BITS WILP AIML Knowledge Base: Semester 2 Resources & Elective Strategy

Rationale for Semester II Core Courses

Choosing Your Specialisation Track

Semester 2 Evaluation Model & Academic Timeline

Semester 2 typically runs from May through September (or November through March depending on your cohort admission cycle). Continuous evaluation is split into three rigorous components:

ACI expands AI beyond purely statistical machine learning. You explore classical search algorithms (A*, IDA*, minimax, alpha-beta pruning, constraint satisfaction), knowledge representation, first-order logic, probabilistic reasoning with Bayesian inference, fuzzy logic controllers, and genetic algorithms. Exams feature a blend of tree-traversal traces, algorithmic proofs, and reasoning calculations.

DRL investigates sequential decision making under uncertainty. Topics cover Markov Decision Processes (MDPs), Bellman equations, Dynamic Programming (Policy & Value Iteration), Monte Carlo methods, Temporal Difference learning (Q-Learning, SARSA), Deep Q-Networks (DQN, Double DQN, Dueling DQN), Policy Gradients (REINFORCE), and Actor-Critic architectures (A2C, PPO, DDPG). Webinars and assignments are Python-based (Gymnasium / PyTorch), while examinations are heavily numerical.

NLP is the gateway course for language engineering, LLMs, and RAG architectures. Topics cover N-gram language modeling, Part-of-Speech (POS) tagging with Hidden Markov Models (Viterbi decoding), Maximum Entropy models, Vector Semantics (TF-IDF, Word2Vec, GloVe), Dependency Parsing, Sequence-to-Sequence models, Attention mechanisms, and Transformer encoders/decoders. Assignments are group-based; examinations combine algorithmic calculations (Viterbi lattices, perplexity) with reasoning.

Information Retrieval covers the architectural building blocks of modern search engines, recommendation systems, and RAG retrieval stages. Topics include Inverted Indexing, Boolean retrieval, Vector Space Models, TF-IDF scoring, BM25 ranking, Web Crawling architectures, Link Analysis (PageRank, HITS), Evaluation metrics (MAP, MRR, NDCG), and Latent Semantic Indexing. This course is widely regarded as one of the least math-intensive electives, making it a high-scoring choice alongside NLP.

DMML is an essential applied course for engineers who build data platforms for AI models. It bridges traditional databases with modern big data engineering: Kafka for stream ingestion, Apache Spark for distributed ETL, Airflow and DBT for workflow orchestration, Snowflake and Lakehouse architectures, feature stores (Feast), data versioning (DVC), and data quality monitoring. Practical, industry-aligned, and low math intensity.

PGM is designed for students who thoroughly enjoyed the mathematical rigour of MFML in Semester 1. It investigates compact representations of joint probability distributions over high-dimensional spaces using graphs. Topics cover Directed Graphical Models (Bayesian Networks), Undirected Graphical Models (Markov Random Fields), D-separation, Variable Elimination, Clique Tree algorithms, Exact & Approximate Inference (MCMC, Gibbs sampling), and parameter learning. Heavily theoretical and proof-driven.

FAccT ML focuses on ethical and trustworthy AI. Topics cover demographic parity, equal opportunity, disparate impact metrics, model explainability (SHAP values, LIME, partial dependence plots, counterfactual explanations), algorithmic bias auditing, and differential privacy. Heavy programmatic assignments in Python with numerical exam calculations.