mlsanityData to 7 Oct 2026

Certified Interface Aliases: Exact Collisions in Vision-Language Preprocessing, and When They Exist

arXiv 2609.33003 ↗First posted 26 Sep 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Vision-language verifiers and routers must distinguish errors repairable by more reasoning from those caused by visual evidence never reaching the language model. This distinction lacks ground truth because annotators see full-resolution images while models receive preprocessed tensors. We introduce AliasForge to create cases where the relevant fact is provably absent from the interface. Fixed-point resampling makes the pre-rounding resize an exact integer linear map that can send nonzero integer perturbations to zero. Hiding a label-flipping perturbation there produces images with opposite step-correctness labels but bit-identical interface states. Every verifier therefore has the same output law on both members, giving pair-balanced accuracy exactly one half and zero gain from language-side repair. We prove that every fixed-point downscaler has such null vectors and bound their smallest size at the most common ratios, which rules out an 8-bit fit whenever the bound exceeds 255. From the resize configuration alone, a lattice criterion supplies realizable collisions and certifies their absence within the specified construction family. It resolves all 20 screened configurations, 17 as constructible and 3 as non-constructible. We construct certified pairs across three architectures and certify four additional processors, with zero decision-logit gap on all 18 scored pairs and none of the 18 controls. The pairs also screen routers that waste computation on re-attention or further reasoning. On natural items, per-item routing headroom exists, but no tested interface-only router improves over stopping. Our fiber ceiling bounds the headroom recoverable from the interface. Code: https://github.com/KurbanIntelligenceLab/aliasforge.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 39 of its 51 references that are in the map. First posted on arXiv 26 Sep 2026.

Builds on

39 of 51 references.

References · 39

  1. 2023Improved Baselines with Visual Instruction Tuningcited by 4,362
  2. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,151
  3. 2024LLaVA-OneVision: Easy Visual Task Transfercited by 2,669
  4. 2024How Far Are We to GPT-4V? Closing the Gap to Commercial Multimodal Models with Open-Source Suitescited by 1,033
  5. 2024Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMscited by 719
  6. 2019Designing and Interpreting Probes with Control Taskscited by 608
  7. 2023V*: Guided Visual Search as a Core Mechanism in Multimodal LLMscited by 459
  8. 2017Mitigating Adversarial Effects Through Randomizationcited by 382
  9. 2023Patch n' Pack: NaViT, a Vision Transformer for any Aspect Ratio and Resolutioncited by 226
  10. 2024Vision language models are blind: Failing to translate detailed visual features into wordscited by 144
  11. 2025VisualPRM: An Effective Process Reward Model for Multimodal Reasoningcited by 134
  12. 2025PRMBench: A Fine-grained and Challenging Benchmark for Process-Level Reward Modelscited by 98
27 more
  1. 2021Bad Characters: Imperceptible NLP Attackscited by 69
  2. 2024Learning How Hard to Think: Input-Adaptive Allocation of LM Computationcited by 67
  3. 2025Seeing but Not Believing: Probing the Disconnect Between Visual Attention and Answer Correctness in VLMscited by 48
  4. 2025Qwen Look Again: Guiding Vision-Language Reasoning Models to Re-attention Visual Informationcited by 27
  5. 2025Perceptual-Evidence Anchored Reinforced Learning for Multimodal Reasoningcited by 20
  6. 2025Lost in Embeddings: Information Loss in Vision-Language Modelscited by 18
  7. 2025Training Vision-Language Process Reward Models for Test-Time Scaling in Multimodal Reasoning: Key Insights and Lessons Learnedcited by 12
  8. 2026When and How Much to Imagine: Adaptive Test-Time Scaling with World Models for Visual Spatial Reasoningcited by 12
  9. 2026VLMs Need Words: Vision Language Models Ignore Visual Detail In Favor of Semantic Anchorscited by 12
  10. 2026Are VLMs Seeing or Just Saying? Uncovering the Illusion of Visual Re-examinationcited by 12
  11. 2024Intriguing Equivalence Structures of the Embedding Space of Vision Transformerscited by 8
  12. 2026Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning Optimizationcited by 6
  13. 2026Bad Seeing or Bad Thinking? Rewarding Perception for Multimodal Reasoningcited by 6
  14. 2026Improving Visual Reasoning with Iterative Evidence Refinementcited by 5
  15. 2026From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Modelscited by 5
  16. 2026TikArt: Stabilizing Aperture-Guided Fine-Grained Visual Reasoning with Reinforcement Learningcited by 4
  17. 2026Deeper Thought, Weaker Aim: Understanding and Mitigating Perceptual Impairment during Reasoning in Multimodal Large Language Modelscited by 4
  18. 2026On Test-Time Scaling for Vision-Language Modelscited by 4
  19. 2024Unaligning Everything: Or Aligning Any Text to Any Image in Multimodal Modelscited by 3
  20. 2025LLMs Can Compensate for Deficiencies in Visual Representationscited by 3
  21. 2026Improving Vision-language Models with Perception-centric Process Reward Modelscited by 3
  22. 2025Chameleon: Adaptive Adversarial Agents for Scaling-Based Visual Prompt Injection in Multimodal AI Systemscited by 2
  23. 2026Disentangling Perception and Reasoning in Multimodal LLMs via Reward Designcited by 2
  24. 2026A Fiber Criterion for Representation Identifiability in Supervised Learningcited by 2
  25. 2026Steal the Patch Size: Adversarially Manipulate Vision-Language Modelscited by 2
  26. 2026Visual Access Boundaries in Vision-Language Model Reasoningcited by 2
  27. 2026Failing to See or Failing to Know? Attributing Errors in Vision-Language Modelscited by 1

Built on by

Not cited yet.