Search signals, briefings, benchmarks and glossary terms.
Search signals, briefings, benchmarks and glossary terms.
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
Chronofy, a new RAG architecture, introduces a temporal-logical decay mechanism to prevent LLMs from using obsolete facts, addressing 'temporal hallucination'.
Research introduces CASC, a new deep subspace clustering method for multivariate spatiotemporal data, focusing on causal dependencies and dynamics.
Research introduces heat-kernel entropy profiles for weighted measures on manifolds to account for geometric support in effective sample size.
Research finds AI assistants, when used as tutors, can over-intervene, hindering user learning and cognitive engagement during problem-solving.
ReliableTableQA is a framework training LLMs to annotate the statistical reliability of tabular QA results, assessing if answers are statistically meaningful.
Researchers propose a counterfactual explainability framework using CycleGAN for retinal disease classification to link model decisions to retinal regions.
New research introduces DFAH-Bench, a replay benchmark designed to measure observable behavioral instability in financial agent decision-making processes.
Researchers propose CEDAR, a constraint-based method for lagged causal edge discovery in sparse autoregressive time series data.
New research from arXiv proposes a novel approach to evaluate LLM performance by aligning models with diverse, real-world user expectations.
DC-Leap proposes a training-free framework to accelerate Diffusion Large Language Models (dLLMs) by addressing conservative confidence thresholds.
Researchers developed a neural-symbolic pipeline combining compact transformers with a linguistically-informed mediator for gaming toxicity detection.
Researchers released GLAN-QnA-KR, a 303K-row Korean instruction-QA corpus generated by Microsoft's Phi-3.5-MoE-instruct model.
Domyn-Small, a 10B parameter multilingual open-weight reasoning LLM developed in Europe, achieves 32K context window via continued pre-training.
Research introduces an attention-based experience replay framework for continual learning in time series forecasting models, addressing data distribution shifts.
Research proposes using B-Splines to improve the smoothness and training of Neural Temporal Point Processes (TPPs) for modeling event sequences.
A research paper proposes a knowledge-injection framework to improve zero-shot delirium prediction using LLMs and electronic health records.
Researchers propose Multimodal CoLRAG-TF, a quadruple-fusion RAG architecture combining text, keywords, knowledge graphs, and image similarity for complex PDFs.
SonicSampler presents unified tile-aware kernels for LLM sampling and speculative verification, improving efficiency for dynamic serving workloads.
Research explores using biokinetic knowledge priors for deep learning in biomanufacturing to address data scarcity with ODE models.
Codec-Gauge introduces a post-training cache-coordinate layer to improve KV-cache compression and quantization in long-context Transformer models.
New research proposes GKR-HND protocols to verify outsourced Transformer inference, addressing model substitution and incomplete execution risks.
Research finds combining regex filters with LLM alignment provides no measurable safety coverage gain against adversarial input, even when designed to bypass regex.
Research introduces a Graph Neural Network (GNN) framework for 'Zero-Shot Digital Twins' that adapts to changing physical domains without retraining.
Research introduces CT-Merging, a method to combine multiple LoRA adapters into one multi-task adapter, reducing storage and inference complexity.
Research introduces 'inspect_permute' to quantify LLM position bias in multiple-choice benchmarks via exhaustive answer-order permutations.
CudaPerf, a new reflective RL framework, enhances LLM-based CUDA kernel generation by incorporating structural and performance-aware rewards beyond mere outcome signals.
Research finds LLMs struggle with dynamic user intent in iterative tasks, as current evaluation focuses on single-turn, fully-specified settings.
Research explores how architectural conditions in active inference agents lead to 'causal emergence,' linking predictive organization to information theory.
Researchers introduced TOUR, a benchmark to evaluate trajectory-level unlearning in Offline Reinforcement Learning (RL) agents, addressing deletion challenges.
Research shows dense prediction rewards cause LLM agents trained with GRPO to fail, driving them to degenerate states rather than task success.
Research quantifies the memorization capacity of LoRA adapters, finding it significantly smaller and less predictable than full fine-tuning.
Research proposes a method for uncertainty-aware trust estimation in multi-LLM systems using structured expert judgment to improve reliability.
Research explores 'looped Transformers' with adaptive depth, analyzing how learned halting gates influence inference computation and trajectory formation.
Research proposes "Best-of-Evidence" for model output selection under partial verification, improving on Best-of-N for vision-language tasks.
LeanFlow is an LLM agent system designed to translate mathematical papers into formal, buildable Lean projects, evaluating runtime mechanisms.
Researchers introduced HypNO, a graph-based neural operator using physics-informed message passing for hyperbolic conservation laws, benchmarked on traffic-flow models.
Research proposes an unsupervised, context-augmented language model using perplexity for Advanced Persistent Threat (APT) detection in security logs.
Research proposes a spectral transformation method to improve LoRA aggregation in federated fine-tuning of Vision Transformers, addressing cross-term errors.
Research proposes Generative Bayesian Filtering, a new method for state estimation in dynamic systems with complex observation models.
Research explores multi-task learning (MTL) for heterogeneous prediction from video game state data, aiming to improve generalization and reduce costs.
Gumbel Distillation is a new technique for parallel text generation, improving non-autoregressive model quality to rival sequential models.
AG-REPA introduces a causal layer selection method for representation alignment, improving generative flow models, particularly for token-conditioned audio.
Research introduces 'swarm-attack,' an open-source adversarial framework using coordinating LLM agents to bypass safety and discover software vulnerabilities.
Researchers introduced TOPReward, a training-free method using LLM token probabilities as zero-shot rewards for general-purpose robot learning, reducing manual effort.
Research introduces VPWEM, a visuomotor policy using working and episodic memory to handle non-Markovian robotic tasks, addressing long-term memory limitations.
Researchers developed M$^3$-Gen, a framework to generate interpretable gene expression profiles from clinical and imaging data, addressing data cost and privacy.
Research finds Reinforcement Learning with Verifiable Rewards (RLVR) can improve one-sample accuracy while decreasing distinct problem-solving ability.
New research proposes APEX, a polynomial framework for exact Aumann-Shapley attribution in Graph Neural Networks (GNNs), addressing GNN explainability challenges.
Research proposes sparse autoencoders (SAEs) to create interpretable embeddings for analyzing large text corpora, addressing limitations of LLM-based and dense embedding methods.
Research presents enhanced Membership Inference Attacks (MIAs) on diffusion models using a frequency-domain approach, improving data privacy breach detection.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion