# Video Question Answering

2 related articles

Subscribe to updates

arXiv Oct 5, 2026
BPO replaces single-response rewards with counterfactual-view behavior packs, lifting Qwen2.5-VL-7B temporal-hard accuracy by 7.8 points
The work proposes Behavior Pack Optimization (BPO), which replaces the single-response reward in video MLLM post-training with a jointly scored behavior pack of outputs across counterfactual views chosen by question type, requiring stability under irrelevant interventions, sensitivity when key evidence is removed, and abstention when no evidence remains, using the original-view response as an anchor-relative advantage; on TempCompass, MVBench, and NExT-QA, BPO improves Qwen2.5-VL-7B-Instruct macro accuracy by 4.7 pp, the temporal-hard subset by 7.8 pp, and abstention F1 by 20.0 pp over a budget-matched vanilla GRPO baseline from the same SFT checkpoint, transferring to Video-MME, LongVideoBench, and LLaVA-Video-7B.
The work proposes Behavior Pack Optimization (BPO), which replaces the single-response reward in video MLLM post-training with a jointly scored behavior pack of outputs across counterfactual views chosen by question type, requiring stability under irrelevant interventions, sensitivity when key evidence is removed, and abstention when no evidence remains, using the original-view response as an anchor-relative advantage; on TempCompass, MVBench, and NExT-QA, BPO improves Qwen2.5-VL-7B-Instruct macro accuracy by 4.7 pp, the temporal-hard subset by 7.8 pp, and abstention F1 by 20.0 pp over a budget-matched vanilla GRPO baseline from the same SFT checkpoint, transferring to Video-MME, LongVideoBench, and LLaVA-Video-7B.
Stanford University The University of Melbourne Peking University
Zhaolu Kang（First author）
Behavior Pack Optimization Post-training Multimodal Large Language Models Video Question Answering Qwen2.5-VL
arXiv Oct 2, 2026
MemLife builds entity-grounded first-person text memories read by a time-indexed agentic reader, beating the strongest training-free baseline by 4.6–12.0% on four long-horizon egocentric video benchmarks, with MemOpt adding 2.7–5.0% by training only the memory writer under the FIRM reward
The work introduces MemLife, a multimodal memory system that compacts egocentric video into time- and entity-anchored first-person text episodes retrieved by a time-indexed agentic reader, improving over the strongest training-free baseline by 4.6–12.0% across four long-horizon benchmarks without training or query-time video access, and further proposes MemOpt, which applies reinforcement learning only to the memory writer under the faithful, informative, and retrievable FIRM reward, adding 2.7–5.0% with gains that transfer across writer and reader backbones, memory systems, and out-of-domain video and question distributions.
The work introduces MemLife, a multimodal memory system that compacts egocentric video into time- and entity-anchored first-person text episodes retrieved by a time-indexed agentic reader, improving over the strongest training-free baseline by 4.6–12.0% across four long-horizon benchmarks without training or query-time video access, and further proposes MemOpt, which applies reinforcement learning only to the memory writer under the faithful, informative, and retrievable FIRM reward, adding 2.7–5.0% with gains that transfer across writer and reader backbones, memory systems, and out-of-domain video and question distributions.
University of Virginia University of Illinois Urbana-Champaign
Guangzhi Xiong（First author）
MemLife MemOpt FIRM Agentic Memory Video Question Answering

You’re all caught up

BPO replaces single-response rewards with counterfactual-view behavior packs, lifting Qwen2.5-VL-7B temporal-hard accuracy by 7.8 points

MemLife builds entity-grounded first-person text memories read by a time-indexed agentic reader, beating the strongest training-free baseline by 4.6–12.0% on four long-horizon egocentric video benchmarks, with MemOpt adding 2.7–5.0% by training only the memory writer under the FIRM reward