notes · a working notebook
Things I keep
re-explaining to myself.
A working ML notebook — architectures, training mechanics, LLMs, vision, MLOps. Each entry is something I’ve actually had to reason about, not a textbook chapter. Some pages are written. Most are stubs I’m filling in over time.
1 written·143 drafts·13 categories
Model Architecture
Classical ML, neural network families, and architectural primitives.
- Linear and Logistic Regressiondraft
- k-Nearest Neighborsdraft
- Clusteringdraft
- Support Vector Machines (SVM)draft
- Naive Bayesdraft
- Decision Trees and Ensemble Methodsdraft
- ML Algorithms Comparative Analysisdraft
- DL Architectures Comparative Analysisdraft
- Neural Architecturesdraft
- Encoder vs. Decoder vs. Encoder-Decoder Modelsdraft
- Generative Adversarial Networks (GANs)draft
- Diffusion Modelsdraft
- Graph Neural Networksdraft
- Attentiondraft
- Transformers
- Parameter Efficient Fine-Tuningdraft
- Dropoutdraft
- Token Sampling Methodsdraft
- Separable Convolutionsdraft
- Inductive Biasdraft
- Convolutional Neural Networksdraft
- Reinforcement Learningdraft
- Mixture-of-Experts (MoE)draft
- State Space Modelsdraft
- FlashAttentiondraft
- Model Accelerationdraft
Data Foundations
Everything that happens to data, gradients, and loss surfaces during training.
- Data Samplingdraft
- Data Imbalancedraft
- Data Quality / Filteringdraft
- Standardization vs. Normalizationdraft
- Inter-Annotator Agreementdraft
- Learning Paradigmsdraft
- Xavier Initializationdraft
- Padding and Packingdraft
- Regularizationdraft
- Gradient Descent and Backpropdraft
- Activation Functionsdraft
- Loss Functionsdraft
- Fine-tuning Modelsdraft
- Splitting Datasetsdraft
- Batchnormdraft
- Double Descentdraft
- Fine-Tuning and Evaluating BERTdraft
- Training Loss Greater Than Validation Lossdraft
- Bias-Variance Tradeoffdraft
- Gradient Accumulation and Checkpointingdraft
- Personally Identifiable Information (PII)draft
- Hypernetworksdraft
- Distributed Training Parallelismdraft
- k-Fold Cross Validationdraft
- Knowledge Graphsdraft
NLP / LLMs / Agents
Language modelling, retrieval, agents, and the post-training stack.
- Embeddingsdraft
- Prompt Engineeringdraft
- Context Engineeringdraft
- NLP Tasksdraft
- Preprocessingdraft
- Tokenizationdraft
- Overview of Large Language Models (LLMs)draft
- Diffusion LLMs / Discrete Diffusion Modelsdraft
- Policy / Preference Optimizationdraft
- Agentsdraft
- Agentic Reinforcement Learningdraft
- Agentic Design Patternsdraft
- Agent Skillsdraft
- LLM-as-a-Judge / Autoratersdraft
- Speculative Decodingdraft
- Reinforcement Fine-Tuningdraft
- Machine Translationdraft
- Factuality in LLMsdraft
- Reasoning in LLMsdraft
- Hallucination Detection and Mitigationdraft
- AI Text Detection Techniquesdraft
- Named Entity Recognitiondraft
- Textual Entailmentdraft
- Retrieval Augmented Generation (RAG)draft
- LLM Context Length Extensiondraft
- Document Intelligencedraft
- Code Mixing and Switchingdraft
- Large Language Model Ops (LLMOps)draft
Vision
Backbones, structural primitives, and image generation.
Speech
Multimodal AI / VLMs
Vision-language models and computer-control agents.
Offline / Online Evaluation
MLOps
On-Device AI
Project Planning
Frameworks for shipping things, on time.
Models
Specific model writeups — what they are, what they changed.
Miscellaneous
Foundations, tooling, and one-offs.
- Ilya Sutskever's Top 30draft
- GPU Architecturedraft
- Debugging Model Trainingdraft
- ML Runtimesdraft
- Chain Ruledraft
- Bayes' Theoremdraft
- Probability Calibrationdraft
- Multiclass vs. Multilabel Classificationdraft
- N-Dimensional Tensor Productdraft
- PyTorch vs. TensorFlowdraft
- Approximate Nearest Neighbors – Similarity Searchdraft
- Transferability Estimationdraft
- TensorBoarddraft
- Convolutional Neural Networks for Text Classificationdraft
- Relationship between Hidden Markov Models and Naive Bayesdraft
- Maximum Entropy Markov Modelsdraft
- Conditional Random Fieldsdraft
- Hyperparametersdraft
- Hyperparameter Tuningdraft
- Hyperparameter Loggingdraft
Practice
Spot an error? Reach me at videetnimsarkar21@gmail.com.