AI & Machine Learning · Thu, Oct 29, 2026

Machine learning, AI, natural language processing, computer vision and multi-agent systems.

RSS feed
arXiv cs.CV · cs.CL

AskStream: When and What to Ask Proactively under Uncertainty in Streaming Video Understanding

In briefAskStream: A Benchmark for When and What to Ask in Streaming Video Understanding

SummaryThis paper introduces AskStream, the first multi-turn benchmark for streaming video understanding that covers both video and user uncertainty. It adds an ask action alongside wait and answer, and includes 438 videos with 2,714 manually annotated frame-level time windows. The strongest model reaches only 52.18% on the proposed Ask-F1 metric.

Lu Wang, Zhuoran Jin, Longxiang Wang et al.

arXiv cs.LG · cs.DS

ExplainCC: Explaining Correlated Clusters by Node Labels

In briefExplainCC: Explaining Correlated Clusters Using Node Labels

SummaryThis paper studies Correlation Clustering, where objects are grouped by pairwise similarity, and requires the clusters to be explainable by node labels. The authors restrict attention to clusters that a label tree, a decision tree splitting objects by labels, can describe. They propose a top-down greedy algorithm that finds locally optimal splits, using maintained counters for efficiency. Experiments show it processes graphs with over a million edges in minutes, recovers ground truth on synthetic data, and yields interpretable clusters on real networks.

Nadeesha Shyami Kumari Herath Mudiyanselage, Florian Adriaens, Nikolaj Tatti

arXiv cs.RO · cs.AI · cs.LG

Explore Where It Matters: Targeted Exploration for Efficient Rover Adaptation

In briefTargeted Exploration for Efficient Rover Adaptation: Exploring Where Errors Matter Most

SummaryThis paper presents TERRA, an online framework that helps planetary rovers correct errors between simulation and real deformable terrain during deployment. It updates a world model with a reward-weighted prediction loss and trains a loss predictor to estimate where performance-relevant errors are likely, guiding exploration during MPC. In simulation, TERRA reaches about 90% slope-climbing success within five trials under severe out-of-distribution pretraining, and on the ERNEST rover it recovers coordinated active-suspension locomotion on a Mars-simulant slope within three to five real-world trials.

Jacob Levy, Arthur Bouton, Tristan D. Hassler et al.