Eylam Tagor

Bio

I am an M.S. student in Computer Science at Columbia University, advised by Prof. Elias Bareinboim. My research focuses on reinforcement learning and causal inference, particularly in building reliable decision-making systems that can handle uncertainty, partial observability, and real-world imperfections.

Before Columbia, I completed my undergraduate degree at UT Austin, where I worked with Prof. Peter Stone on multi-agent RL and sim2real as part of the RoboCup team.

I am currently applying for Ph.D. programs for Fall 2026, and seeking research internships from now until Fall 2026.

Research

Tagor, E., Li, M., & Bareinboim, E. CILBench: Benchmarking Robust Imitation Learning in Confounded High-Dimensional Control Environments: A benchmark and methodology for causal imitation learning in high-dimensional, long-horizon tasks.
2025 · Preprint · PDF · Environment Code · Algorithm Code

Gymnasium for Causal Imitation Learning: Developed imitation learning algorithms grounded in causal inference; built SCM-parameterized RL environments supporting observational, interventional, and counterfactual evaluation.
2025 · PDF

Multi-Agent RL for RoboCup @ LARG: Implemented multi-agent deep RL behaviors for NAO robot soccer; improved sim2real transfer using high-fidelity simulation, contributing to a first-place finish in the 2024 RoboCup SPL Challenge Division.
2024 · Code

Residual Neural Causal Model: Designed ResNCM, a neural causal model using PyTorch and Causal-Learn; conducted counterfactual analyses on education datasets, outperforming classical and deep-learning baselines.
2025 · PDF · Code

Generalist Embodied Video Game Agent @ NVIDIA: Built embodied agents and multimodal training environments using OpenAI Gym; integrated RL algorithms, LLMs, and large-scale pretraining to improve cross-task generalization.
2023 · Code

Teaching

• ORCS 4529: Reinforcement Learning (Fall 2025)
• COMS 4995: Causal Inference for Data Science (Spring 2025)