Artificial Intelligence
Artificial Intelligence news — Wednesday 30 September 2026
150 stories from 10 sources, every one in English. Each links to its publisher.
-
Gemini 4 Argon: our next era of frontier intelligence
Google DeepMind
-
Anthropic plugs regulation, AI leaders meet with Trump, and open models
AI Business
The AI world buzzes about stopping out-of-control agents. President Trump appears to slightly modify his anti-AI regulation stance. Open source AI could lose out. Or not.
-
"The Korean version of the core is above". .'1 body value' Elise Group, Cosmic Challenge
AI Times (Korea)
Translated from Korean
-
'RRSI' released to improve your own ‘Hans’ without re-learning of AI models
AI Times (Korea)
Translated from Korean
-
Requid AI launches ‘d1’ for decision-making.. . . . . . .
AI Times (Korea)
Translated from Korean
-
Lisa Sue AMD CEO, October 8th...Innovation with NPU Corporation
AI Times (Korea)
Translated from Korean
-
Alababa reveals the voice AI ‘Crew-Audio-3.1-Real Time’ to determine conversation timing
AI Times (Korea)
Translated from Korean
-
Google, AI responds to contributors to cost over 100 digital publishers
AI Times (Korea)
Translated from Korean
-
AI is now ‘super intelligence,’ but federal oversight is still missing
AI Business
President Trump calls AI by a new name and asserts the U.S. will not be slowing down AI development.
-
Pope Leo 14-year-old, Jenson Sullivan..."AI Dangerous Alerts Can't Go"
AI Times (Korea)
Translated from Korean
-
AI industry copes with out-of-control agents. Here’s OpenAI’s response
AI Business
The generative AI vendor delayed the release of GPT-6.1 Astra after testing exposed new safety concerns, showing why AI models need continued evaluation and enterprise runtime controls.
-
[Bulletin] Choi Su-yeon NAVER representative, UNGC attending the society
AI Times (Korea)
Translated from Korean
-
NYT "OpenAI, staff, external researcher AI security alert"
AI Times (Korea)
Translated from Korean
-
COPYRIGHT(C) 2015 Kwangyang Cyber University. ALL RIGHTS RESERVED.
AI Times (Korea)
Translated from Korean
-
2071, U.S.A., by 2029, expanding its business...
AI Times (Korea)
Translated from Korean
-
AIIA, cooperating with Korean evaluation data to expose and grow AI company
AI Times (Korea)
Translated from Korean
-
AWS provides the latest ‘Callload’ model in Seoul. ."Data in the border"
AI Times (Korea)
Translated from Korean
-
The 10th and 19th Anniversary of the Seventeenth Anniversary of the Concert...
AI Times (Korea)
Translated from Korean
-
Google, “Android and search data can be opened”... EU Commissioning Instructions
AI Times (Korea)
Translated from Korean
-
S2W, Ontology-Based Technology Alteration...’Enterprise AX’ Reduction Acceleration
AI Times (Korea)
Translated from Korean
-
"Accounts 50%"... AI Researcher 22 'Kyunggo interview' released
AI Times (Korea)
Translated from Korean
-
How Many Stories Can Your Data Tell?
Towards Data Science
How the way we represent data can change what we think the data is saying The post How Many Stories Can Your Data Tell? appeared first on Towards Data Science .
-
OpenAI, the Australian government network unauthorized access apple..."Prohibition prevention·Special investigation team composition"
AI Times (Korea)
Translated from Korean
-
Kakao, ‘AI National Library’, expands its national Pension inquiry and application
AI Times (Korea)
Translated from Korean
-
Google·MS Joins AI Data Standard ‘Apache Oshi’... ‘The Blockchain Ecosystem’
AI Times (Korea)
Translated from Korean
-
Introducing SynthID Bio
Google DeepMind
Proof of concept for watermarking AI-generated proteins while preserving biological function.
-
Insight Is Still the Currency of Data Science
Towards Data Science
Coding agents give us more time for discovery and our review practices should follow the analysis The post Insight Is Still the Currency of Data Science appeared first on Towards Data Science .
-
'Smart Ring' Oura, IPO smoke in market uncertainty
AI Times (Korea)
Translated from Korean
-
Copyright © 2018. All Rights Reserved.
AI Times (Korea)
Translated from Korean
-
Gwangju Gourmet Week, Food·Golf·Culture...
AI Times (Korea)
Translated from Korean
-
"AI's real risk" official warning on an IPO application
AI Times (Korea)
Translated from Korean
-
How to Solve Issues When You Nest Measures While Overwriting the Same Filter
Towards Data Science
In DAX, we reuse existing measures all the time when writing new measures. What happens when we write a measure based on another measure and try to change a filter already set in the nested measure? The post How to Solve Issues When You…
-
The Download: OpenAI’s chief research officer explains its hacking response
MIT Technology Review Biotech
This is today’s edition of The Download , our weekday newsletter that provides a daily dose of what’s going on in the world of technology. “We’re not going to shoot ourselves in the foot” over hack fallout, says OpenAI’s chief research…
-
Meta Launches AI Agent for Small and Medium Business...
AI Times (Korea)
Translated from Korean
-
Altman "Accounting AI Safety Without Corporate Public Press"
AI Times (Korea)
Translated from Korean
-
OpenAI additional investment attracting goal is to be ‘valued in 1900’. .Year-round sales: 95 trillion
AI Times (Korea)
Translated from Korean
-
Trump reconciliation of BIGTECH's 'Notification'...It's a reconciliation of the industry
AI Times (Korea)
Translated from Korean
-
Towards Spec-Driven Test Automation: Part 2
Towards Data Science
One run, and what it actually proves The post Towards Spec-Driven Test Automation: Part 2 appeared first on Towards Data Science .
-
“Solar Farmers”, which was turned off...It will be a rice farmer under the panel
AI Times (Korea)
Translated from Korean
-
Copyright (c) 2015 ILSHIN ELECTRONICS CO., LTD. All Rights Reserved.
AI Times (Korea)
Translated from Korean
-
“We’re not going to shoot ourselves in the foot” over hack fallout, says OpenAI’s chief research officer
MIT Technology Review Biotech
Two months after the bombshell news that a swarm of its agents had broken their containment and hacked into the computers of the AI company Hugging Face , OpenAI is still putting out fires. A steady drip of disclosures about other hacks in…
-
India Double-sided solar, up to 27.1% more power than single-sided
AI Times (Korea)
Translated from Korean
-
Disrupting a coordinated model-distillation campaign
OpenAI Blog
Learn how OpenAI disrupted a campaign to extract protected model reasoning and is strengthening defenses against adversarial distillation.
-
Helping small businesses put AI to work
OpenAI Blog
OpenAI is partnering with America’s SBDC to expand hands-on AI training and local support for small businesses, alongside a new report on how small teams are using AI.
-
[AIReader’s Letter] AI Agent Development Complete Guide Learning to Spring AI 2.0
AI Times (Korea)
Translated from Korean
-
OpenAI, a 24-hour self-adhesive agent ‘Dottz’ released...No B2B winner for Consumer
AI Times (Korea)
Translated from Korean
-
[September 29] AI ended mathematical problems...It’s a real concern to learners
AI Times (Korea)
Translated from Korean
-
"Astra Advanced Intelligence at a price of a quarter"...OpenAI, 'GPT-6.1 Sol' released
AI Times (Korea)
Translated from Korean
-
When Demonstrations Fail: Diagnosing the Limits of In-Context Learning in Large Audio-Language Models with Progressive Cue Removal
arXiv cs.CL
arXiv:2603.20433v2 Announce Type: replace-cross Abstract: While Large Audio-Language Models (LALMs) have been shown to exhibit degraded instruction-following capabilities, their ability to infer task patterns from in-context examples with…
-
Suppression Is Not Forgetting: Residual Recoverability in Visual Concept Unlearning for VLMs
arXiv cs.CV
arXiv:2604.03114v2 Announce Type: replace Abstract: Vision-language models (VLMs) may need to forget visual concepts after deployment because of privacy, copyright, licensing, safety, or policy changes. Conventional machine unlearning…
-
DocHop-QA: Towards Multi-Hop Reasoning over Multimodal Document Collections
arXiv cs.CL
arXiv:2508.15851v4 Announce Type: replace Abstract: Despite rapid progress in large language models (LLMs), current QA benchmarks still overlook the core challenge of real-world scientific information seeking: synthesizing multimodal…
-
VEX-Bench: Benchmarking Verification Complexity of LLM-Generated Misinformation
arXiv cs.CL
arXiv:2609.35028v1 Announce Type: cross Abstract: Large language models (LLMs) have made misinformation inexpensive to produce but not to verify, creating a growing asymmetry in the information ecosystem. Under tight time, labor, and…
-
Med-RADIO: Reducing All Medical Domains Into One via Multi-Teacher Distillation
arXiv cs.CV
arXiv:2609.37682v1 Announce Type: new Abstract: The rapid expansion of large-scale medical datasets and computational resources has driven significant progress in medical foundation models. Given the inherent heterogeneity of medical…
-
Exemplar2VQA: A Scalable Exemplar-Driven Visual Question Answering Generation Framework via Multi-Agent Coding
arXiv cs.CV
arXiv:2609.37655v1 Announce Type: new Abstract: Advancing spatial intelligence in Multimodal Large Language Models (MLLMs) is bottlenecked by the scarcity of complex, scalable 3D question-answer (QA) data. While manual annotation is…
-
CounterSteer: Suppressing Indirect Prompt Injection with Activation Steering
arXiv cs.AI
arXiv:2609.36570v1 Announce Type: cross Abstract: Indirect prompt injection makes an LLM agent treat untrusted retrieved text as instructions. We present CounterSteer, an inference-time defense that suppresses this behavior inside the…
-
Soft Geometric Inductive Bias for Object Centric Dynamics
arXiv cs.LG
arXiv:2512.15493v2 Announce Type: replace Abstract: Physical systems are naturally parameterized in terms of geometric entities and their transformations. While exact equivariance can be a powerful inductive bias, real-world applications…
-
Accelerated surrogate dynamics for dynamical, stochastic system evolution
arXiv cs.AI
arXiv:2609.37184v1 Announce Type: new Abstract: Dynamic simulations are an entrenched way of gaining insight into the evolution of system dynamics. Their computational cost however is often prohibitively high, especially in cases of…
-
Learning how to Forget: Fine-tuning for Long-Context Sparse Attention
arXiv cs.CL
arXiv:2608.19920v2 Announce Type: replace Abstract: A lot of prior work addressed key-value (KV) cache selection and compression by sparse attention to enable long-context inference for transformer language models without excessive…
-
MTLiquid: Enabling Efficient Multi-Task Learning using Liquid Neural Networks for Lightweight Healthcare Monitoring Systems
arXiv cs.LG
arXiv:2609.33232v1 Announce Type: new Abstract: Continuous-time sensing and monitoring with timely and accurate decision-making are critical for many real-world applications. In healthcare monitoring systems, physiological signals are…
-
Screening Is Enough
arXiv cs.AI
arXiv:2604.01178v4 Announce Type: replace-cross Abstract: We call query--key relevance absolute when its values lie on a fixed bounded scale, depend on neither competing keys nor sequence length, require no sequence-length-dependent…
-
MoGround: Measuring and Mitigating Modality Distraction in Vision-Language Models
arXiv cs.LG
arXiv:2609.33431v1 Announce Type: new Abstract: We release MoGround, a vision-language dataset spanning four visual domains in which the answer to every question is guaranteed to be available from exactly one modality. This guarantee…
-
MassAlloc Attention: Let Attention Allocate Its Own Compute
arXiv cs.CL
arXiv:2609.32712v1 Announce Type: cross Abstract: FullAttn often assigns negligible normalized mass to much of the causal score space, yet dense kernels execute the complete post-score path after forming each QK tile. We introduce MALA, a…
-
Efficient Dynamic Algorithms for Graph Neural Networks with Non-Linear Propagation
arXiv cs.LG
arXiv:2609.32929v1 Announce Type: new Abstract: Graph Neural Networks (GNNs) are widely used for representation learning on graphs, but most methods assume static topologies, making them inefficient on evolving networks where edges change…
-
Context without Commitment: Robust Dense Correspondence under Non-Rigid Deformation
arXiv cs.CV
arXiv:2609.37071v1 Announce Type: new Abstract: Non-rigid point-cloud registration aims to find the corresponding target point for each point on a deforming source surface. Point-level matching keeps the full target cloud available, but…
-
JumpStart Your Policy Learning with Lessons from 160,000 Training Runs
arXiv cs.LG
arXiv:2609.13730v2 Announce Type: replace Abstract: Reliable progress in offline policy learning depends on careful reporting, well-tuned baselines, and evaluation across diverse conditions. Prior work has shown that results can be…
-
AraDynFact: Dynamic Evaluation of Factual Knowledge in Arabic
arXiv cs.CL
arXiv:2609.35461v1 Announce Type: new Abstract: As Large Language Models (LLMs) continue to scale both in size and capabilities, their proficiency in the Arabic Language has seen significant advancement. However, a critical gap remains…
-
Dual-Mode Low-Rank Learner with Bridge-Prototype Ensemble for Vision-Language Class-Incremental Learning
arXiv cs.CV
arXiv:2609.36759v1 Announce Type: new Abstract: Benefiting from transferable visual-textual alignment, CLIP has been widely adopted for class-incremental learning (CIL). However, existing learners either repeatedly update components shared…
-
Integrating Language Models into Listened and Imagined Speech Decoding from MEG
arXiv cs.LG
arXiv:2609.31997v1 Announce Type: new Abstract: Decoding imagined speech is an important goal for brain-computer interfaces but remains challenging due to weak neural responses, low signal-to-noise ratio, and limited imagined-speech…
-
Transformer MLP Gate Thresholds Are Couplings to a Carried Reference Direction
arXiv cs.CL
arXiv:2609.31956v1 Announce Type: new Abstract: The corpus-mean direction of a transformer's residual stream is a component shared across all inputs, and is commonly removed by mean-centering before representational analysis. We present…
-
Arithmetic Simplicity in Stochastic Gradient Methods
arXiv cs.LG
arXiv:2609.32240v1 Announce Type: new Abstract: A gradient descent method is arithmetically simple if the operations are limited to $+,-, \times$, and division $x/2^t$ with integer $t$. An arthmetically simple gradient method is easy to…
-
Why Fine-Tuning Encourages Hallucinations and How to Fix It
arXiv cs.LG
arXiv:2604.15574v3 Announce Type: replace-cross Abstract: Large language models are prone to hallucinating factually incorrect statements. A key source of these errors is exposure to new factual information through supervised fine-tuning…
-
AnchorRep: Defending LLMs Against Cross-Model Adversarial Transfer via Representation Repulsion
arXiv cs.LG
arXiv:2609.32602v1 Announce Type: new Abstract: Adversarial attacks optimized on a single open-weight LLM can transfer to and jailbreak architecturally different models, allowing an attacker with white-box access to one model to compromise…
-
Understanding and Mitigating Under-Confidence in GNNs from the Final Layer
arXiv cs.LG
arXiv:2505.11335v4 Announce Type: replace Abstract: Graph Neural Networks (GNNs) have demonstrated remarkable effectiveness on graph-based tasks. However, their predictive confidence is often miscalibrated, typically exhibiting…
-
DCEmbed: Scalable Optimization over Neural Surrogates
arXiv cs.LG
arXiv:2609.33879v1 Announce Type: cross Abstract: Neural surrogates can accelerate large-scale optimization by replacing expensive or intractable model components with efficient learned approximations, but solving the resulting embedded…
-
Action Shaping: Policies Absorb What They Can Express
arXiv cs.LG
arXiv:2609.32752v1 Announce Type: cross Abstract: Reward shaping has a theorem: a potential-based term can be removed without changing the optimal policy. The same practice on the action channel, an offset added in training and dropped at…
-
Jet-Long: Efficient Long-Context Extension with Dynamic Bifocal RoPE
arXiv cs.LG
arXiv:2607.07740v4 Announce Type: replace Abstract: Modern LLMs are increasingly deployed in long-context applications such as retrieval-augmented generation, repository-level coding, and agentic workflows whose accumulated reasoning and…
-
Leveraging Soft Prompts for Privacy Attacks in Federated Prompt Tuning
arXiv cs.LG
arXiv:2601.06641v3 Announce Type: replace Abstract: Membership inference attacks (MIAs) pose a serious privacy threat in federated learning (FL). While MIAs have been extensively studied in standard FL, the recent shift toward federated…
-
Adaptive Latent Capacity for World Models
arXiv cs.LG
arXiv:2609.32921v1 Announce Type: new Abstract: We introduce Adaptive LeWorldModel (ALeWM), a world model based on a joint-embedding predictive architecture (JEPA) that learns to concentrate predictive information in compact prefixes of a…
-
Understanding and Exploiting Anisotropy in Post-Training
arXiv cs.CL
arXiv:2609.32792v1 Announce Type: cross Abstract: LLM post-training combines supervised fine-tuning (SFT), a mode-covering forward-KL objective, with reinforcement learning (RL), a mode-seeking reverse-KL objective. Frequency-weighted…
-
Locating Answer-Correctness Signals in Frozen Large Language Models
arXiv cs.AI
arXiv:2609.37700v1 Announce Type: new Abstract: Language models expose internal signals that predict whether an answer is correct, readable from a single forward pass of a frozen model without additional generations. Yet existing probes…
-
BridgeMatch: Conditional Transport Bridges in Matching Matrix Space for 3D Deformable Registration
arXiv cs.CV
arXiv:2609.11472v2 Announce Type: replace Abstract: Reliable matching between partially observed, deforming point clouds requires global context and fine geometric detail. Coarse candidate selection can exclude correct fine-level…
-
The Temporal Tug-of-War: Visualizing and Detecting RAG Conflicts in Diffusion Models via Trajectory Variance
arXiv cs.CL
arXiv:2609.31684v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) introduces a specific failure mode in discrete diffusion language models: when retrieved context contradicts parametric knowledge, the iterative denoising…
-
Convergence-Aware Pareto Selection of Covariate Scaling Transformations for Markov Deterioration Hazard Models: Evidence from Bridge Inspection Data
arXiv cs.LG
arXiv:2609.31786v1 Announce Type: cross Abstract: Maximum likelihood estimation of Markov deterioration hazard models for infrastructure asset management is sensitive to the numerical conditioning of explanatory covariates, yet production…
-
ARCagent: An Adaptive Retrieval Calibration Agent for Clinical Question Answering
arXiv cs.AI
arXiv:2609.36392v1 Announce Type: new Abstract: In diseases where clinical guidelines are incomplete, contested, or mutually contradictory, knowledge completeness and dynamic conflict-aware synthesis are two safety-critical properties that…
-
Cost-Sensitive Online Window Size Selection for Portfolio Management
arXiv cs.LG
arXiv:2609.29887v2 Announce Type: replace-cross Abstract: This paper investigates cost-sensitive online window size selection for portfolio management under changing market conditions. Specifically, we propose a two-level framework that…
-
When Keywords Drop but Classifiers Hold: Soft Refusals under KV Cache Compression
arXiv cs.LG
arXiv:2609.31678v1 Announce Type: new Abstract: KV cache compression is widely used for long context LLM inference under memory constraints, while deployed systems typically score refusals after generation with keyword filters or learned…
-
A literature-guided descriptor-based framework for filtering composition search spaces
arXiv cs.CL
arXiv:2609.31650v1 Announce Type: new Abstract: Scientific literature contains latent knowledge about materials behavior, but much of this knowledge is expressed through words, contexts, and recurring associations rather than explicit…
-
Tokens Change, Structure Endures: Spectral Watermarking for Generated Speech
arXiv cs.LG
arXiv:2609.33774v1 Announce Type: cross Abstract: Watermarking is a promising tool for establishing the provenance of AI-generated speech. While many neural audio watermarking methods rely on a separately trained watermark generator…
-
Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It
arXiv cs.AI
arXiv:2609.36585v1 Announce Type: new Abstract: Pretrained transformers use little of their depth to follow references in context. Thirteen base models reliably follow only 1.4-3.6 lines, and extra pretrained loops add little. A…
-
FineART: Fine-grained Annotated Robotic Trajectory Dataset and Vision-Language-Action Model for Bimanual Manipulation
arXiv cs.AI
arXiv:2609.36416v1 Announce Type: cross Abstract: Robots operating in real-world environments must execute complex, multi-step bimanual tasks over long horizons rather than single, isolated actions. Current manipulation datasets struggle…
-
Making Cross-Continental Federated Learning Repeatable with FLIP: a Multi-Application Study
arXiv cs.CV
arXiv:2609.36001v1 Announce Type: cross Abstract: Federated learning (FL) in healthcare remains challenging, as the overhead of rebuilding governance guarantees for every collaboration stops most projects at the proof-of-concept stage…
-
Utility-Aware Data Pricing: Token-Level Quality and Empirical Training Gain for LLMs
arXiv cs.AI
arXiv:2604.22893v2 Announce Type: replace-cross Abstract: Traditional ``row-count $\times$ quality coefficient'' approaches fail to capture the nonlinear utility of data for Large Language Model (LLM) capabilities. We present a…
-
GRAVITY: Architecture-Agnostic Structured Anchoring for Long-Horizon Conversational Memory
arXiv cs.AI
arXiv:2605.01688v2 Announce Type: replace-cross Abstract: Long-horizon memory systems increasingly improve how evidence is stored and retrieved, yet the generator must still reason over fragments whose cross-session relationships are…
-
Predicting the Financial Impact of Supply Chain Risk for Major AI-Related Semiconductor Firms: A Heterogeneous Graph Patch Transformer Approach
arXiv cs.LG
arXiv:2609.32741v1 Announce Type: new Abstract: Modern semiconductor production relies on a globally distributed, multi-tier supply chain in which financial stress at one firm spreads with a delay and eventually affects the revenue…
-
Learned Relay Representations for Forward-Thinking Discrete Diffusion Models
arXiv cs.LG
arXiv:2605.22967v4 Announce Type: replace Abstract: When Masked Diffusion Models (MDMs) generate sequences through iterative refinement, the rich internal computation over masked positions is discarded, forcing every subsequent refinement…
-
Do Audio LLMs Listen Before They Act? Diagnosing Acoustic-Context Gating in Voice Agents
arXiv cs.CL
arXiv:2609.32536v1 Announce Type: cross Abstract: Audio language models can recognize spoken commands and invoke tools, but an agent must first decide whether the acoustic and conversational context warrants action. We introduce VGBench, a…
-
Rethinking Cross-Channel Importance in Time-Series Forecasting
arXiv cs.LG
arXiv:2609.32187v1 Announce Type: new Abstract: Cross-channel modeling is central to multivariate time-series forecasting, yet channels that are statistically related, predictively useful, and actually used by a trained forecaster are…
-
On-Policy or Off-Policy Learning? A Systematic Study of Distillation Dynamics
arXiv cs.LG
arXiv:2609.35259v1 Announce Type: new Abstract: On-policy learning has been argued to reduce catastrophic forgetting, produce sparser parameter updates, and improve generalisation. However, existing comparisons between supervised…
-
K-Bench: a clinically calibrated benchmark for evaluating large language models in high-risk mental health conversations
arXiv cs.LG
arXiv:2609.15855v3 Announce Type: replace-cross Abstract: People increasingly use large language models (LLMs) for mental health support, yet their safety in evolving, high-risk conversations remains poorly characterised. We developed…
-
Chinese-Jev: Bringing System One Model to Chinese-Language Tasks
arXiv cs.CV
arXiv:2609.36965v1 Announce Type: cross Abstract: System One models such as Jev offer an efficient alternative to generative language models for tasks that require decisions rather than open-ended responses. However, existing Jev models…
-
TriO: Tri-Modal Unsupervised Occupancy World Model for Anything Perception
arXiv cs.CV
arXiv:2609.32013v2 Announce Type: replace Abstract: We present TriO, a multi-modal unsupervised world model that predicts 4D occupancy, obstacle segmentation, flow and LiDAR. In contrast to prior work, TriO utilizes three distinct sensor…
-
SAGE: A Statistical Acceptance Gate for Self-Evolving Agents
arXiv cs.AI
arXiv:2609.36043v1 Announce Type: new Abstract: Large Language Model (LLM)-based agents increasingly self-evolve by editing a persistent skill document that encodes their workflow, tool-use rules, and decision logic. This loop has two…
-
PowerZooJax: A JAX-based Power System Benchmark for Reinforcement Learning
arXiv cs.AI
arXiv:2609.36052v1 Announce Type: new Abstract: Power system operation is a safety-critical sequential decision-making problem, making it a natural testbed for reinforcement learning (RL). However, existing RL environments for power…
-
AdaST: Adaptive Coupling for Spatial-Temporal Forecasting
arXiv cs.AI
arXiv:2609.36119v1 Announce Type: new Abstract: Spatial-temporal (ST) forecasting underpins many real-world systems such as traffic, climate, and energy networks. While existing methods implicitly assume strong spatiotemporal coupling, we…
-
Distinguish or Homogenize: Last-Chance Policy Identification and Risk-Budgeted Recovery under Irreversible Resource Depletion
arXiv cs.AI
arXiv:2609.36741v1 Announce Type: new Abstract: Under irreversible resource depletion, an agent can spend resources to distinguish among latent fault models, or to change the system state so that the remaining models admit a common…
-
Aperture: Merge-Consistent Rotary States for Compressed Tokens
arXiv cs.AI
arXiv:2609.36781v1 Announce Type: new Abstract: Token compression combines content from several positions, yet rotary position embeddings usually assign the merged token one coordinate. We ask what positional information must survive later…
-
From Learner Behavior to Reusable Skills for Effective and Efficient Learner Simulation
arXiv cs.AI
arXiv:2609.37157v1 Announce Type: new Abstract: Learner simulation aims to reproduce how a particular learner behaves on new tasks. Although Large Language Models (LLMs) can generate increasingly fine-grained learning behaviors, existing…
-
Towards Breaking the Learning System Wall Using Multimodal Tutoring Transcriptions
arXiv cs.AI
arXiv:2609.36502v1 Announce Type: cross Abstract: Past research using log data has faced the "learning system wall," whereby few methods exist for generalizing models of student learning across platforms. Increasingly, online learning is…
-
Protected Cores Are Not Enough: Certifying AI-Proposed Revisions of Temporal Specifications
arXiv cs.LG
arXiv:2609.33461v1 Announce Type: cross Abstract: Runtime monitoring traditionally evaluates a specification that is fixed before execution or externally modified when requirements change. In learning-enabled and data-intensive systems…
-
CoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings
arXiv cs.LG
arXiv:2609.35645v1 Announce Type: cross Abstract: Code-switching (CS), a seamless alternation between languages within a single utterance, remains a critical challenge in automatic speech recognition (ASR). While prior works focus on…
-
Capability Provenance in Language Models: A Case Study in Social Reasoning
arXiv cs.LG
arXiv:2606.19625v4 Announce Type: replace-cross Abstract: We use training-data attribution as an interpretable tool for capability discovery, mapping which regions of the pretraining corpus support social reasoning versus STEM reasoning in…
-
Loss-Guided Pretraining Data Selection for Time-Series Foundation Models
arXiv cs.AI
arXiv:2609.37255v1 Announce Type: cross Abstract: Time series foundation models (TSFMs) are pretrained on heterogeneous collections containing billions of observations, yet their training windows are typically sampled without estimating…
-
LEGO-Anything: Coding Agents for 3D Scene Reconstruction
arXiv cs.CV
arXiv:2609.36380v1 Announce Type: new Abstract: A 3D scene reconstructed from a single image is most useful when represented not as a rendering or a fixed 3D output, but as an explicit scene program whose execution yields a scene that can…
-
Rethinking Contextualization by Reinterpreting Attention Head Channels
arXiv cs.CL
arXiv:2609.33851v1 Announce Type: cross Abstract: Contextualization, the core operation of language modeling, transmits information across words to build sentence-specific word representations. Prior works mainly study contextualization…
-
Program-Verified Self-Evolution for Vision-Language Models
arXiv cs.CL
arXiv:2609.33855v1 Announce Type: cross Abstract: Self-evolving vision-language models train on questions they generate from unlabeled images. Since these questions have no gold answers, prior methods label them by majority vote over…
-
Knowing When Thinking Is Not Enough: Teaching Small Reasoning Models to Reason Beyond Their Parametric Knowledge
arXiv cs.CL
arXiv:2609.34327v1 Announce Type: cross Abstract: Scaling test-time computation is a powerful way to improve language-model reasoning, and is particularly appealing for small reasoning models (sRMs) that are cheap to serve. However, is…
-
Parallel Tokenizers: Rethinking Vocabulary Design in Cross-Lingual Transfer of Low-Resource Languages
arXiv cs.CL
arXiv:2510.06128v3 Announce Type: replace Abstract: Tokenization forms the basis of multilingual language models, yet existing methods often limit cross-lingual transfer by mapping semantically equivalent words to different embeddings. For…
-
RMB: Reward Model Boosting Mitigates Reward Hacking
arXiv cs.LG
arXiv:2609.33221v1 Announce Type: new Abstract: Reinforcement Learning from Human Feedback (RLHF) is a powerful technique for aligning large language models (LLMs) with human preference. However, it often suffers from the reward hacking…
-
Shared Doubt: Zero-Shot Cross-Lingual Confidence Estimation for Language Models
arXiv cs.LG
arXiv:2605.31220v3 Announce Type: replace-cross Abstract: Confidence estimation (CE), i.e. quantifying the reliability of a model's prediction, has attracted great interest in the context of large language models (LLMs). However, most…
-
How Much Imprecision is Enough Imprecision in my Classifier? A Practical Elicitation Procedure
arXiv cs.LG
arXiv:2609.33352v1 Announce Type: new Abstract: Set-valued classifiers, whether derived from precise probabilities and an adapted cost function, from convex sets with a robust inference mechanism, or from conformal methods, are routine…
-
Optimal Transport Dropout for Structured Predictive Uncertainty
arXiv cs.LG
arXiv:2609.33377v1 Announce Type: new Abstract: Deterministic neural networks and neural operators provide point predictions with no intrinsic measure of reliability. Yet, predictive uncertainty may stem from irreducible outcome…
-
A Cheap Verifier is Good Enough: LLM Post-training is Robust to Erroneous Rewards
arXiv cs.LG
arXiv:2609.33467v1 Announce Type: new Abstract: When post-training large language models on tasks with semi-verifiable rewards, there are many factors (training steps, base model size, training order, data quality, verifier accuracy, etc.)…
-
Deep kernel hedging
arXiv cs.LG
arXiv:2609.34474v1 Announce Type: new Abstract: We introduce a deep kernel hedging framework that combines the flexibility of deep learning with the structural inductive bias of kernel methods. The hedging functional is restricted to a…
-
Kolmogorov-Arnold networks in nuclear binding energy prediction
arXiv cs.LG
arXiv:2407.20737v3 Announce Type: cross Abstract: This study explores the application of Kolmogorov-Arnold networks (KANs) in predicting nuclear binding energies, leveraging their ability to decompose complex multiparameter systems into…
-
When and How to Canonize: A Generalization Perspective
arXiv cs.LG
arXiv:2605.11008v2 Announce Type: replace Abstract: While invariant architectures are standard for processing symmetric data, there is growing interest in achieving invariance by applying group averaging or canonization to non-invariant…
-
Product-Aware Deterministic Rounding for Quantized Matrix Multiplication
arXiv cs.LG
arXiv:2609.31641v1 Announce Type: new Abstract: Scalar rounding decisions interact through matrix multiplication. We study deterministic product-aware rounding after scales, clipping bounds, and grids are fixed, with each active scalar…
-
TACO: Training-free Sound Prompted Segmentation via Semantically Constrained Audio-visual CO-factorization
arXiv cs.LG
arXiv:2412.01488v4 Announce Type: replace-cross Abstract: Large-scale pre-trained audio and image models demonstrate an unprecedented degree of generalization, making them suitable for a wide range of applications. Here, we tackle the…
-
Mitigating Hallucination in Large Language Models: A Capability-Oriented Survey on RAG, Reasoning, and Agentic Systems
arXiv cs.CL
arXiv:2510.24476v2 Announce Type: replace Abstract: Hallucination remains one of the key obstacles to the reliable deployment of large language models (LLMs). Although various mitigation approaches have been proposed, existing studies…
-
FlashSampling: Fast and Memory-Efficient Exact Sampling
arXiv cs.LG
arXiv:2603.15854v3 Announce Type: replace Abstract: Sampling from a categorical distribution is mathematically simple, but in large-vocabulary decoding, it often triggers extra memory traffic and extra kernels after the LM head. We present…
-
Beyond Multimodal Alignment: Shared Physical Representations Across Sensors and Action Orders
arXiv cs.LG
arXiv:2608.19492v2 Announce Type: replace Abstract: Multimodal world models are often evaluated by whether different sensors produce similar representations. However, similar representations do not necessarily imply that the models make…
-
Learning to Recorrupt: Noise Distribution Agnostic Self-Supervised Image Denoising
arXiv cs.CV
arXiv:2603.25869v2 Announce Type: replace-cross Abstract: Self-supervised image denoising methods have traditionally relied on architectural constraints, pseudo-pair constructions, or specialized loss functions to avoid the trivial…
-
Hybrid Approach for Enhancing Lesion Segmentation in Fundus Images
arXiv cs.CV
arXiv:2509.25549v3 Announce Type: replace Abstract: Choroidal nevi are common benign pigmented lesions in the eye, with a small risk of transforming into melanoma. Early detection is critical to improving survival rates, but misdiagnosis…
-
SB-TRPO: Towards Safe Reinforcement Learning with Hard Constraints
arXiv cs.LG
arXiv:2512.23770v4 Announce Type: replace Abstract: In safety-critical domains, reinforcement learning (RL) systems must satisfy strict, zero-cost safety constraints while achieving meaningful task performance. Existing model-free methods…
-
When Text Matters: Design Principles for Visual Token Pruning in Vision-Language Model
arXiv cs.LG
arXiv:2609.34861v1 Announce Type: cross Abstract: Visual token pruning has been widely studied as a practical approach to reducing the computational cost of large vision-language models. However, it struggles to preserve essential visual…
-
Towards Scalable Context-Aware Single-Cell Spatial Transcriptomics Prediction from Histology Images
arXiv cs.CV
arXiv:2609.36429v1 Announce Type: new Abstract: Predicting gene expression from H&E-stained histology images offers a scalable alternative to costly spatial transcriptomics, yet most existing methods operate at the spot level, where…
-
QwenGyre: An Elastic Reinforcement Learning Framework for Training xLong-Horizon Agents
arXiv cs.LG
arXiv:2609.33848v1 Announce Type: new Abstract: Large language model (LLM) agents increasingly undertake extreme-long (xlong) horizon tasks, where a single execution can span hours, hundreds of model--environment interactions, and nearly…
-
Does Execution Require Target KV Fidelity? A Mixed-Fidelity KV Runtime for LLM Serving
arXiv cs.LG
arXiv:2609.33536v1 Announce Type: new Abstract: Large language model (LLM) serving is increasingly constrained by the GPU memory consumed by key-value (KV) caches. Existing compression, eviction, and offloading techniques alleviate this…
-
Equivalent Flows, Unequal Learning: Clean-Latent Prediction in Transformers
arXiv cs.LG
arXiv:2605.27102v3 Announce Type: replace-cross Abstract: Flow samplers consume velocity, but the neural network can predict the clean endpoint and convert it to velocity through a fixed affine readout. We study this choice with JLT, a…
-
3-D Emissions Mapping and Social Cost Estimation for US Domestic Aviation at West Coast Hubs
arXiv cs.LG
arXiv:2609.31686v1 Announce Type: new Abstract: Existing aviation emissions inventories lack accurate trajectory data for high-resolution social cost and health impact assessment. This paper develops a 3-D emissions map by reconstructing…
-
From Behavior to Mechanism: Tracing Divergent Response Modes in Frontier Language Models
arXiv cs.LG
arXiv:2608.06578v2 Announce Type: replace-cross Abstract: Frontier language models are trained with distinct data, objectives, and safety pipelines, but whether those differences produce measurably different behavior under steering…
-
What If TSF: Reframing Time Series Forecasting as Scenario-Guided Multimodal Forecasting
arXiv cs.CL
arXiv:2601.08509v2 Announce Type: replace-cross Abstract: Recent advances in large language models (LLMs) have enabled time series forecasting to move beyond numerical observations and incorporate external information in multimodal…
-
UserProxyBench: Evaluating LLM User Simulators for Agent Benchmarks and Training
arXiv cs.AI
arXiv:2609.38043v1 Announce Type: new Abstract: Interactive agent benchmarks and multi-turn reinforcement learning increasingly place a second language model in the role of the user. This simulated user controls what information the agent…
-
Reading Is Not Leaking: Local, Auditable Measurement and Reduction of Inference Exposure from Public Footprints
arXiv cs.CL
arXiv:2609.32565v1 Announce Type: cross Abstract: Anyone with a public footprint leaks facts that were never stated, and language models make the inference cheap. We present a framework for measuring and reducing this inference exposure…
-
Nonparametric Contextual Pricing and Inventory Learning under Censored Demand
arXiv cs.LG
arXiv:2608.30944v2 Announce Type: replace Abstract: In online retailing, when a product sells out, a retailer often sees only the units sold, not how many customers would have bought it had inventory been available. However, the inventory…
-
Beyond Token Scale: Chunk-Level Sparse Autoencoders for Reliable Semantic Feature Discovery
arXiv cs.CL
arXiv:2609.35521v1 Announce Type: new Abstract: Sparse autoencoders (SAEs) expose features that help us understand and steer language models, but faithful reconstruction does not guarantee informative concepts. Token-level objectives…
-
SimWAM: A Simple World Action Model for End-to-End Autonomous Driving
arXiv cs.CV
arXiv:2608.07468v5 Announce Type: replace Abstract: In autonomous driving, World-Action Models (WAMs) have improved end-to-end planning by transferring video dynamics priors to action prediction, but many still couple planning with…
-
One Rollout Is All You Get: Fully Test-Time Adaptation for GUI Agents
arXiv cs.LG
arXiv:2609.34321v1 Announce Type: new Abstract: GUI agents are deployed with frozen weights and discard everything they experience on the job. Existing ways to update an agent's weights assume something deployment withholds: ground truth…
-
When Should a Satellite Estimate Be Changed? Stress-Testing Neural Corrections for Evapotranspiration
arXiv cs.LG
arXiv:2609.35314v1 Announce Type: new Abstract: Neural residuals can improve satellite evapotranspiration (ET) estimates, but selectors must predict when a correction helps and reject unsupported inputs. We evaluate ten-member models on…
-
A Function-Level Vulnerability Score Measures Flag Rate More Than the Model: Protocol Effects on Paired Benchmarks
arXiv cs.LG
arXiv:2609.32890v1 Announce Type: new Abstract: Language models are increasingly evaluated as vulnerability detectors, and scores reported for similar models differ widely between papers. We measured how much of that difference evaluation…
-
VISTA: Value-Informed Event Appraisal for Multimodal Emotion Conflict
arXiv cs.AI
arXiv:2609.37324v1 Announce Type: new Abstract: Conflicting emotional cues can be individually valid: a subdued voice may reflect a blocked goal while a smile satisfies a social obligation. Their interpretation depends on what the event…