Code

Research code

• DTSemNet · JMLR 2026, ECAI 2024

  Differentiable oblique decision trees and tree policies for RL.

• InK-BWTS · ECML-PKDD 2026

  Neurosymbolic hierarchical RL with Belief World Tree Search.

• F1Tenth

  ROS 2 race stack (SLAM, particle-filter localisation, pure pursuit) on Jetson AGX.

Giving back

Notebook-style tutorials that build things from scratch on toy problems. Ongoing.

• Deep RL Jutsu

  Deep RL algorithms from scratch (DQN, DDPG, SAC, PPO), with a curated guide to RL libraries.

• GenAI

  Small-scale VAE, GAN, diffusion, state-space models, CLIP and VLA on toy problems.

• RLFT

  RL fine-tuning of foundation models — RLHF, DPO and GRPO, from a from-scratch GPT to a pretrained LLM.

• Neurosymbolic

  Neurosymbolic approaches implemented on toy problems.