TopER is a low-dimensional, interpretable graph embedding method based on topological evolution rates, enabling intuitive visualization and strong performance on clustering and classification tasks.

We introduce the first benchmark of 84 real-world temporal graphs and MiNT, the first multi-network temporal model pre-trained on this collection, demonstrating strong transferability to unseen token networks.

Propose benchmark and sataset to evaluate LLMs extraction of state machines from network protocols standards, curated the datasets and benchmark on several open and close LLMs.

The paper proposes a neural network–based learned heuristic that accelerates MPE inference in high-treewidth PGMs by predicting impactful variable assignments from solver traces, significantly reducing search space while preserving solution quality.

We introduce MUelim, an algorithm for efficient motor unit decomposition that uses approximate joint diagonalization with a subtractive approach to rapidly identify motor units for advanced neuroprosthetic control.

Combinatorial Open-World Detection

In this paper, we study how preference data distribution influences DPO, from both theoretical and empirical perspectives.