Deep Thought Alignment: Trajectory-Level Latent Distillation for Video Reasoning
cs.CV, cs.AI, cs.CL
Submitted: 2026-08-17
Updated: 2026-09-01
License: http://creativecommons.org/licenses/by-nc-sa/4.0/
The gist: Large Multimodal Models (LMMs) for video reasoning have long been hindered by the high computational cost of processing vast amounts of visual information.
Terminology
Abstract
Large Multimodal Models (LMMs) for video reasoning have long been hindered by the high computational cost of processing vast amounts of visual information. This dilemma motivates the transfer of the reasoning capabilities of large models to smaller, more efficient ones. On-Policy Distillation (OPD) offers a promising solution by matching output-token distributions along student-generated trajectories. However, video reasoning often depends on evidence accumulated across multiple frames. In this context, output-level supervision only captures information expressed through token predictions and does not directly constrain the latent representations formed during reasoning. To address this limitation, we propose Latent-OPD, which augments OPD with trajectory-level latent distillation. Specifically, our method focuses on the position at the end of each trajectory, where hidden states effectively summarize the accumulated visual evidence and reasoning context. Furthermore, we introduce a progressive teacher-lookahead strategy, which aligns middle-to-late student layers with increasingly deeper teacher layers. Experiments on six video reasoning benchmarks show that Latent-OPD consistently outperforms output-only OPD. Notably, the improvements are particularly pronounced in scenarios with limited frames, long videos, or tasks requiring complex evidence aggregation. These results establish Latent-OPD as a highly effective approach to frame-efficient video reasoning.
Sources
- Toward Native Multimodal Modeling: A Roadmap
- Qwen3-VL Technical Report
- VOLD: Reasoning Transfer from LLMs to Vision-Language Models via On-Policy Distillation
- VideoLLaMA 2: Advancing Spatial-Temporal Modeling and Audio Understanding in Video-LLMs
- MatchLM2Lite: A Scalable MLLM-to-Lite Framework for Reproduced Content Identification
- Video-of-Thought: Step-by-Step Video Reasoning from Perception to Cognition
- Stabilizing On-Policy Distillation for MLLM Reasoning with Global Normalization
- Training Large Language Models to Reason in a Continuous Latent Space
- Video-MMMU: Evaluating Knowledge Acquisition from Multi-Discipline Professional Videos
- LUT: Latent Utility Training for Visual Reasoning
- Latent Visual Reasoning
- LLaVA-OneVision: Easy Visual Task Transfer
- Video-OPD: Efficient Post-Training of Multimodal Large Language Models for Temporal Video Grounding via On-Policy Distillation
- Cognitive Mismatch in Multimodal Large Language Models for Discrete Symbol Understanding
- VISD: Enhancing Video Reasoning via Structured Self-Distillation
- TangramPuzzle: Evaluating Multimodal Large Language Models with Compositional Spatial Reasoning
- Visual-Advantage On-Policy Distillation for Vision-Language Models
- Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models
- Efficient Reasoning with Hidden Thinking
- Kimi K3: Open Frontier Intelligence
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models