Researchers have introduced SeT-Diff, the first foundational model for high-performance computing (HPC) node telemetry and time-series data, achieving a Mean Absolute Error (MAE) of 0.0470 on reconstruction tasks. The di…
Researchers present QFoldAgent, a closed-loop multi-agent framework for 5-residue tetrahedral-lattice protein folding that uses a design agent to propose sequence-conditioned Hamiltonian penalties, a VQE-based quantum-cl…
A new AI agent called DeepLens Diagnosis Agent, designed by researchers and centered on the small medical reasoning model JSL Medical Small 7B v2, achieved 60.14% top-1 diagnostic accuracy on the 915-case DiagnosisArena …
Researchers propose MIITA, a Memory-Induced Inference-Time Adaptation framework that enables small language models to perform continual learning under constrained storage without catastrophic forgetting. MIITA stores sup…
Researchers introduce program distillation as a cheaper, faster, and more transparent alternative to LLM-as-a-judge for automated evaluation, presenting PAJAMA, a system that synthesizes programmatic judges from an LLM's…
Researchers propose SF-AMS, a framework for strategic forgetting in LLM agent memory systems that replaces static retrieval with a utility-driven survival mechanism, achieving consistent gains over baselines including Li…
Researchers extended AssetOpsBench with a Smart Grid Transformer asset class and four IEC-grounded diagnostic tools, and introduced ScenarioGeneratorAgent, a pipeline for synthetic industrial-agent scenario generation th…
Researchers propose a loss-aware feature-map pruning framework for convolutional neural networks that uses multi-armed bandits (UCB1 and Thompson Sampling) to identify and remove redundant feature maps while preserving a…
A new middleware layer called Schema-Aware Localisation (SAL) achieves 62.6% execution-grounded truth (EGT) on 500 TPC-H natural language questions against Oracle Autonomous Database 23c using GPT-4o-mini, up from 2.2% w…
Researchers propose DSTFView, a dual-input spatio-temporal-frequency multi-view workload forecasting framework for collaborative cloud-edge environments, which jointly models closeness and period dependencies and extract…
Researchers released MedLoCoMo, a medical long-context benchmark for evaluating large language models on patient-specific clinical reasoning over multi-admission dialogues, built from deidentified MIMIC-IV records. The b…
A new empirical study from arXiv reveals that prompt wording significantly affects energy consumption for on-device large language models (LLMs), with imperative keywords and instruction structure influencing decoding le…
A new execution-grounded red-team testing framework reveals that coding agents integrated into system operations can be induced to carry out unsafe actions on the surrounding environment, with verified unsafe execution r…
A new framework called SCEPTER (Single-Case Evidence-driven PubMed-To-rEcommendation Reasoner) reduces clinical literature search from 576 papers to 53 retained papers, 7 Pareto-optimal claims and 3 final recommendations…
Researchers introduced KANEx, the first framework leveraging Kolmogorov-Arnold Networks (KANs) to ground Vision-Language Model (VLM) reasoning for medical explainability, achieving a 10% improvement in visual localizatio…
A new framework called Cloud Decoy AI Agent pairs high-fidelity cloud decoys with autonomous language model agents to compress intrusion investigation from suspicious activity to analyst-ready reports, addressing challen…
Researchers propose MXAttention, a data-free post-training quantization framework for MXFP4 attention that introduces Universal Optimal Scaling (UOS) with a distribution-independent optimal scaling boundary Qmax=7.25 and…
AI systems already govern by ranking speech, filtering applicants, and triaging claims, but the dominant frame of alignment fails to address their legitimacy, according to a new article on arXiv (2607.24391v1). The artic…
A new hybrid model called MTSF-ANO, integrating variational quantum circuits with adaptive non-local observables, ranks first or second in mean squared error in 17 of 20 settings on the four ETT datasets, improving over …
Researchers propose DraftExpert, an expansion-aware self-speculative decoding framework for expert-offloaded Mixture-of-Experts (MoE) inference on end devices, achieving 1.45x average decode throughput improvement, 84-87…