An independent AI research program created by Harshit.
I am an educator at
I have designed this program to be thorough, exciting, gamified and hands-on. The program also draws inspiration from various Research Engineer roles at companies like OpenAI, Anthropic, Meta, and the likes.
The following curriculum is designed to help me get a deeper and thorough understanding of all important concepts in AI.
- AI Research Program
- Table of contents
- My learning approach
- How to consume this?
- Pillar 1: Foundational Knowledge
- Pillar 2: Programming and Tools
- Pillar 3: Deep Learning Fundamentals
- Pillar 4: Reinforcement Learning
- Pillar 5: NLP / LLM Research
- Pillar 6: Research Paper Analysis and Replication
- Pillar 7: High-Performance AI Systems and Applications
- Pillar 8: Large-scale ETL and Data Engineering
- Pillar 9: Ethical AI and Responsible Development
- Pillar 10: Community Engagement and Networking
- Pillar 11: Research and Publication
- Resources
I have a background in Data Science and ML and I've decided to take a recursive approach(going deeper as needed) based on my level of understanding.
My primary purpose of starting this program is to dig deeper into LLM research and engineering problems.
So, I'll start from Pillar 5, NLP/LLM research, how LLMs work, how they are pre-trained and how to finetune them. In the process, I will hop onto other Pillars to learn and re-learn a other fundamental concepts.
Keep an eye on weekly-notes for my learnings and thoughts.
I'll dedicate at least 1 hours daily to my learning.
⚠️ This program is still under development and will be updated as I progress in my journey. Feel free to contribute to it by opening issues or suggesting improvements.
- Beginners should start with the Foundational Knowledge section.
- For all others who have some exposure to ML/DL, tweak the curriculum as per your level of understanding.
- This curriculum is not supposed to be consumed in a linear fashion. Treat it more like a checklist.
- You should start with the basics and gradually move towards more complex topics as per your goals and interests.
To follow me in this journey, consider subscribing to my:
- weekly newsletter.
- Learning vlogs at youtube
- Discord for discussions and sharing your learnings
| Topic | Subtopics | Learning Resources | Projects/Assignments | Evaluation Criteria |
|---|---|---|---|---|
| Linear Algebra | Vectors, matrices, eigenvalues | MIT OpenCourseWare, 3Blue1Brown videos | Implement basic operations, solve systems of equations | Quiz, problem set, subjective evaluation |
| Calculus | Derivatives, integrals, gradients | Khan Academy, Coursera | Optimization problems, gradient descent implementation | Quiz, problem set, subjective evaluation |
| Probability | Distributions, Bayes theorem | Stanford CS109 | Probabilistic modeling project | Quiz, project review, subjective evaluation |
| Statistics | Hypothesis testing, regression | StatQuest videos, EdX courses | Data analysis project | Project review, subjective evaluation |
| Information Theory | Entropy, mutual information | Elements of Information Theory (book) | Implement compression algorithm | Project review, subjective evaluation |
| Skill | Subtopics | Learning Resources | Projects/Assignments | Evaluation Criteria |
|---|---|---|---|---|
| Python | Data structures, OOP, functional programming | Python.org tutorials, Real Python | Build a data processing pipeline | Code review, subjective evaluation |
| PyTorch | Tensors, autograd, distributed training | PyTorch documentation, Fast.ai course | Implement and distribute a neural network | Project review, subjective evaluation |
| Git | Version control, branching, merging | Git documentation, GitHub Learning Lab | Contribute to an open-source project | Contribution review, subjective evaluation |
| Linux | Command line, shell scripting, OS internals | Linux Journey, "Operating Systems: Three Easy Pieces" book | Implement a simple process scheduler | Practical test, code review, subjective evaluation |
| Cloud Platforms | AWS, GCP, or Azure basics | Cloud provider documentation | Deploy a scalable web application | Project review, subjective evaluation |
| High Performance Computing | GPU programming, CUDA | NVIDIA CUDA tutorials | Optimize a neural network for GPU computation | Performance benchmarks, subjective evaluation |
| Kubernetes | Container orchestration, scaling | Kubernetes documentation, KodeKloud courses | Deploy and scale an ML model using Kubernetes | Project review, subjective evaluation |
| Topic | Subtopics | Learning Resources | Projects/Assignments | Evaluation Criteria |
|---|---|---|---|---|
| Neural Networks | Perceptrons, activation functions, backpropagation | Deep Learning book by Goodfellow et al. | Implement a multi-layer perceptron | Code review, subjective evaluation |
| CNNs | Convolution, pooling, modern architectures | CS231n course materials | Build an image classification model | Project review, subjective evaluation |
| RNNs | LSTM, GRU, sequence modeling | D2L.ai tutorials | Implement a language model | Code review, subjective evaluation |
| Attention Mechanisms | Self-attention, multi-head attention | "Attention Is All You Need" paper | Implement and optimize a new attention mechanism | Performance analysis, subjective evaluation |
| Transformer Variants | Transformer-XL, Reformer, Performer | Recent papers on arXiv | Compare compute efficiency of two Transformer variants | Comparative analysis, subjective evaluation |
| Optimization Techniques | SGD, Adam, learning rate schedules | Sebastian Ruder's blog | Compare optimizers on a benchmark task | Report review, subjective evaluation |
| Regularization | Dropout, batch normalization, data augmentation | Papers with Code | Implement and compare regularization techniques | Project review, subjective evaluation |
| Topic | Subtopics | Learning Resources | Projects/Assignments | Evaluation Criteria |
|---|---|---|---|---|
| RL Fundamentals | MDP, value functions, policy gradients | Sutton & Barto book, David Silver's course | Implement and solve classic RL problems (e.g., CartPole, Mountain Car) | Algorithm performance, subjective evaluation |
| Deep RL | DQN, A3C, PPO | OpenAI Spinning Up | Implement a deep RL algorithm for a complex environment (e.g., Atari games) | Agent performance, subjective evaluation |
| RL for NLP | RLHF, sequence generation | Recent papers (e.g., InstructGPT) | Fine-tune a language model using RL | Model improvement, subjective evaluation |
| Topic | Subtopics | Learning Resources | Projects / Assignments | Evaluation Criteria |
|---|---|---|---|---|
| 1. Foundations | - Word Vectors/Embeddings - Tokenization - Preprocessing - Data Sampling |
- Book: "Speech and Language Processing" by Jurafsky & Martin - Course: Stanford CS224N: NLP with Deep Learning - Paper: "Efficient Estimation of Word Representations in Vector Space" by Mikolov et al. |
- Implement word2vec from scratch - Build a custom tokenizer for a specific language or domain - Create a data preprocessing pipeline for a large text corpus |
- Accuracy of word embeddings on analogy tasks - Efficiency and coverage of tokenization - Quality and cleanliness of preprocessed data |
| 2. Classical NLP Techniques | - Hidden Markov Models - Naive Bayes - Maximum Entropy Markov Models - Conditional Random Fields |
- Book: "Foundations of Statistical Natural Language Processing" by Manning & Schütze - Tutorial: "A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition" by Rabiner |
- Implement a part-of-speech tagger using HMMs - Build a spam classifier using Naive Bayes - Develop a named entity recognition system using CRFs |
- Accuracy on standard POS tagging datasets - Precision, recall, and F1 score for spam classification - F1 score on CoNLL 2003 NER dataset |
| 3. Neural Architectures | - Feed-forward Neural Networks - Recurrent Neural Networks - Convolutional Neural Networks - Attention Mechanisms - Transformers |
- Book: "Deep Learning" by Goodfellow, Bengio, & Courville - Paper: "Attention Is All You Need" by Vaswani et al. - Course: fast.ai Practical Deep Learning for Coders |
- Implement a sentiment analysis model using CNNs - Build a language model using LSTMs - Create a machine translation system using Transformers |
- Accuracy on sentiment analysis benchmarks (e.g., IMDb) - Perplexity of language model on test set - BLEU score for machine translation |
| 4. Language Models | - N-gram Models - Neural Language Models - Autoregressive vs. Autoencoder Models - Large Language Models (LLMs) - Vision-Language Models (VLMs) |
- Paper: "Language Models are Few-Shot Learners" (GPT-3) - Blog: "The Illustrated Transformer" by Jay Alammar - Course: Hugging Face NLP Course |
- Fine-tune GPT-2 for text generation - Implement a BERT-based question answering system - Create a multimodal model for image captioning |
- Perplexity and cross-entropy loss - F1 and Exact Match scores for QA - BLEU, METEOR, and CIDEr scores for image captioning |
| 5. Advanced LLM Concepts | - LLM Alignment - Token Sampling Methods - Context Length Extension - Personalization |
- Paper: "Constitutional AI: Harmlessness from AI Feedback" by Anthropic - Blog: "How to sample from language models" by Hugging Face - Paper: "LoRA: Low-Rank Adaptation of Large Language Models" |
- Implement different decoding strategies (greedy, beam search, top-k, top-p) - Develop a method to extend context length of a pre-trained LLM - Create a personalized language model using adapters |
- Human evaluation of model alignment - Quality and diversity of generated text - Perplexity on long-context tasks - Personalization accuracy on user-specific tasks |
| 6. NLP Applications | - Machine Translation - Named Entity Recognition - Textual Entailment - Retrieval Augmented Generation (RAG) - Document Intelligence |
- Paper: "Neural Machine Translation by Jointly Learning to Align and Translate" - Tutorial: spaCy's Named Entity Recognition - Paper: "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks" |
- Build an end-to-end neural machine translation system - Develop a RAG system for question answering - Create a document understanding system for invoice processing |
- BLEU, METEOR scores for MT - F1 score for NER - Accuracy on textual entailment datasets (e.g., SNLI) - Relevance and accuracy of RAG responses |
| 7. Knowledge Representation | - Knowledge Graphs - Semantic Networks |
- Book: "Knowledge Representation and Reasoning" by Brachman & Levesque - Tutorial: Neo4j Graph Database |
- Construct a knowledge graph from a text corpus - Develop a question answering system using a knowledge graph |
- Coverage and accuracy of extracted knowledge - Precision and recall of graph-based QA system |
| 8. Challenges and Mitigation | - Hallucination Mitigation - AI Text Detection - Bias Detection and Mitigation |
- Paper: "TruthfulQA: Measuring How Models Mimic Human Falsehoods" - Blog: "Detecting Machine-Generated Text" by OpenAI - Paper: "Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings" |
- Develop a fact-checking system for LLM outputs - Create an AI text detector - Implement bias mitigation techniques in word embeddings |
- Reduction in hallucination rate - Accuracy of AI text detection - Reduction in bias measures (e.g., WEAT score) |
| 9. Evaluation and Benchmarking | - LLM/VLM Benchmarks - Task-specific Metrics |
- Paper: "GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding" - Tutorial: Hugging Face's Evaluate library |
- Evaluate an LLM on multiple NLP tasks using GLUE benchmark - Implement and compare different evaluation metrics for a specific NLP task |
- Performance across multiple benchmarks - Inter-annotator agreement for human evaluation |
| 10. Practical Aspects | - Large Language Model Ops (LLMOps) - Ethical Considerations |
- Book: "Building Machine Learning Pipelines" by Hapke & Nelson - Paper: "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" |
- Design and implement an LLMOps pipeline - Conduct an ethical audit of an NLP system |
- Efficiency and reliability of deployment pipeline - Compliance with ethical AI principles |
| Activity | Approach | Resources | Deliverables | Evaluation Criteria |
|---|---|---|---|---|
| Paper Reading | Weekly reading of latest papers | arXiv, Papers With Code | Paper summaries, critiques | Quality of analysis, subjective evaluation |
| Research Replication | Reproduce results of significant papers | Original papers, open-source implementations | Replicated experiments, results comparison | Accuracy of replication, subjective evaluation |
| Trend Analysis | Identify and analyze research trends | NeurIPS, ICML, ACL proceedings | Trend reports, blog posts | Insight quality, subjective evaluation |
| Paper Presentation | Present papers to peers or online | Conference recordings | Video explanations, slide decks | Presentation quality, subjective evaluation |
| Extension Projects | Extend or combine ideas from papers | Related work sections of papers | Novel experiments, blog posts | Originality, subjective evaluation |
| Visualization Projects | Develop interactive visualizations for ML concepts | D3.js, Plotly | Create an interactive visualization of attention between tokens in a language model | Visualization quality, insight provided, subjective evaluation |
| Topic | Subtopics | Learning Resources | Projects/Assignments | Evaluation Criteria |
|---|---|---|---|---|
| Model Optimization | Quantization, pruning, distillation | TensorFlow Model Optimization Toolkit | Optimize a large model for edge devices | Performance benchmarks, subjective evaluation |
| Efficient Architectures | MobileNet, EfficientNet, BERT-tiny | Papers, GitHub implementations | Implement and benchmark efficient models | Comparative analysis, subjective evaluation |
| Hardware Acceleration | GPU programming, TPU utilization | NVIDIA Deep Learning Institute | Optimize a model for specific hardware | Performance improvement, subjective evaluation |
| Scalable AI Systems | Microservices, containerization, orchestration | Kubernetes documentation, Docker tutorials | Design a scalable AI service architecture | Architecture review, subjective evaluation |
| ML Ops | CI/CD for ML, model monitoring | Google MLOps guides | Set up an MLOps pipeline | Pipeline effectiveness, subjective evaluation |
| Topic | Subtopics | Learning Resources | Projects/Assignments | Evaluation Criteria |
|---|---|---|---|---|
| ETL Fundamentals | Data extraction, transformation, loading | Udacity Data Engineering course | Design and implement an ETL pipeline | Pipeline efficiency, subjective evaluation |
| Distributed Data Processing | Apache Spark, Hadoop | Spark documentation, Coursera courses | Process and analyze a large dataset using Spark | Processing speed, scalability, subjective evaluation |
| Data Warehousing | Star schema, OLAP | Kimball Group resources | Design a data warehouse for ML experiments | Design review, subjective evaluation |
| Streaming Data | Kafka, Flink | Confluent Kafka tutorials | Implement a real-time data processing pipeline | System performance, subjective evaluation |
| Topic | Subtopics | Learning Resources | Projects/Assignments | Evaluation Criteria |
|---|---|---|---|---|
| AI Ethics | Fairness, accountability, transparency | MIT Moral Machine, ethicalML.org | Develop an AI ethics framework | Framework review, subjective evaluation |
| Bias in AI | Dataset bias, algorithmic bias | "Gender Shades" paper, AI Fairness 360 | Audit a model for bias | Report review, subjective evaluation |
| Privacy in ML | Differential privacy, federated learning | OpenMined tutorials | Implement privacy-preserving ML | Project review, subjective evaluation |
| Explainable AI | LIME, SHAP, counterfactual explanations | Interpretable Machine Learning book | Develop model explanations | Project review, subjective evaluation |
| AI Governance | Regulations, guidelines, best practices | EU AI Act, IEEE Ethically Aligned Design | Propose AI governance structure | Proposal review, subjective evaluation |
| Activity | Approach | Resources | Deliverables | Evaluation Criteria |
|---|---|---|---|---|
| Conference Participation | Attend or present at AI conferences | Conference websites, Call for Papers | Presentation slides, trip reports | Engagement quality, subjective evaluation |
| Open Source Contribution | Contribute to AI/ML open source projects | GitHub, PyPi | Code contributions, pull requests | Contribution impact, subjective evaluation |
| AI Community Building | Organize meetups, study groups | Meetup.com, local tech communities | Event reports, community growth metrics | Community impact, subjective evaluation |
| Online Presence | Blog writing, social media engagement | Medium, Twitter, LinkedIn | Blog posts, thread discussions | Reach and engagement, subjective evaluation |
| Collaborative Projects | Partner with other researchers/engineers | Academic collaborations, hackathons | Joint projects, co-authored papers | Collaboration quality, subjective evaluation |
| Activity | Approach | Resources | Deliverables | Evaluation Criteria |
|---|---|---|---|---|
| Research Ideation | Brainstorming, literature review | arXiv, Google Scholar | Research proposals | Originality, feasibility, subjective evaluation |
| Experiment Design | Hypothesis formulation, methodology planning | Research design books, mentorship | Experimental protocols | Rigor, subjective evaluation |
| Data Collection and Analysis | Data gathering, statistical analysis | Kaggle datasets, statistical tools | Datasets, analysis reports | Data quality, analysis depth, subjective evaluation |
| Paper Writing | Scientific writing, LaTeX | Overleaf, writing workshops | Draft papers | Writing quality, subjective evaluation |
| Peer Review | Understand and participate in peer review process | PubliONS, Elsevier Researcher Academy | Review reports | Review quality, subjective evaluation |
| Publication and Presentation | Submit to journals/conferences, create posters | Journal guidelines, conference deadlines | Published papers, conference posters | Impact factor, presentation quality, subjective evaluation |
