Projects

[WIP] Nano Optax
JIT-compiled first- and second-order optimization algorithms in JAX, blazingly fast and scalable.

Model-agnostic Meta-learning for RL
Implementation of MAML algorithm for RL for continuous control tasks using TorchRL.

Grokking Modular Addition
Exploring the 'grokking' phenomenon for modular addition using a one-layer transformer with bidirectional attention.

Skipgram word2vec model with Negative Sampling
Training word embeddings using negative sampling.

Law of Large Numbers for Markovian epidemic model
Proof and simulation of convergence of (scaled) stochastic SIR epidemic model to solution of the deterministic SIR ODE system.