Entity · paper

Test-Time Training for Modality Order Consistency in Vision-Language Models

paperactiveprovisionaltest-time-training-for-modality-order-consistency-in-vision-language-models-9721b0f0·1 events·first seen 4d ago

Aliases: Test-Time Training for Modality Order Consistency in Vision-Language Models

More like this (12)

ENTRAP-VL: A Taxonomic Probe for Dual Contextual Entrainment in Vision-Language Models contrastive vision-language pretraining Modeling Complex Behaviors: Multi-Personality Composition and Dynamic Switching in Vision-Language Models visual language model Vision-Language Models Training-Free Semantic Correction for Autoregressive Visual Models Visually Grounded Self-Reflection for Vision-Language Models via Reinforcement Learning Evolution of Accuracy and Visual-Cognitive Errors in a Decade of Vision-Language AI Models TempoVLA: Learning Speed-Controllable Vision-Language-Action Policies How Robust is OCR-Reasoning? Evaluating OCR-Reasoning Robustness of Vision-Language Models under Visual Perturbations LabVLA: Grounding Vision-Language-Action Models in Scientific Laboratories Same Evidence, Different Answer: Auditing Order Sensitivity in Multimodal Large Language Models

Recent events (1)

5arXiv · cs.CL·4d ago·source ↗

Test-time training corrects modality order sensitivity in vision-language models

Researchers find that vision-language models consistently perform better when images are presented before questions rather than after, a semantically irrelevant prompt-order artifact that persists across three models and three benchmarks. The authors propose an order-consistent test-time training method that closes this gap and also improves the stronger image-first baseline. Activation patching localizes the failure to a narrow mid-network region where representations diverge between orderings, framing modality-order sensitivity as a circuit-level failure. The work contributes both a diagnostic finding about VLM robustness and a lightweight mitigation technique.

Evaluation and Benchmarking Multimodal Progress Test-Time Training for Modality Order Consistency in Vision-Language Models