subtopic","position":{"x":28,"y":6509},"data":{"label":"Talks and Posters","description":"Conference Talks · Poster Sessions"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":28,"y":6509}},{"id":"stage-10","type":"section","position":{"x":864,"y":5349},"data":{"label":"Capstone and Career","description":"Turn skills into a research portfolio through projects, open-source work and targeted interview preparation.","number":10},"width":824,"height":1273,"style":{"width":824,"height":1273},"measured":{"width":824,"height":1273},"zIndex":-999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":864,"y":5349}},{"id":"topic-10-1","type":"topic","position":{"x":892,"y":5470},"data":{"label":"Capstone Projects","number":"10.1"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":5470}},{"id":"sub-10-1-1","type":"subtopic","position":{"x":892,"y":5538},"data":{"label":"Train a Small LLM from Scratch","description":"Tokenizer · Pre-training · SFT · Evaluation"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":5538}},{"id":"sub-10-1-2","type":"subtopic","position":{"x":1152,"y":5538},"data":{"label":"Post-Training Project","description":"Preference Data · DPO · GRPO · Evaluation"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":5538}},{"id":"sub-10-1-3","type":"subtopic","position":{"x":1412,"y":5538},"data":{"label":"Systems Project","description":"Custom Kernel · Distributed Training · Inference Optimization"},"width":248,"height":106,"style":{"width":248,"height":106},"measured":{"width":248,"height":106},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":5538}},{"id":"sub-10-1-4","type":"subtopic","position":{"x":892,"y":5656},"data":{"label":"Novel Research Project","description":"Hypothesis · Experiments · Workshop Paper"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":5656}},{"id":"topic-10-2","type":"topic","position":{"x":892,"y":5769},"data":{"label":"Open Source and Portfolio","number":"10.2"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":5769}},{"id":"sub-10-2-1","type":"subtopic","position":{"x":892,"y":5837},"data":{"label":"Releasing Code and Models","description":"Hugging Face Hub · Model Cards · Licenses"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":5837}},{"id":"sub-10-2-2","type":"subtopic","position":{"x":1152,"y":5837},"data":{"label":"Contributing to Open Source","description":"PyTorch · Transformers · vLLM"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":5837}},{"id":"sub-10-2-3","type":"subtopic","position":{"x":1412,"y":5837},"data":{"label":"Technical Writing","description":"Blog Posts · Paper Explainers · Research Notes"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":5837}},{"id":"topic-10-3","type":"topic","position":{"x":892,"y":5950},"data":{"label":"Interviews and Career Growth","number":"10.3"},"width":768,"height":56,"style":{"width":768,"height":56},"measured":{"width":768,"height":56},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":5950}},{"id":"sub-10-3-1","type":"subtopic","position":{"x":892,"y":6018},"data":{"label":"Coding Interviews","description":"Algorithms · ML Coding · Implementing Attention"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":6018}},{"id":"sub-10-3-2","type":"subtopic","position":{"x":1152,"y":6018},"data":{"label":"ML Depth Interviews","description":"Paper Discussions · Research Proposals · Debugging Scenarios"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1152,"y":6018}},{"id":"sub-10-3-3","type":"subtopic","position":{"x":1412,"y":6018},"data":{"label":"Research Presentations","description":"Job Talk · Past Work Deep Dive"},"width":248,"height":104,"style":{"width":248,"height":104},"measured":{"width":248,"height":104},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":1412,"y":6018}},{"id":"sub-10-3-4","type":"subtopic","position":{"x":892,"y":6134},"data":{"label":"Career Paths","description":"Residencies · PhD Programs · Industry Labs · Startups"},"width":248,"height":85,"style":{"width":248,"height":85},"measured":{"width":248,"height":85},"zIndex":999,"selected":false,"selectable":true,"focusable":true,"dragging":false,"resizing":false,"positionAbsolute":{"x":892,"y":6134}}],"edges":[]}
01 Mathematical Foundations Build the linear algebra, calculus, probability and information theory fluency that ML papers assume. 02 Programming and Computer Science Write efficient, well-tested Python and understand the algorithms and systems your experiments run on. 03 Core Machine Learning Master classical learning algorithms and the theory that explains when and why models generalize. 04 Deep Learning Foundations Understand how neural networks are built, trained and implemented, from backpropagation to the Transformer. 05 Deep Learning Domains Apply deep learning across vision, language, multimodal and graph data, the main arenas of modern research. 06 Generative Models and Reinforcement Learning Learn the generative and decision-making frameworks behind modern foundation models and post-training. 07 Large Language Models Understand how frontier LLMs are pre-trained, aligned, extended with reasoning and tools, and made safe. 08 Infrastructure and Efficiency Train and serve models at scale by mastering GPUs, distributed training and efficient inference. 09 Research Methodology Run the full research loop: read critically, design rigorous experiments, evaluate honestly and communicate. 10 Capstone and Career Turn skills into a research portfolio through projects, open-source work and targeted interview preparation. AI Research Engineer For engineers aiming at frontier AI research. Build math, ML and systems depth, then train, align, scale and evaluate models and publish reproducible results. 10 STAGES · 42 TOPICS · 148 SUBTOPICS 1.1 Linear Algebra Vectors, Matrices and Tensors Vector Spaces · Norms · Rank · Tensor Operations Matrix Decompositions Eigendecomposition · SVD · PCA · Cholesky Matrix Calculus Jacobians · Hessians · Vector-Jacobian Products 1.2 Calculus and Optimization Multivariate Calculus Partial Derivatives · Gradients · Chain Rule · Taylor Expansion Convex Optimization Convexity · Duality · Lagrange Multipliers · KKT Conditions Gradient-Based Optimizers SGD · Momentum · Adam · AdamW · Muon Learning Rate Schedules Warmup · Cosine Decay · WSD Schedules 1.3 Probability and Statistics Probability Theory Bayes' Theorem · Expectations · Moments · Common Distributions Estimation MLE · MAP · Estimator Bias Hypothesis Testing p-values · Confidence Intervals · Bootstrapping · Multiple Comparisons 1.4 Information Theory Entropy and Cross-Entropy KL Divergence and Mutual Information Compression and Perplexity Bits per Token · Minimum Description Length 2.1 Scientific Python Python Proficiency Typing · Generators · Decorators · Packaging NumPy and Vectorization Broadcasting · Advanced Indexing · einsum Data Handling pandas · Polars · Apache Arrow · Parquet Plotting Matplotlib · Seaborn 2.2 Algorithms and Complexity Data Structures Trees · Graphs · Heaps · Hash Maps Algorithm Design Dynamic Programming · Graph Search · Divide and Conquer Complexity Analysis Big O · Time Complexity · Memory Complexity 2.3 Systems and Parallel Computing Computer Architecture Memory Hierarchy · Caches · Bandwidth Concurrency and Parallelism Threads · Processes · Async · SIMD Distributed Systems Basics Networking · Collective Communication · Fault Tolerance 2.4 Research Software Engineering Version Control Git · Branching · Code Review Testing and Debugging pytest · Debuggers · Profilers Reproducible Environments uv · Conda · Docker · Random Seeds 3.1 Supervised Learning Linear Models Linear Regression · Logistic Regression · Regularization Kernel Methods Kernel Trick · SVMs · Gaussian Processes Tree Ensembles Random Forests · XGBoost · LightGBM 3.2 Unsupervised Learning Clustering K-Means · DBSCAN · Gaussian Mixtures · EM Algorithm Dimensionality Reduction PCA · t-SNE · UMAP 3.3 Model Evaluation Data Splits and Cross-Validation Train-Val-Test · K-Fold · Data Leakage Metrics Accuracy · F1 · ROC-AUC · Calibration Bias-Variance Tradeoff Underfitting · Overfitting · Model Capacity 3.4 Statistical Learning Theory PAC Learning and VC Dimension Generalization Bounds Rademacher Complexity · PAC-Bayes Empirical Risk Minimization ERM · Structural Risk · Regularization 4.1 Neural Network Basics Perceptrons and MLPs Activation Functions · GELU · SiLU · Universal Approximation Backpropagation Computational Graphs · Autodiff · Chain Rule Loss Functions Cross-Entropy · MSE · Contrastive Losses 4.2 Training Deep Networks Initialization Xavier · Kaiming · muP Normalization Layers BatchNorm · LayerNorm · RMSNorm · GroupNorm Regularization Dropout · Weight Decay · Data Augmentation · Label Smoothing Training Stability Gradient Clipping · Residual Connections · Loss Spikes 4.3 Deep Learning Frameworks PyTorch Fundamentals Tensors · Autograd · nn.Module · DataLoader Advanced PyTorch torch.compile · Hooks · Custom Autograd · Profiler JAX Ecosystem jit · vmap · shard_map · Flax NNX · Optax 4.4 Core Architectures Convolutional Networks Convolutions · Pooling · Receptive Fields Recurrent Networks RNN · LSTM · GRU Attention Mechanisms Self-Attention · Cross-Attention · Multi-Head Attention The Transformer Positional Encodings · RoPE · KV Cache · Decoder-Only 4.5 Deep Learning Theory Neural Tangent Kernel Infinite Width · Lazy Training · Feature Learning Generalization Phenomena Double Descent · Grokking · Lottery Tickets Loss Landscape Geometry Sharpness · Mode Connectivity · Edge of Stability 5.1 Computer Vision Vision Backbones ResNet · ConvNeXt · ViT · Swin Detection and Segmentation DETR · Mask R-CNN · SAM 2 3D Vision NeRF · 3D Gaussian Splatting · Depth Estimation 5.2 Natural Language Processing Tokenization BPE · WordPiece · SentencePiece · Vocabulary Design Word Embeddings Word2Vec · GloVe · Contextual Embeddings Language Modeling Perplexity · Masked LM · Causal LM 5.3 Self-Supervised and Multimodal Learning Contrastive Learning SimCLR · CLIP · SigLIP Masked and Joint-Embedding Methods MAE · DINOv2 · I-JEPA Vision-Language Models Cross-Modal Attention · LLaVA · Qwen-VL Any-to-Any Multimodality Audio · Video · Image · Text 5.4 Graph and Geometric Deep Learning Message Passing Networks GCN · GraphSAGE · GIN Graph Attention and Transformers GAT · Graph Transformers · Positional Encodings Equivariant Networks Symmetry Groups · E(3) Equivariance · Geometric Priors Physics-Informed ML PINNs · Neural Operators 6.1 Generative Models VAEs and GANs ELBO · Reparameterization · Adversarial Training Diffusion Models DDPM · Score Matching · SDEs · Latent Diffusion Flow Matching Rectified Flow · Normalizing Flows · Optimal Transport Autoregressive Generation Next-Token Prediction · VQ-VAE · Discrete Tokenizers 6.2 Reinforcement Learning Fundamentals Markov Decision Processes Bellman Equations · Value Functions · Dynamic Programming Value-Based Methods TD Learning · Q-Learning · DQN Policy Gradient Methods REINFORCE · Actor-Critic · PPO · SAC 6.3 Advanced Reinforcement Learning Offline RL CQL · IQL · Decision Transformer Model-Based RL World Models · MuZero · Dreamer Multi-Agent and Hierarchical RL Self-Play · Options · Cooperative Agents 7.1 Pre-training Data Web-Scale Data Pipelines Crawling · Deduplication · Quality Filtering · PII Removal Data Mixtures and Curricula Domain Weighting · Annealing · Code and Math Synthetic Data Self-Instruct · Rejection Sampling · Distillation Benchmark Contamination Leakage Detection · N-gram Overlap · Decontamination 7.2 Pre-training and Scaling Training Objectives Causal LM · Masked LM · Multi-Token Prediction Scaling Laws Compute-Optimal Training · Chinchilla · Data-Constrained Scaling Long Context RoPE Scaling · Context Extension · Sparse Attention In-Context Learning Few-Shot Prompting · Emergent Abilities 7.3 Emerging Architectures Mixture of Experts Routing · Load Balancing · Expert Parallelism State Space Models Mamba · Linear Attention · Hybrid Models Attention Variants MQA · GQA · MLA · Sliding Window 7.4 Post-Training and Alignment Supervised Fine-Tuning Instruction Tuning · Chat Templates · Data Quality RLHF and RLAIF Reward Modeling · PPO · Constitutional AI Preference Optimization DPO · KTO · ORPO · SimPO RL with Verifiable Rewards GRPO · RLVR · Process Reward Models 7.5 Reasoning, Agents and Tool Use Chain-of-Thought Reasoning Self-Consistency · Tree of Thoughts · Reasoning Models Test-Time Compute Best-of-N · Search · Verifiers · Budget Forcing Retrieval-Augmented Generation Embeddings · Vector Search · Chunking · Re-ranking Tool Use and Function Calling Structured Outputs · MCP · Code Execution Agentic Systems ReAct · Planning · Reflection · Multi-Agent 7.6 Interpretability and Safety Mechanistic Interpretability Circuits · Probing · Sparse Autoencoders · Activation Patching Red Teaming Jailbreaks · Prompt Injection · Adversarial Attacks Alignment Research Reward Hacking · Scalable Oversight · Honesty 8.1 GPU Programming GPU and TPU Architecture SMs · HBM · Tensor Cores · Memory Bandwidth CUDA Basics Threads · Blocks · Shared Memory · Kernels Custom Kernels Triton · Kernel Fusion · CUTLASS Performance Analysis Roofline Model · Nsight · PyTorch Profiler 8.2 Distributed Training Data Parallelism DDP · FSDP2 · ZeRO Model Parallelism Tensor · Pipeline · Sequence · Expert Training Frameworks Megatron-LM · DeepSpeed · TorchTitan Cluster Computing Slurm · Kubernetes · NCCL · Fault-Tolerant Checkpointing 8.3 Memory and Compute Efficiency Mixed Precision BF16 · FP8 · Loss Scaling Activation Checkpointing and Offloading Efficient Attention FlashAttention · Ring Attention Parameter-Efficient Fine-Tuning LoRA · QLoRA · Adapters 8.4 Efficient Inference Inference Engines vLLM · SGLang · TensorRT-LLM Serving Optimizations PagedAttention · Continuous Batching · Speculative Decoding Quantization GPTQ · AWQ · FP8 · INT4 Pruning and Distillation Structured Pruning · Knowledge Distillation 8.5 Experiment Management Experiment Tracking Weights & Biases · MLflow · TensorBoard Checkpoints and Artifacts Model Versioning · Hugging Face Hub · Resumable Training Configuration and Sweeps Hydra · Hyperparameter Search · Sweep Tools 9.1 Reading and Synthesizing Papers Tracking the Literature arXiv · OpenReview · Conference Proceedings · Semantic Scholar Reading Papers Critically Claims · Assumptions · Baselines · Limitations Reproducing Papers Reimplementation · Matching Baselines · Reporting Discrepancies Formulating Hypotheses Identifying Gaps · Research Questions 9.2 Experiment Design Compute Budgeting Small-Scale Proxies · Scaling Experiments Baselines and Ablations Strong Baselines · Controlled Comparisons Statistical Rigor Multiple Seeds · Confidence Intervals · Significance Tests Rapid Prototyping Sanity Checks · Overfitting Tiny Batches · Debugging Training 9.3 Evaluation and Benchmarking Static Benchmarks MMLU-Pro · GPQA · SWE-bench Verified · Benchmark Saturation LLM-as-a-Judge Pairwise Comparison · Rubrics · Judge Bias Human Preference Evaluation LMArena · Elo Ratings · Crowdsourcing Open-Ended Evaluation Long Context · Tool Use · Agentic Tasks Evaluation Harnesses lm-evaluation-harness · Inspect · HELM 9.4 Scientific Communication Writing Papers Structure · Related Work · NeurIPS · ICLR · ICML Figures and Visualization Plots · Diagrams · Tables Peer Review Reviewing · Rebuttals · Workshops Talks and Posters Conference Talks · Poster Sessions 10.1 Capstone Projects Train a Small LLM from Scratch Tokenizer · Pre-training · SFT · Evaluation Post-Training Project Preference Data · DPO · GRPO · Evaluation Systems Project Custom Kernel · Distributed Training · Inference Optimization Novel Research Project Hypothesis · Experiments · Workshop Paper 10.2 Open Source and Portfolio Releasing Code and Models Hugging Face Hub · Model Cards · Licenses Contributing to Open Source PyTorch · Transformers · vLLM Technical Writing Blog Posts · Paper Explainers · Research Notes 10.3 Interviews and Career Growth Coding Interviews Algorithms · ML Coding · Implementing Attention ML Depth Interviews Paper Discussions · Research Proposals · Debugging Scenarios Research Presentations Job Talk · Past Work Deep Dive Career Paths Residencies · PhD Programs · Industry Labs · Startups