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.
Research evaluates an interpretable ML framework for predicting respiratory disease and air quality, analyzing socioeconomic subgroups.
Apeliotes, a new diffusion-based modeling framework, offers machine learning-based downscaling for high-resolution atmospheric data.
Research introduces a diagnostic for evaluating LLM constraint reasoning, distinguishing between problem density and inherent solver hardness for SAT problems.
Research explores Bayes-filtered transformers (BFTs), showing their next-token prediction can approximate Bayesian posterior predictive distributions.
Research details fine-tuning Whisper models for robust Automatic Speech Recognition (ASR) in low-resource languages like Assamese.
Research demonstrates improving vanilla Transformer LLM arithmetic performance by tuning models with human-like strategies.
DADIR is a new framework for imbalanced regression, addressing challenges in continuous target variables and heterogeneous density distributions.
Research explores 'Distilled Reinforcement Learning' for LLM post-training, aiming to improve reasoning and alignment beyond current RL and OPD methods.
Node4All introduces a new node representation learning method designed for generalization across arbitrary graph datasets without dataset-specific optimization.
AIGB-R1 is a research paper proposing a self-evolving generative auto-bidding system using hierarchical planner-executor optimization with LLMs to improve online advertising strategy.
Research investigates lookahead branching strategies to improve neural network verification, integrating them into branch-and-bound verifiers.
Research paper axiomatizes the softmax policy, a core reinforcement learning component, resolving a tension with the Independence axiom.
Research explores the practical efficiency and robustness of Reinforcement Learning algorithms under real-world transfer constraints and dynamics mismatch.
Research highlights that ML drift detectors, though accurate on synthetic shifts, often generate high false alarm rates in continuous production monitoring.
Researchers propose a multiverse-consensus pipeline for reproducible feature selection in untargeted LC-MS metabolomics, addressing pipeline decision variability.
ChebyMA, a new parameter-efficient adaptation method using Chebyshev polynomial bases for weight matrices, is proposed in research.
Researchers propose Kernelized Linear Attention Activations (KATA) to improve linear attention's associative recall, formulating it as a spherical-packing problem.
Research introduces persistent homology and split-conformal methods for calibrated Alzheimer's conversion risk prediction from MCI with uncertainty estimates.
Research introduces RGMR, an inference-time framework adapting pre-trained foundation models for multi-scale temporal analysis and iterative refinement.
Researchers introduced Retrieval-Augmented Interpretable Learning (RAIL), a meta-learning framework for zero-shot generation of interpretable models.
Research explores lightweight wrappers for adapting Time Series Foundation Models (TSFMs) to regional forecasting, addressing proprietary weights and limited local data.
Research explores lowest-variance selection in Gaussian random-field models, motivated by confidence-guided parallel unmasking in masked diffusion.
Research introduces FailureAtlas, a taxonomy classifying failure modes in multi-provider LLM serving infrastructure based on origin layer (Network/Transport).
Research combines LLMs with neural simulation-based inference for joint model selection and parameter estimation from natural language.
A research paper proposes a volatility-aware machine learning approach to detect extreme price movements in high-frequency financial markets.
Research introduces CoCurve, a training-free structured pruning method for LLMs that considers cross-module dependencies, addressing limitations of independent unit pruning.
Research characterizes the expressive power of global-attention graph transformers for Mixed-Integer Linear Programs (MILPs) using graph isomorphism.
Research introduces Jacobian-Aggregated Group Gradient (AGG) to reduce computational cost for Group Relative Policy Optimization (GRPO) in diffusion models.
ANNLib is a new programming framework for efficient and flexible Approximate Nearest Neighbor Search (ANNS) in deep learning pipelines.
Research proposes a multi-engine fusion method for protein-ligand binding affinity prediction, improving accuracy with interpretable confidence.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion