Search signals, briefings, company results, benchmarks and glossary terms.
Search signals, briefings, company results, 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.
Research proposes online conformal inference with retrospective adjustment, enhancing prediction set coverage guarantees under evolving data distributions.
New research proposes Likelihood Matching to train diffusion models by establishing equivalence between target data likelihood and reverse diffusion path likelihood.
Research proposes funOCLUST, an extension of the OCLUST algorithm, for robust clustering of functional data with outlier detection capabilities.
Researchers propose PnPBO, a plug-and-play framework for stochastic bilevel optimization, integrating unbiased and biased estimators into single-loop methods.
New research proposes "50%-task-completion time horizon" to benchmark AI's ability to complete long software development tasks against human performance.
Research on learned optimizers (LOs) aims to improve traditional optimizers by meta-learning neural networks, focusing on efficient long-horizon training.
New research suggests reinforcement learning (RL) may be more prone to catastrophic forgetting in continual learning than previously believed.
Researchers propose FAST, a new framework to improve sampling efficiency and reduce re-initialization latency in parallel reinforcement learning for autonomous driving.
Research studies gradient descent dynamics in feed-forward ReLU networks, deriving explicit recursions for multi-hidden-layer cases.
IntraShuffler is a privacy-preserving framework for Heterogeneous Differential Privacy in Federated Learning, improving model utility.
Research explores improving Neural Processes for modeling unknown functions from sparse, irregular data, addressing limitations in translation-equivariant NPs.
Research proposes "Value-Filtered Decoding" to enhance LLM safety by selectively applying safety steering, intervening only when necessary.
Research explores Causal Foundation Models for continuous treatments, addressing a gap in causal inference for variables with continuous ranges.
A new momentum schedule for SGD, derived from critical damping, achieves 90% test accuracy 2.34x faster on ResNet-18/CIFAR-10, with a minor final accuracy deficit.
Turbo Connection (TurboConn) is a novel Transformer architecture designed to overcome fixed-depth computation limits in complex reasoning tasks.
Research details the optimal trade-off between predictive utility and fairness (separation) using an information-theoretic approach.
Research proposes TIC-FM, an in-context learning framework for zero-shot time series classification, moving beyond task-specific classifiers.
Researchers introduced Spectral Filtering Operator (SFO), a neural operator using a universal spectral basis for efficient PDE solution, addressing long-range interactions.
Research paper proposes an algorithm with optimal regret for dynamic service fee pricing on third-party platforms with demand learning under confounding.
Research explores enhancing adversarial transferability using 'block stretch and shrink' transformations, explaining efficacy from an implicit ensemble perspective.
Researchers propose a 'Graph Optimization Foundation Model' extending the LLM pretrain-transfer paradigm to combinatorial optimization problems on graphs.
Researchers propose a data-driven interpolation method for functions on smooth manifolds, using Gaussian Nadaraya–Watson and Voronoi tessellations.
Researchers propose Riemannian Denoising Diffusion Probabilistic Models (RDDPMs) for generative modeling on complex data manifolds, reducing reliance on explicit geometric information.
Research explores using stochastic sharpness-aware minimization (SAM) with isotropic normal noise to analyze diagonal linear network training dynamics.
Research paper shows that training AI models on additional, even irrelevant, states can improve generalization to new contexts in contextual MDPs.
Research explores federated topic modeling with VAEs for privacy-preserving, multi-party document analysis, addressing communication/performance.
Research identifies a novel 'input-aware dynamic backdoor attack' vulnerability in Quantum Neural Networks, circumventing fixed-trigger defenses.
Research proposes a climate-invariant conformal prediction method for multi-horizon solar and wind forecasting, addressing challenges in uncertainty quantification.
Research paper re-mathematizes backpropagation as a single nilpotent linear system, offering a global operator theory instead of layer-wise recursion.
Researchers propose LaGuadia, a language-guided adaptive distillation framework to create compact pathology image encoders from large foundation models.
Research proposes a hierarchical architecture for LLM agent orchestration, using stack-based execution and lazy discovery to manage tool complexity.
Research ablates coding agents' tool use, comparing bash, code execution, and no restriction on synthetic and SWE-bench tasks to assess efficacy.
RecRec proposes recursive refinement for sequential recommendation systems, modeling user preferences as a persistent latent state for iterative updates.
Research introduces TSCoNet, a two-stage Copula CNN-LSTM model for uncertainty-aware spatio-temporal forecasting of correlated environmental variables.
Research addresses runtime safety for learned sUAS separation policies under degraded GNSS, highlighting fundamental architectural questions for deployment.
A research paper proposes a multi-agent, knowledge-based decision support system for recommending security controls in IT environments.
Research finds quantum reservoir computing performance parity with carefully tuned classical methods, challenging prior claims of quantum advantage.
GATAS introduces a black-box testing method for ASR systems, generating phoneme-level adversarial inputs to induce failures while maintaining naturalness.
Research explores Task-Conditioned Synthetic Data Generation (SDG) to enhance machine learning performance, addressing data scarcity in agricultural tasks.
Researchers propose ECG-LDC, a low-dimensional computing framework for energy-efficient ECG arrhythmia classification on resource-constrained hardware.
Research explores 'Requential Coding,' a method using self-generated training data for model compression, suggesting efficiency gains.
New research proposes Active Offline-to-Online Reinforcement Learning (O2O-RL) for training and improving policies using historical data and limited online interaction.
DiffEEG, a 9.6M-parameter self-supervised diffusion model, is proposed for EEG seizure detection, addressing data scarcity and class imbalance.
Research introduces a new method for anytime-valid and feature-aware auditing of black-box conditional quantile forecasters, addressing data drift challenges.
Research finds common post-hoc calibration methods for probabilistic classifiers, Temperature Scaling and Isotonic Regression, lack robustness across varied operating conditions.
Researchers propose CDFM, a 'Causal Discovery Foundation Model' approach for recovering causal structures from diverse observational data.
Researchers propose PRISM Edit, a new LLM editing method allowing temporal facts to be updated while retaining historical accuracy without full retraining.
Research introduces PolicySynth, a framework for generating synthetic data that ensures decision alignment for campaign decision support, beyond just distributional similarity.
Research proposes ToolAtlas, a system for LLM agents where tool knowledge and memory are maintained by the tool provider, enabling reuse across agents.
Research explores a new training priority weighting framework for physics-informed neural networks (PINNs) to improve stability and accuracy.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion