Astrit Tola
Postdoctoral Scholar · Department of Mathematics · Florida State University
I work at the intersection of Topological Data Analysis, Machine Learning, and AI, with applications spanning drug discovery, biological networks, and cybersecurity. My research designs topology-aware AI methods that make modern ML more interpretable, robust, and grounded in mathematical structure.
Previously, I completed my Ph.D. in Mathematics at the University of Texas at Dallas (2020–2025), advised by Prof. Baris Coskunuzer. My dissertation, "Graph Representation Learning with Topological Data Analysis," developed new topological methods for learning expressive, interpretable representations of graphs. I currently work with Asst. Prof. Malbor Asllani at FSU.
Education
Ph.D. in Mathematics · University of Texas at Dallas
Dissertation: Graph Representation Learning with Topological Data Analysis
Advisor: Prof. Baris Coskunuzer
B.S. in Mathematics · Middle East Technical University (METU)
Ankara, Turkey.
Upcoming
Artificial Intelligence Seminar · FSU Department of Scientific Computing
Talk: Topology-Aware Graph Representation Learning · Tallahassee, FL
Learning on Graphs Conference (LoG 2026) · Boston, MA
Presenting PROXI: Challenging the GNNs for Link Prediction (TMLR Journal Track) · Northeastern University
NeurIPS 2026 · Atlanta, GA
Attending for TopoGED: A Topological Encoder-Decoder Framework for Temporal Graph Learning
News
LoG 2026 PROXI selected for the LoG 2026 TMLR Journal Track
Our TMLR paper "PROXI: Challenging the GNNs for Link Prediction" was accepted for presentation at the Learning on Graphs Conference (LoG 2026) in the TMLR Journal Track. See you in Boston, November 20–22!
New Course Introduction to Computational Topology and Machine Learning
This Spring 2027 I'm teaching a new course at FSU, open to all majors, with no prior topology, Python, or machine learning required. It is offered as an undergraduate section (MAP 4934.02) and a dual-listed graduate section (MAP 5932.04), meeting Monday and Wednesday at 3:05 PM in LOV 107.
→ Undergraduate section · Graduate section · Course flyer (PDF)
NeurIPS 2026 TopoGED accepted at NeurIPS 2026
Our paper "A Topological Encoder-Decoder Framework for Temporal Graph Learning" (TopoGED) was accepted at NeurIPS 2026. Joint work with R. Buck, K. Shamsi, T. G. B. Ngo, B. Coskunuzer, and C. G. Akcora. See you in Atlanta, December 9–13!
UROP HOT-KG: Harmonic Optimal Transport for Knowledge Graph Completion
New Grant ReliaQuest Innovation Challenge Award
Excited to share that our proposal, "TopoCorr-KG: Topology-Aware Retrieval and Correlation Across Disparate Cybersecurity Knowledge Graphs", has been selected for the ReliaQuest Innovation Challenge at Florida State University. Looking forward to collaborating with Co-PI Malbor Asllani, with Ece Karacam (GTA) and Sebastian Powers (UTA) joining the team.
TopoFormer accepted at ICLR 2026
Our paper "Topology Meets Attention for Graph Learning" was accepted at ICLR 2026.
TopER presented at NeurIPS 2025
Poster on topological embeddings for graphs at NeurIPS 2025, San Diego.
Research Highlights
Topological Data Analysis
Persistent homology, Euler characteristic curves, and topological summaries as inductive biases for modern ML.
Graph Classification & Embeddings
Topology-aware Transformers and embeddings for graph-level tasks (TopoFormer, TopER).
Temporal Graph Learning
Topological encoder–decoder frameworks for learning on graphs that evolve over time (TopoGED).
Node Classification
Spatial and contextual node representations for node-level prediction tasks (SCNode).
Link Prediction
Strong, interpretable baselines and dual-perspective methods for link prediction in graphs (PROXI, DuoLink), and knowledge graph completion (HOT-KG).
Drug Discovery & Biology
Topological machine learning for identifying molecular compounds, e.g., compounds targeting bacterial propionate metabolism.
AI for Cybersecurity
Retrieval and correlation across cybersecurity knowledge graphs (TopoCorr-KG, ReliaQuest Innovation Challenge).