Research

∑AI for Mathematics

sat

• Evaluating SageMath-Augmented LLM Agents for Computational and Experimental Mathematics — development and benchmarking of agentic systems with computational tools for solving real-world mathematical problems ICML 2026
• Applying language models to algebraic topology: generating simplicial cycles using multi-labeling in Wu's formula — applying Transformers to modeling sequences with group structure, with applications to homotopy groups of spheres ICML 2024
• Proving and Computing: Multi-Agent Systems with Lean and CAS for Research Mathematics — a multi-agent system that uses Lean and computer algebra as complementary tools for research-level mathematical problem solving NIPS 2026 MATH AI Workshop under review

♟LLM Agents in Games and Interactive Environments

unknown

Development and evaluation of LLM agentic pipelines for long-horizon planning and reasoning in games and interactive environments.

⊢Neuro-Symbolic Verified Reasoning

unknown

LLM verifiable reasoning with feedback from symbolic solvers (SMT/SAT) and compilers.

∂Geometry and Topology for Deep Learning

sat

Geometric and topological methods for studying, improving, and developing deep learning models.

• PhD project: Methods of geometry and topology in the study of deep learning models PhD thesis, 2025
• Sheaf theory: from deep geometry to deep learning — a survey of sheaf theory applications across computer science, data analysis, and mathematics arXiv 2025
• Study of Foundation Models Knowledge Representations: Geometry Perspective — investigating the geometric properties of embeddings in foundation models
• Chordal Embeddings Based on the Topology of Tonal Space — improving RNN performance by integrating domain-specific topological information into embeddings EvoMUSART 2023

⌖AI-Generated Content Detection

sat

• Robust AI-generated text detection by restricted embeddings — improving cross-domain and cross-generator robustness by removing harmful linear subspaces from Transformer embeddings EMNLP 2024 Findings
• AI-generated text boundary detection with RoFT — detecting the transition from human-written to machine-generated text using perplexity- and topology-based features under domain and model shifts COLM 2024 · best paper award
• Improving interpretability and robustness for the detection of AI-generated images — interpreting CLIP-based detectors and improving cross-generator transfer through embedding-component removal and attention-head selection arXiv 2024