⌂ Me Now Musings Mission Artifacts Notes Readings Experiments ML Projects Resume

Projects

Research sprints at the intersection of learning, causality, and dynamical systems.


2025–2026 Reinforcement Learning · Multi-Agent

Zero-Shot Coordination in Multi-Agent RL

Convention emergence via IPPO in grid environments. Analyzing when ad-hoc teamwork fails and what structural symmetry-breaking enables coordination without prior agreement.

PyTorch JAX PettingZoo
2025–2026 Causal Inference · Time Series

Causal State-Space Models for Time Series

Structural causal models integrated with SSM architectures for counterfactual forecasting. Granger-causal baselines compared against do-calculus interventions in finance and climate data.

PyMC Mamba DoWhy
2025–2026 Physics-Informed ML · Dynamical Systems

Port-Hamiltonian Neural Networks

Energy-preserving neural ODEs constrained to Hamiltonian structure. Enforcing symplectic symmetry as inductive bias for learning conservative physical systems from trajectory data.

torchdiffeq JAX Diffrax
2025 LLM Agents · Social Simulation

Generative Social Network Simulation

LLM-driven agent simulation of graduate student archetypes. Evaluating opinion dynamics and belief propagation across synthetic social graphs under heterogeneous prior distributions.

GPT-4 NetworkX Mesa
2025 Vision-Language Models · Alignment

Personality-Aligned Vision-Language Model

Fine-tuning VLMs with RLHF to align visual description style with Myers-Briggs personality dimensions. Studying whether personality-conditioned reward signals improve downstream task coherence.

LLaVA TRL PEFT
GitHub · LinkedIn · Twitter · pc3197@columbia.edu

© 2024–2026 Prabakaran Chandran