Embeddings and representation learning

Reads "king − man + woman ≈ queen"; understands dense vs sparse features.

Trains word2vec / GloVe; uses cosine similarity for lookups.

Uses sentence-transformers / BGE for semantic search; fine-tunes embeddings with triplet loss.

Analyses isotropy, anisotropy; matrix-factorisation and contrastive framings.

Multi-modal embeddings, retrieval-aware pretraining, memory-efficient embedding tables.