🧮 Topological Data Analysis

Persistent homology, Euler characteristic curves, filtration-based descriptors, and stable topological summaries as inductive biases for ML.

🕸️ Graph Classification & Embeddings

Topology-aware Transformers and embeddings for graph-level tasks — TopoFormer, TopER.

⏱️ Temporal Graph Learning

Topological encoder–decoder frameworks for graphs whose nodes and edges evolve over time — TopoGED.

📍 Node Classification

Spatial and contextual coordinates for node-level prediction — SCNode.

🔗 Link Prediction

Strong proximity-based baselines and dual-perspective methods on graphs and knowledge graphs — PROXI, DuoLink; knowledge graph completion via harmonic optimal transport — HOT-KG.

🧬 AI for Drug Discovery & Biology

Topological ML for identifying molecular compounds — including compounds targeting bacterial propionate metabolism.

🛡️ AI for Cybersecurity

Topology-aware retrieval and correlation across disparate cybersecurity knowledge graphs — TopoCorr-KG (ReliaQuest Innovation Challenge).

TopoCorr-KG 2026 – present · ReliaQuest

Topology-Aware Retrieval and Correlation Across Disparate Cybersecurity Knowledge Graphs. AI-driven methods for retrieving and correlating information across fragmented cybersecurity knowledge graphs (MITRE ATT&CK, CVE/NVD, internal SOC graphs) to improve cyber threat analysis and decision-making.

PI: A. Tola · Co-PI: M. Asllani · GTA: Ece Karacam · UTA: Sebastian Powers

HOT-KG Aug 2026 – present · FSU UROP · Knowledge Graphs

Harmonic Optimal Transport for Knowledge Graph Completion.

FSU Undergraduate Research Opportunity Program (UROP) · Mentor: A. Tola · 3 undergraduate researchers.

TopoGED NeurIPS 2026 · Temporal Graphs

A Topological Encoder-Decoder Framework for Temporal Graph Learning. Brings topological encodings to temporal graphs — networks whose structure changes over time — within an encoder–decoder architecture.

With R. Buck, K. Shamsi, T. G. B. Ngo, B. Coskunuzer, C. G. Akcora.

TopoFormer ICLR 2026

Topology Meets Attention for Graph Learning. A lightweight, scalable Transformer framework that encodes graph topology into ordered token sequences via a novel Topo-Scan module. Bypasses persistence diagrams, comes with stability guarantees, and matches state-of-the-art on graph classification.

With MD J. Uddin (★ equal contribution), C. G. Akcora, B. Coskunuzer.

TopER NeurIPS 2025

Topological Embeddings in Graph Representation Learning. An interpretable, low-dimensional (2D) graph embedding grounded in topology. TopER fits a line to the (nodes, edges) trajectory across a filtration, producing a (pivot, growth) descriptor that scales to 100K-node graphs.

With F. M. Taiwo, C. G. Akcora, B. Coskunuzer. Project page.

SCNode TMLR 2025 · Node Classification

Spatial and Contextual Coordinates for Graph Representation Learning. A node-embedding framework that jointly captures spatial geometry and contextual neighborhood signal for richer node-level representations.

With MD J. Uddin, C. G. Akcora, B. Coskunuzer. PDF.

PROXI TMLR 2025 · LoG 2026 (TMLR Journal Track) · Link Prediction

Challenging the GNNs for Link Prediction. A surprisingly strong proximity-based baseline that exposes how simple structural features can rival — and often beat — more complex GNNs on link prediction benchmarks.

With J. A. Myrich, B. Coskunuzer. PDF.

Bacterial Propionate Metabolism J. Mol. Graph. Model. 2026 · Drug Discovery

Identification of Molecular Compounds Targeting Bacterial Propionate Metabolism with Topological Machine Learning. Uses topological descriptors and ML to screen molecular compounds that disrupt bacterial propionate metabolic pathways.

With S. Aziz, D. Zhabilov, D. Winkler, M. Candas, B. Coskunuzer. bioRxiv.

DuoLink In preparation · Link Prediction

A Dual Perspective on Link Prediction via Line Graphs. Reformulates link prediction by jointly modeling a graph and its line graph, capturing both node–node and edge–edge interactions for stronger predictive performance.

With S. Kumar, B. Coskunuzer.