All projects
Generative Models, Transformers & Deep RL
An academic deep-learning project spanning generative modeling and reinforcement learning: a conditional DDPM with a flow-matching head, GAN/VAE families, a from-scratch MiniGPT transformer, and DQN/DDPG agents.
An academic deep-learning project covering both generative modeling and deep reinforcement learning:
- Generative models — a conditional DDPM with a flow-matching head on a class-conditioned Residual U-Net, plus GAN, LSGAN, and DCGAN variants, a VAE and a CVAE, and a from-scratch MiniGPT transformer.
- Value-based RL — DQN and Double DQN agents on
CarRacing-v3, using hard-target updates and Polyak averaging to curb Q-value overestimation, with an ablation over the soft-update rate (τ). - Policy-gradient RL — solved the continuous-action Inverted Double Pendulum with a DDPG actor–critic, using an Ornstein–Uhlenbeck noise process for exploration.