Publications
See also my Google Scholar profile.
Papers
2026
-
ICML 2026
WorkshopISM: Self-Improving Strategy Memory for Continual Mathematical ReasoningAI for Math Workshop (AI4Math) at ICML 2026A strategy memory that learns from both successful and failed episodes to improve mathematical reasoning with a frozen language model, evaluated on MATH-Hard and OlympiadBench.
-
ICML 2026
WorkshopHonest Lying: Understanding Memory Confabulation in Reflexive AgentsWorkshop on Failure Modes in Agentic AI at ICML 2026Studies why reflexive agents produce plausible but incorrect recollections of their own past, and what this implies for the reliability of long-horizon agents.
-
IGARSS
2026WildfireVLM: AI-powered Analysis for Early Wildfire Detection and Risk Assessment Using Satellite ImageryIEEE International Geoscience and Remote Sensing Symposium (IGARSS) 2026A vision-language framework for early wildfire detection and risk assessment from satellite imagery.
2024
-
NeurIPS 2024
WorkshopSBI-RAG: Enhancing Math Word Problem Solving for Students through Schema-Based Instruction and Retrieval-Augmented GenerationMATH-AI Workshop at NeurIPS 2024Combines schema-based instruction with retrieval-augmented generation to produce step-by-step, schema-guided solutions to math word problems.
2023
-
AAAI 2023
WorkshopReProHRL: Towards Multi-Goal Navigation in the Real World using Hierarchical AgentsReinforcement Learning Ready for Production Workshop at AAAI 2023Hierarchical RL for multi-goal navigation, transferred from simulation to a real nano-drone.
2022
Thesis & Other Writing
-
M.S.
ThesisDynamic Edge Weighting in Relational Graph Convolutional Networks: Enhancing Sample Efficiency via Graph Attention in Reinforcement LearningM.S. Thesis, University of Maryland, Baltimore County, 2024Integrates graph attention into an R-GCN framework for model-free RL, improving sample efficiency by 20% on Box-World and MiniGrid LavaGap.
-
Blog
2025SOTA Results in Agentic Memory on LongMemEvalEmergence AI Research Blog, July 2025A long-term memory system for conversational agents reaching 86% accuracy on LongMemEval, 15 points above the previous best.