News & Updates
Research highlights, publications, student achievements, and activities from the DYNAMO group.
2026
September 2026 — DYNAMO Paper Accepted to NeurIPS 2026
Concorde: Geometry-Aware Link Prediction via Decoupled Energy Minimization accepted to NeurIPS 2026. The paper develops a geometric approach to link prediction that combines spectral, mixed-curvature, and nonlocal representations to improve prediction in sparse networks.
April 2026 — Two DYNAMO Papers at ICLR 2026
DYNAMO researchers presented two papers at ICLR 2026, highlighting recent advances in physics-informed graph learning and human–AI systems.
FlowSymm: Physics-Aware, Symmetry-Preserving Graph Attention for Network Flow Completion develops a graph learning approach that respects physical conservation laws when reconstructing incomplete network flows.
Learning to Lie: Adversarial Attacks on Human-AI Teams and LLMs investigates how adversarial AI assistants can manipulate trust and influence collective decision-making.
2025
August 2025 — Two Papers at KDD 2025
DYNAMO researchers contributed two papers to KDD 2025: Normalized Space Alignment: A Versatile Metric for Representation Analysis, by Danish Ebadulla, Aditya Gulati, and Ambuj Singh, and Attribute-Enhanced Similarity Ranking for Sparse Link Prediction, with collaborators João Mattos, Zexi Huang, Mert Kosan, and Arlei Silva. These works advance the comparison of learned representations and link prediction in sparse graphs.
July 2025 — Evaluating Graph Generation by Large Language Models
Congratulations to Ege Demirci and Rithwik Kerur on their ACL 2025 Student Research Workshop paper, Are LLMs Truly Graph-Savvy? A Comprehensive Evaluation of Graph Generation.
The paper evaluates the ability of language models to generate graphs that satisfy structural and logical constraints, revealing important limitations in graph reasoning.
April 2025 — GraphEval36K Published at NAACL 2025
GraphEval36K: Benchmarking Coding and Reasoning Capabilities of Large Language Models on Graph Datasets was published in Findings of NAACL 2025.
The work introduces a benchmark of 40 graph programming problems and 36,900 test cases for assessing the ability of language models to reason about graph structure and algorithms.
February 2025 — DYNAMO Leadership at KDD 2025
Ambuj Singh served as Chair of the Datasets and Benchmarks track at KDD 2025 and KDD 2026.
2023
November 17, 2023 — Ambuj gives a distinguished lecture in the Charles L. and Ann Brown Distinguished Colloquium series in the ECE department at the University of Virginia.
August 7, 2023 — Ambuj serves as the General Chair for KDD 2023. A number of current students and alumni contributed to the organization of the event.
June 12, 2023 — Congratulations to Saurabh for passing his Major Area Examination!