Search: “routing”

Visual Tokens on Demand: Visual Budgeting for Scalable and Efficient Multi-Image Reasoning in Multimodal LLMs

arXiv:2610.18858v1 Announce Type: new Abstract: Multimodal large language models have advanced rapidly, but most remain optimized for single-image inputs. Multi-image reasoning is costly because each

local-aiarxiv-cs-cv

14 Oct 2026

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Research

When Dynamic Scaling Fails: A Scale-State Feedback Governs Recurrent-State Quantization in Linear-Attention and SSM LLMs

DGX agent

arXiv:2610.15139v1 Announce Type: new Abstract: Linear-attention and state-space LLMs carry long context in a fixed-size recurrent state, and quantizing that state is the natural way to cut its cost d

researcharxiv-cs-lg

14 Oct 2026

Which Geometry on Which Layer? Measured Sign-Spectral Assignment in Mixed-Optimizer Training

arXiv:2610.17285v1 Announce Type: new Abstract: Modern neural-network training uses mixed recipes, assigning AdamW-like or sign-based updates to embedding and output layers and spectral updates to hid

researcharxiv-cs-lg

14 Oct 2026

An Input-Dependent Readout Phase for Resonant Sequence Models

arXiv:2610.12496v1 Announce Type: cross Abstract: Sequence models with a fixed-size state avoid the quadratic cost of attention, but face a design choice. An input-dependent state transition gives con

Attention Sees, MLP Refuses: Tracing the Attention-to-MLP Safety Pathway in LLMs

arXiv:2610.13078v1 Announce Type: new Abstract: Safety-aligned large language models (LLMs) remain vulnerable to jailbreaks and can over-refuse benign requests. Understanding these failures requires d

Agents

Competence-Aware Orchestration for LLM-Based Multi-Agent Systems

arXiv:2610.13144v1 Announce Type: new Abstract: LLM-based multi-agent systems (MAS) have emerged as a promising paradigm for enhancing complex reasoning through role specialization and collaboration,

agentsarxiv-cs-ai

DistressBench: Evaluating Multi-Dimensional Crisis Support in Large Language Models

arXiv:2610.12758v1 Announce Type: new Abstract: Language models have been adopted faster than evaluation practices that reliably assess their risks, and a growing share of use is affective rather than

safetyarxiv-cs-cl

From Benchmark to Bench: Can Agents Survive Real-World Drug Discovery?

arXiv:2610.06411v2 Announce Type: replace Abstract: Agentic systems increasingly coordinate molecular-design tools, but it is unclear which layer of the stack limits outcomes on real projects. We deve

Hindsight Structured Advice Distillation: Steering Executor Agents with Advisors Trained on Their Failures and Task Characteristics

arXiv:2610.13000v1 Announce Type: new Abstract: Language-model agents are used for specialized work such as financial research, healthcare and law. Such agents are often capable enough for a task but

Humanize Olympiad Agents: A Multi-Session Workflow for Olympiad Problem Solving with Solve, Review, Wiki, Verify

arXiv:2610.13632v1 Announce Type: new Abstract: Large language model (LLM) agents are rapidly improving on long-horizon scientific and olympiad tasks. Recent systems have produced a Lean-formalized re

JoyInside-VLM-Agent: Product-Conditioned Multimodal Policy Learning for Embodied Agents

arXiv:2610.12957v1 Announce Type: cross Abstract: Embodied agents must translate multimodal observations and user requests into executable actions while adapting to product constraints, changing goals

safetyarxiv-cs-ai

Knowing When to Stop: Efficient Context Processing via Latent Sufficiency Signals

arXiv:2502.01025v4 Announce Type: replace Abstract: Large language models (LLMs) process entire input contexts indiscriminately, which is inefficient when the information required to answer a query is

MetaJoin: Adaptive Join Selections for Heterogeneous Tables

model-releasesarxiv-cs-lg

arXiv:2512.02444v2 Announce Type: replace-cross Abstract: Joining heterogeneous tables requires different forms of correspondence, including exact agreement, syntactic transformation, semantic similar

Hardware

RAMP: Recurrent Adaptive Mixed Precision for Efficient Prefill in Recurrent-Depth Language Models

arXiv:2610.13041v1 Announce Type: new Abstract: Recurrent-depth language models improve inference-time reasoning by repeatedly applying a shared network to refine internal representations, but the add

hardwarearxiv-cs-cl

Taking Tokenization off the Host for Agentic LLM Serving

arXiv:2610.13325v1 Announce Type: cross Abstract: Agentic workloads resend long contexts that barely change between calls. Prefix caching now skips most of their prefill, yet every request must still

Tracing and Taming Hallucinated Terms via Subsequence-Association Filtering and Breadth-Aware Selection for LLM Query Expansion

arXiv:2610.13810v1 Announce Type: cross Abstract: Large language models (LLMs) are now a standard source of query expansion terms but they hallucinate. They inject non existent entities and off-topic

safetyarxiv-cs-cl

When LLMs Sit Above Diagnostic Tools: Unrealized Complementarity in Industrial Fault Diagnosis

arXiv:2610.05031v2 Announce Type: replace Abstract: Large language models are increasingly used as integration layers above specialized tools, but a stronger component does not necessarily produce a s

researcharxiv-cs-cl

Causal-fate dynamics of unrealized influence

arXiv:2610.11422v1 Announce Type: cross Abstract: Many dynamical systems generate influences whose consequences are not fully exhausted in the realized trajectory at the moment they arise. Such conseq

local-aiarxiv-cs-lg

EgoPhys: Estimating Peak Contact Force and Mechanical Work from Egocentric Manipulation Video

arXiv:2610.11347v1 Announce Type: new Abstract: Physically grounded manipulation of articulated objects requires understanding both the maximum forces encountered during contact and the work performed

researcharxiv-cs-cv

Applications

Masked Generative Motion Planning with Geometry-Guided Token Search

arXiv:2610.10646v1 Announce Type: cross Abstract: Generative motion planners typically use learned trajectory priors for initial generation, while leaving test-time repair to local continuous refineme

applicationsarxiv-cs-ai

Recompose and Refine Latent Reasoning Flows for Vision-Language-Action Models

arXiv:2610.12090v1 Announce Type: new Abstract: Latent reasoning enables vision-language-action (VLA) models to transform multimodal observations into task-relevant internal states before generating c

safetyarxiv-cs-ai

SpaceFlow: Locally Controllable 3D Generation

arXiv:2610.12399v1 Announce Type: cross Abstract: Current 3D generation methods lack explicit local control: geometric adherence is often defined by a global control strength, and appearance cannot be

TAPNAV: Humanoid Navigation through Tactile Active Perception

arXiv:2610.10748v1 Announce Type: new Abstract: Navigation in vision-denied environments is challenging for humanoid robots because proprioceptive odometry drifts and localization uncertainty accumula

researcharxiv-cs-ro

Itgan at NADI 2026 shared task: Parameter-Efficient Whisper Adaptation for Robust, Mixed-Dialect and Code-Switched Arabic ASR

arXiv:2610.09934v1 Announce Type: new Abstract: We describe the Itgan systems for the three ASR subtasks of NADI 2026, namely robust country-level ASR (1.1), mixed-dialect ASR (1.2), and Tunisian code

Agents

Multi-Agent Coordination via Support-Preserving Distillation

arXiv:2610.10087v1 Announce Type: new Abstract: Offline MARL increasingly relies on generative policies to model multimodal joint behavior, typically by distilling a centralized teacher into decentral

agentsarxiv-cs-lg

Quad-State Safety Evaluation of Open-Weight Large Language Models on Non-Canonical Inputs

arXiv:2610.09033v1 Announce Type: new Abstract: Standard safety evaluations of large language models assess harmful requests written in canonical plain text, while models in real-world deployment rout

Spatial Induction Heads: In-Context Learning of Multidimensional Cellular Automata

arXiv:2610.09124v1 Announce Type: new Abstract: Induction heads provide a mechanistic account of in-context learning in sequential data, but existing theory largely assumes that the context relevant t

safetyarxiv-cs-lg

When Rank Rises as LLMs Degrade

arXiv:2610.09647v1 Announce Type: cross Abstract: Post-training adapts language models in non-stationary environments. Practitioners monitor representation health with RankMe and related spectral stat

researcharxiv-cs-cl

Agent in a Bottle: Can LLM Agents Turn Their Capabilities Into Cheap, Scalable Artifacts?

arXiv:2610.08775v1 Announce Type: new Abstract: Large language models (LLMs) can solve many narrow tasks, but querying them separately for millions of related instances can be prohibitively expensive.

Anchor-driven Multi-modal Multi-scale Expert Selection for Survival Prediction

model-releasesarxiv-cs-cv

arXiv:2610.07694v1 Announce Type: new Abstract: The integrative analysis of histopathological Whole-Slide Images (WSIs) and transcriptomic profiles holds significant promise for cancer survival predic

Building Rome from a Single Image

arXiv:2610.08790v1 Announce Type: new Abstract: Single-image scene generation aims to produce a complete 3D scene mesh from a single image, including surfaces the camera did not observe. While pretrai

researcharxiv-cs-cv

Compositional Concept Erasure in Text-to-Image Diffusion Models via Hierarchically Grounded Semantic Surgery

arXiv:2610.07337v1 Announce Type: new Abstract: Removing copyrighted, unsafe, or user-specified concepts from a deployed text-to-image diffusion model is now a practical requirement. Weight-editing me

researcharxiv-cs-cv

Agents

Decoupling What from Where: How Should a Small GUI Grounding Model Receive the Action Type?

arXiv:2610.07444v1 Announce Type: cross Abstract: A GUI agent decides which action to take and where to take it; we ask how a small grounding model should receive the action type. Fine-tuning Qwen2-VL

agentsarxiv-cs-cv

DIPrune: Task-Aware Token Pruning with Dual Importance for Efficient Multimodal Language Models

model-releasesarxiv-cs-cv

arXiv:2610.08341v1 Announce Type: new Abstract: Recent training-free pruning approaches for Multimodal Large Language Models (MLLMs) effectively cut computational overhead by exploiting visual redunda

Dynamic Budget Allocation for LLM Evaluation under Hard Resource Constraints

model-releasesarxiv-cs-lg

arXiv:2610.07362v1 Announce Type: new Abstract: We evaluate large language models (LLMs) in multi-turn interactions through their time-to-event: the number of interaction steps required to produce an

DySCo: Dynamic Sharding for Collaborative Edge-Cloud LLM Inference with Depth-Synchronized Batching

arXiv:2610.08268v1 Announce Type: cross Abstract: Pervasive intelligent applications are increasingly deployed on mobile and Internet of Things (IoT) edge devices. Consequently, Large Language Models

local-aiarxiv-cs-ai

Evidence-Bound Reasoning: Neuro-Semantic Verification of Biomedical AI in Glioblastoma Radiogenomics

arXiv:2610.08660v1 Announce Type: new Abstract: Background: Biomedical AI can generate plausible explanations without reliably verifying whether each statement is supported by patient-specific evidenc

Federated Mixture-of-Experts Alignment on Mobile Edge Networks under Data Heterogeneity

arXiv:2603.21276v2 Announce Type: replace-cross Abstract: The growing demand for on-device large language model (LLM) services on mobile edge devices has driven the adoption of Mixture-of-Experts (MoE

From Shared Demand Patterns to Local Uncertainty: Probabilistic Load Forecasting by Mixing Compact Adaptations

arXiv:2610.08538v1 Announce Type: cross Abstract: Probabilistic load forecasting has been widely studied for power-system operation and planning, but customer- and transformer-level forecasting introd

local-aiarxiv-cs-ai

Have I Seen Enough? Frozen Video-Language Models Encode Evidence Readiness

arXiv:2610.08560v1 Announce Type: cross Abstract: Streaming video-language models must decide not only what to answer, but whether the evidence needed for the current question has arrived. Existing sy

Holdout Best-of-N: Unbiased Evaluation and Its Cost

arXiv:2610.08719v1 Announce Type: new Abstract: Reusing the scores that select a Best-of-N winner can overstate its expected reward. We study evaluation from a fixed matrix of K independent scores per

safetyarxiv-cs-cl

Linear Fitness Subspace in Protein Language Models Enables Sample-Efficient Directed Evolution

arXiv:2610.07607v1 Announce Type: cross Abstract: Model-guided directed evolution seeks to identify high-fitness protein variants under limited oracle budgets. Protein language models (PLMs) provide r

Loud and Clear: Dynamic Activation Steering for Improving Speech Intelligibility in Noisy Environments

arXiv:2610.07647v1 Announce Type: cross Abstract: Speech becomes less intelligible in noisy environments, and humans naturally adapt their voice to compensate. Inspired by this behavior, we investigat

researcharxiv-cs-cl

MacJEPA: Missingness-Robust Audio-Visual Recognition from Untrimmed Egocentric Videos

arXiv:2610.08192v1 Announce Type: new Abstract: Audio-visual models improve egocentric action recognition by exploiting complementary cues, yet typically assume that both streams remain available at i

local-aiarxiv-cs-cv

MedZERO: Self-Evolving Agents for Open-Ended Medical Reasoning Through Controlled Knowledge Accumulation

arXiv:2610.08327v1 Announce Type: new Abstract: Large language models (LLMs) have shown promise in medical question answering and clinical reasoning, yet their improvement remains constrained by stati

Rethinking Adapter Placement: A Dominant Adaptation Module Perspective

arXiv:2605.06183v2 Announce Type: replace Abstract: Low-rank adaptation (LoRA) is a widely used parameter-efficient fine-tuning method that places trainable low-rank adapters into frozen pre-trained m

Small Language Models for Smart Data Model Classification at the Edge: A Cost-Aware Hybrid Approach

arXiv:2610.07093v1 Announce Type: new Abstract: The rapid proliferation of heterogeneous data sources within the Internet of Things (IoT) across domains such as smart cities, energy management, and en

Tool-calling retrieval versus vector RAG for a small Greek--English knowledge base: accuracy and robustness to how users type Greek

arXiv:2610.08205v1 Announce Type: cross Abstract: Assistants grounded in a small, frequently edited knowledge base can retrieve through tool calls to a live data interface or through vector retrieval-