{"slug":"reasoning-multimodal","label":"Reasoning Multimodal","item_count":8,"day_count":7,"source_count":4,"first_seen":"2026-10-23T10:43:41+00:00","last_updated":"2026-11-06T17:32:49+00:00","generated_at":"2026-11-10T00:13:24.787725+00:00","sources":["arxiv_agent_systems_research","arxiv_cs_cl","arxiv_cs_lg","arxiv_llm_reliability"],"days":[{"date":"2026-10-23","items":[{"title":"Beyond Post-Hoc Rewards: Privileged Reasoning Exploration for Multimodal Embeddings","url":"http://arxiv.org/abs/2610.32474v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"Advances in Multimodal Large Language Models (MLLMs) have driven substantial progress in multimodal embeddings. Recent methods exploit generative capabilities by augmenting embedded inputs with self-generated reasonin...","why_it_matters":"Matches feed focus: eval.","sid":"23a2048e82ce2894","published":"2026-10-23T10:43:41+00:00","editor_note":"Examines self-generated reasoning for multimodal embeddings, arguing against post-hoc rewards."}]},{"date":"2026-10-27","items":[{"title":"When Visual Evidence Is Insufficient: How Instructions, Post-Training, and Internal Activations Shape Refusal in Multimodal Language Models","url":"http://arxiv.org/abs/2610.37430v1","source":"arxiv_cs_lg","type":"paper","summary_1line":"For reliable deployment, multimodal large language models (MLLMs) should recognize when an image provides insufficient evidence to answer a question and withhold unsupported answers. Yet refusal alone is not necessari...","sid":"1a8c5544b9b567ca","published":"2026-10-27T15:42:20+00:00","editor_note":"Studies how instructions, post-training and internal activations shape refusal when an image gives insufficient evidence."}]},{"date":"2026-10-30","items":[{"title":"Latent-Lens: Visual Perception in Small Language Models Through Latent Communication","url":"http://arxiv.org/abs/2610.41751v1","source":"arxiv_agent_systems_research","type":"paper","summary_1line":"Bridging independent vision-language models (VLMs) and small language models (SLMs) in a modular multi-agent system typically forces them to exchange information as discrete text, which creates a lossy bottleneck for...","why_it_matters":"Matches feed focus: agent, eval.","sid":"abe1652e18273a1a","published":"2026-10-30T13:53:44+00:00","editor_note":"Proposes latent communication so vision-language models and small language models avoid a lossy text bottleneck."},{"title":"Latent Reasoning in Continuous Space for Unified Multimodal Models","url":"http://arxiv.org/abs/2610.42149v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"Unified Multimodal Models (UMMs) achieve remarkable success on diverse multimodal understanding and generation tasks, yet they still struggle to effectively leverage language-based chain-of-thought reasoning for image...","why_it_matters":"Matches feed focus: eval.","sid":"282f057baf2d5931","published":"2026-10-30T17:59:43+00:00","editor_note":"Brings continuous-space latent reasoning to unified multimodal models."}]},{"date":"2026-11-02","items":[{"title":"MIRA: Reinforcing Multimodal Reasoning via Deceptive Contextual Augmentation","url":"http://arxiv.org/abs/2611.03087v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"While Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced multimodal reasoning, existing frameworks suffer from a critical \"perception-reasoning gap.\" Due to an over-reliance on seemingly relevant visual tokens, even minor perceptual perturbations can","sid":"526c07c3667e6cf7","published":"2026-11-02T13:09:45+00:00","editor_note":"Targets the perception-reasoning gap in RLVR with deceptive contextual augmentation."}]},{"date":"2026-11-03","items":[{"title":"VIGOR: Visual Gain Ordering for Hallucination Mitigation in Multimodal Discrete Diffusion Language Models","url":"http://arxiv.org/abs/2611.05066v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Multimodal discrete diffusion language models (dLLMs) generate responses through iterative mask prediction and offer a promising alternative to autoregressive large vision-language models. However, they still suffer from hallucination, producing outputs that are plausible under l","sid":"57b08f9866a26766","published":"2026-11-03T15:28:11+00:00","editor_note":"Addresses hallucination in multimodal discrete diffusion language models."}]},{"date":"2026-11-04","items":[{"title":"Reasoning-Aware Cross-Modal Association Unlearning in Multimodal Large Reasoning Models","url":"http://arxiv.org/abs/2611.06168v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Multimodal large reasoning models (MLRMs) integrate visual inputs with textual and parametric knowledge, allowing sensitive information to be accessed through cross-modal associations during reasoning. This raises privacy concerns when a visual input should no longer provide acce","sid":"c43d3ff3aa20c41c","published":"2026-11-04T08:48:07+00:00","editor_note":"Adds privacy-oriented unlearning of cross-modal associations in multimodal reasoning models."}]},{"date":"2026-11-06","items":[{"title":"ROSE: Risk-Aware Orthogonal Subspace Navigation for Lifelong Knowledge Editing in Multimodal Large Language Models","url":"http://arxiv.org/abs/2611.09770v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Multimodal Large Language Models (MLLMs) require continuous factual updates to stay current, yet lifelong knowledge editing remains a daunting challenge. The primary obstacle is the severe Locality Erosion and Catastr...","sid":"5606760643e4651e","published":"2026-11-06T17:32:49+00:00","editor_note":"Extends the thread to lifelong knowledge editing for multimodal LLMs."}]}],"editorial":{"tldr":"Late-October papers focused on getting reasoning into multimodal models: reasoning-augmented embeddings, latent communication between vision and small language models, and continuous-space reasoning. By early November the research shifted to reliability, covering refusal, perception robustness, hallucination and unlearning.","stale":false,"whats_new":"ROSE (Nov 6) proposes risk-aware orthogonal subspace navigation for lifelong knowledge editing in multimodal LLMs, targeting locality erosion and catastrophic forgetting.","why_it_matters":"Multimodal models are being pushed toward deployable behavior: knowing when to refuse, resisting perceptual perturbation, avoiding hallucination, and being updated or unlearned after release.","take_for_builders":"If you serve multimodal models, add evals for insufficient-evidence refusal and perceptual perturbation before adopting any of these techniques; treat them as preprints until independently reproduced.","status":{"state":"Developing","tone":"rising","changed":"2026-11-06","detail":"A cluster of arXiv papers moving from adding reasoning to multimodal models toward making them reliable and maintainable; all items are preprints."},"beats":[{"kicker":"REASONING IN","tone":"launch","headline":"Reasoning is added to embeddings, small-model perception and unified models","summary":"Three papers on using reasoning to improve multimodal representation and perception.","sids":["23a2048e82ce2894","abe1652e18273a1a","282f057baf2d5931"]},{"kicker":"THE TURN","tone":"turn","headline":"Focus shifts to refusal, robustness and hallucination","summary":"Papers examine when models should withhold answers and how reasoning fails under perceptual perturbation or hallucination.","sids":["1a8c5544b9b567ca","526c07c3667e6cf7","57b08f9866a26766"]},{"kicker":"NOW","tone":"now","headline":"Post-deployment control: unlearning and knowledge editing","summary":"Work on removing and updating what multimodal models know.","sids":["c43d3ff3aa20c41c","5606760643e4651e"]}],"open_questions":["Do any of these methods get released with code or evaluations on production-scale multimodal models?","Can unlearning and knowledge-editing methods preserve reasoning quality in multimodal reasoning models?"],"generated_at":"2026-11-10T00:30:00+00:00"}}