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 explores explainable, lightweight, and compact deep models for Speech Emotion Recognition, focusing on transparent predictions and efficient deployment.
Research explores transferable low-rank convolutional bases to onboard unseen medical imaging modalities without retraining the core model.
Deep Adaptive Bayesian Screening (DABS) introduces an offline policy network for adaptive factorial screening in high-dimensional discrete design spaces.
Research proposes Counterfactual Vision Action Analysis (CVAA) to interpret complex road scenarios for autonomous vehicle vision-language-action models.
New research characterizes sample complexity for estimating min-entropy and Rényi entropy, building on existing Shannon entropy estimation.
Research explores Twisted Schrödinger Bridge Matching for approximating optimal transport dynamics with applications in generative modeling.
PriorProof is a research tool that quantifies the 'nonstandardness' of formal proofs in Lean by analyzing the weighted surprisal of their dependency footprint.
Research explores multi-model hybrid defense to protect Network Intrusion Detection Systems (NIDS) from white-box adversarial attacks.
Research on agentic computer-use RL shows performance is dominated by upstream variance, not evaluation methods, across multiple oracle-graded environments.
New research introduces Rate-Distortion-Perception theory, extending classical rate-distortion to incorporate perceptual quality and semantic validity.
Research empirically analyzes speculative decoding, a technique to accelerate LLM inference by using a small draft model and batched verification.
Research explores interpreting quantum machine learning models, which use coherent quantum evolution, via stochastic processes to understand their decision-making.
Research explores self-modifying agents for formal mathematical reasoning, focusing on evolving proof workflows in Lean using compiler feedback and self-repair.
Taurus introduces a single-machine system to accelerate out-of-core Graph Neural Network inference on billion-scale graphs, reducing memory and I/O costs.
Research presents an experimentally verified formal law for calculating uplift from diversity of thought in LLM ensembles, decomposing lift into rescue and damage masses.
KReTTaH is a new training-data-free, interpretable, nonparametric kernel regression framework for multi-way data imputation using tensor trains.
Research proposes a method for efficiently evaluating new LLMs on question sets using historical data, constructing confidence sequences to track capability.
Researchers propose Feature-Informed Diffusion Evolution (FIDE), a black-box framework for inverse rendering, bypassing differentiable renderers and gradient descent.
DA-MergeLoRA proposes a hypernetwork-based LoRA merging method for few-shot test-time domain adaptation in machine learning.
Researchers introduced FluxBench to systematically evaluate AI agents on full electronic design automation (EDA) workflows from RTL to GDS.
Research identifies and removes an artifact in Nash equilibrium selection by regularized solvers in zero-sum games, aligning empirical results with maximum-entropy selection.
COLIP-2 integrates olfaction, vision, and language into a unified multimodal embedding space for probabilistic aroma localization.
ZifaMem proposes a structured memory system for AI companions, organizing dialogue into session summaries, episodic memories, and a consolidated user model.
Research proposes an end-to-end pipeline using LightGBM and TreeSHAP for explainable money mule detection from transaction and behavioral data.
Research introduces a method for online learning of neural state-space models, improving upon offline identification techniques by enabling real-time adaptation.
Research proposes LFM, a new framework leveraging foundation models for source-free universal domain adaptation, addressing covariate and label shifts.
Research introduces an adaptive algorithm for efficient dimension selection in multidimensional probit graded response models used in assessments.
Research uses specialized Vision Transformers with Sentinel-2 satellite data for early sugar beet yield prediction, integrating domain knowledge.
A research commentary discusses the systematic comparison of test fairness and algorithmic fairness, mapping the entire testing workflow onto AI/ML paradigms.
Researchers developed an adjoint-sensitivity framework to analyze positional influence and 'lost-in-the-middle' phenomena in causal residual transformers.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion