Translate
Live intelligence

Latest AI updates

1,754 updates
AllModelsProductsIndustryResearchTutorialOpinion
Research

Mobile AI 2026 Challenge: Learned Smartphone ISP on Mobile GPUs

This arXiv paper reports on the Mobile AI 2026 challenge, which tasked participants with designing deep learning models to reconstruct RGB photos from RAW Bayer data on smartphone GPUs. Using the Fujifilm UltraISP dataset with Sony IMX586 sensor and GFX100 camera pairs, top solutions achieved high-quality 5MP reconstruction in under 100 ms on Snapdragon and Dimensity chipsets. The paper provides a comprehensive description of the winning models.

arxiv4d agoScore 78
computational photographyefficient deep learningmobile hardware
Research

CoGA2M: A Composite GAM That Beats LightGBM Under Distribution Shift

This paper introduces CoGA2M, a multi-resolution, higher-order Composite Generalised Additive Model, and benchmarks it on the TALENT suite. CoGA2M ranks in the top third across all baselines on regression and classification tasks, matching state-of-the-art tabular methods. Under new out-of-distribution splits, CoGA2M outperforms LightGBM, highlighting a structural advantage of additive models in distribution shift.

arxiv4d agoScore 78
Generalised Additive ModelsTabular LearningDistribution Shift
Research

Mobile AI 2026 Challenge Report: Efficient RGB Photo Enhancement on Smartphone GPUs

This paper presents the Mobile AI 2026 challenge on on-device RGB photo enhancement, targeting efficient deep learning models for smartphone GPUs. Participants used the DPED dataset and optimized for Qualcomm and Arm chipsets, with top solutions enhancing Full HD images in under 300 ms. The report details the models and results, highlighting practical deployment on constrained hardware.

arxiv4d agoScore 75
Mobile AIImage EnhancementOn-device Deep Learning
Research Editor selected

ALBI: LLM-Guided Inductive Logic Programming via Iterative Generate-Test-Constrain Refinement

ALBI is a neuro-symbolic ILP framework that replaces the ASP-based hypothesis generator in the Learning from Failures paradigm with an LLM, while using SWI-Prolog as a formal verifier. It implements an iterative generate-test-constrain loop where the LLM proposes hypotheses, Prolog tests them, and diagnostic feedback drives refinement. On 30 standard ILP benchmarks, ALBI achieves 100% success across three frontier LLMs, matching Popper with only 1.4-1.7 iterations on average. Semantic anonymization tests confirm robust symbolic reasoning, with all models maintaining 100% success despite a moderate increase in iterations.

Why it matters: This paper presents a novel integration of LLMs with formal verification for ILP, demonstrating high success and robustness, which is valuable for researchers and practitioners in neuro-symbolic AI.

arxiv4d agoScore 82
Inductive Logic ProgrammingLarge Language ModelsNeuro-symbolic AI
Research Editor selected

Geometric Abstraction Reduces Multimodal Queries in Agentic Robot Policies

This paper proposes replacing step-by-step visual reasoning in multimodal robot policies with a 3D geometric abstraction that converts scene geometry into typed, evidence-aware symbolic references grounded from RGB-D data. The system refreshes metric values locally and returns compact textual checkpoints after each action, invoking direct visual reasoning only on demand. Experiments show improved multi-step manipulation completion while substantially reducing repeated multimodal queries.

Why it matters: Offers a concrete method to improve efficiency and reliability of agentic robot policies by using geometric abstraction, relevant to AI-driven robotics.

arxiv4d agoScore 85
robot manipulationmultimodal language modelsgeometric reasoning
Research Editor selected

FloMo: 3D Scene Flow as a World Action Model Intermediate for Learning from Human Video

FloMo is a 5B video-generation model fine-tuned to predict 3D scene flow and action tokens, using scene flow as an intermediate representation to transfer video pretraining to robot action prediction. Co-trained on human video and robot teleoperation data, it generalizes to unseen objects and tasks, outperforming video prediction objectives. Experiments show that predicting video wastes model capacity, as action tokens attend more to motion than video.

Why it matters: This paper presents a concrete, recent AI development with empirical evidence that 3D scene flow improves sample efficiency and generalization in robot learning from video.

arxiv4d agoScore 85
Robot learningVideo pretrainingScene flow
Research

VIGOR: Visual Gain Ordering for Hallucination Mitigation in Multimodal Discrete Diffusion Language Models

VIGOR is a training-free, plug-and-play inference strategy that reduces hallucination in multimodal discrete diffusion language models by reordering masked token positions based on counterfactual visual evidence. It compares confidence under original and visual-ablation passes to prioritize positions strongly supported by the image. Experiments on LLaDA-V and Lumina-DiMOO show consistent hallucination reduction and improved multimodal reasoning.

arxiv4d agoScore 78
multimodal diffusionhallucination mitigationtoken ordering
Research Editor selected

RFWorld: Rollout-Free World Model for End-to-End Autonomous Driving

RFWorld introduces a rollout-free world model for trajectory evaluation in end-to-end autonomous driving. It builds a shared scene memory and uses lightweight training-time temporal readout with semantic supervision to create future-readable representations, avoiding costly recursive rollout. At inference, candidates query the memory via sparse spatial readout, cutting computation and improving planning-aligned representation, achieving SOTA on NAVSIM and Bench2Drive.

Why it matters: This is a concrete research advance that reduces inference cost and improves trajectory evaluation accuracy, with benchmark results on standard driving datasets.

arxiv4d agoScore 85
end-to-end planningworld modelstrajectory scoring
Research Editor selected

LatentTriad: Pre-Action Risk Representations for Malicious Agent Skill Detection

LatentTriad is a new method that detects malicious skills in LLM agents without executing them, using hidden states from a single prefill pass. It combines contextual boundary features, group-adjusted operation directions, and departures from benign state relationships to achieve 99.0-99.2% pooled F1 across 5,830 skill packages. The method is execution-free, producing no output tokens and invoking no tools, with a compact implementation running in 259-284 ms median.

Why it matters: This paper presents a concrete, recent AI safety technique with strong empirical results and practical efficiency, valuable for developers and researchers working on agent security.

arxiv4d agoScore 85
AI safetyLLM agentsmalicious skill detection
Research

EEG Foundation Models Show Convergent Representations with Decodability

This paper tests the Platonic Representation Hypothesis on EEG foundation models, finding that models with better EEG decodability show stronger cross-family alignment in local and global structure, though not consistently with parameter count. Cross-modal convergence is modality-dependent: stronger global alignment to language but not vision, with local alignment showing no consistent association. The results suggest a dissociation between local sensory features and global semantic structure.

arxiv4d agoScore 78
EEG foundation modelsrepresentation learningcross-modal alignment
Research Editor selected

Molecular Prior Networks: Single-Pass Affinity Prediction via Transformers

The paper introduces Molecular Prior Networks, a transformer-based model that predicts drug-target affinity in a single forward pass, trained on thousands of existing affinity datasets. It matches state-of-the-art performance on the FS-Mol regression benchmark for small support sets while being faster due to no task-specific training. Comparisons show the model's performance goes beyond simple molecular similarity.

Why it matters: Presents a novel AI method for drug discovery with concrete benchmark results and practical speed advantages.

arxiv4d agoScore 82
drug discoverymolecular modelingfew-shot learning
Research Editor selected

Workload-Static Optical Circuit Switching for Scalable LLM Inference Decode

This paper proposes using workload-static optical circuit switching (OCS) for LLM inference decode, exploiting the fixed communication structure of decode to pre-configure circuits per deployment. Co-designed with model parallelism and topology-aware collective schedules, the approach turns spare accelerator ports into fewer communication phases or higher bandwidth. Evaluations on a 576-accelerator scale-up domain serving 5T-parameter MoE models show up to 65% better interactivity and 119% better energy efficiency over oversubscribed packet-switched fabrics.

Why it matters: Offers a concrete, novel networking architecture for LLM inference with significant measured gains, relevant to scalable AI systems.

arxiv4d agoScore 85
LLM inferenceoptical networkingmodel parallelism
Research

Continual AutoResearch: Do Agents Learn Across Tasks?

This paper introduces Continual AutoResearch, an environment for testing whether autonomous research agents can retain and apply knowledge across tasks. It compares persistent and reset agents across nine task families and five frontier agents, finding that persistent state improves research efficiency but not consistently final solution quality. The work highlights a gap between retaining experience and effectively using it.

arxiv4d agoScore 78
AutoResearchagent memorytask adaptation
Research Editor selected

AI-Mediated Proposals Show Instability and Hidden Alternatives in Collective Decision-Making

A study of four LLM mediators found that proposals generated from citizen opinions are unstable: re-running with different seeds or order changes assignments in 17-20% of cases, and reordering opinions affects assignments nearly as much as replacing them. Human annotators confirmed real differences, though an LLM judge overestimated frequency. Variation persists even with safeguards like temperature zero, and prompts seeking agreement underrepresent minority views. The authors propose audits to expose alternative assignments and disclose the objective behind each proposal.

Why it matters: This research reveals a concrete, practical risk in AI-mediated collective decision-making, with actionable audit recommendations for transparency.

arxiv4d agoScore 85
AI-mediated deliberationLLM reliabilitytransparency
Research Editor selected

Meta$^2$: Precomputing Reusable Knowledge for Online Harness Evolution in Long-Horizon Agents

Meta$^2$ is a new method that precomputes reusable knowledge for evolving agent harnesses (prompts, tools, memory, hooks) online during long-horizon tasks. It learns a textual Meta-Skill offline by contrasting training rollouts with and without privileged verifier information, then applies this knowledge online to decide whether to keep or evolve the harness per task. On Terminal-Bench 2.1 and DeepSWE 1.1, it achieves 81.00% and 44.44% with GPT-5.6-Terra, outperforming baselines.

Why it matters: This paper presents a novel approach to harness optimization that combines offline learning with online adaptation, showing strong benchmark results and addressing a key limitation of existing methods.

arxiv4d agoScore 85
agent harness optimizationmeta-skill learningonline evolution
Research Editor selected

Query Routing Enables Power Flexibility in LLM Datacenters

This paper proposes using query routing across LLM instances to provide power demand flexibility in datacenters without deferring requests. The authors develop an analytical framework linking query arrival rates and sequence lengths to token iteration time, residence time, capacity, and power, validated on four GPU platforms with 2.8% mean absolute error. A trace-driven simulation of a 10,000 H100 GPU datacenter shows up to 16% total power flexibility, with 90% of desired reduction achieved in ~10 seconds.

Why it matters: Provides a concrete, validated mechanism for datacenter power flexibility, directly addressing the growing energy demands of LLM inference.

arxiv4d agoScore 82
LLM inferencedatacenter power managementquery routing
Research Editor selected

WebCatalyst: High-Fidelity RL for Interactive HTML Generation

WebCatalyst is a reinforcement learning framework that improves interactive HTML generation from natural-language specs, using a game-level GUI assessor, Visual Anchor to prevent reward hacking, and Action-Trace Credit for token-level credit assignment. Trained on under 1K requests, it raises Cookie-Bench Mini Games from 68.8 to 75.3 and WebGen-Bench accuracy from 48.8 to 56.9 on Qwen3.6-35B-A3B, outperforming Claude Opus 4.7 in average rank. The work highlights feedback fidelity as key to advancing interactive HTML generation.

Why it matters: Provides a concrete, novel RL method with measurable benchmark gains and analysis, valuable for AI researchers and practitioners in code generation.

arxiv4d agoScore 82
WebCatalystinteractive HTML generationreinforcement learning
Research Editor selected

Asynchronous Robot Policy decouples VLM reasoning from fast action control

ARP is an architectural framework that separates high-capacity VLM semantic reasoning from lightweight action generation using asynchronous processes at different frequencies. It updates a semantic memory at low frequency while a Vision-Action Expert generates actions at high frequency, trained to handle staleness and execution delay. On Franka tasks, ARP improves success rate by 18%, reduces execution time by 20%, and achieves 3-5x baseline performance in dynamic simulations, while allowing the VLM to run at 10x lower frequency without significant loss.

Why it matters: This is a concrete architectural advance addressing the latency-reactivity tradeoff in VLA robots, with clear quantitative gains and practical efficiency benefits.

arxiv4d agoScore 85
robot manipulationvision-language-action modelsasynchronous processing