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 finds relative positional encodings enable transformer models to generalize better to longer sequences than absolute encodings.
Research explores auto-tuning of quantum-inspired annealers for SVMs, converting continuous variables into QUBO models with Optuna.
Researchers introduced PertReason, a new benchmark and framework for evaluating mechanistic reasoning in ML models within scientific domains.
QScheduler proposes adaptive gradient sampling for Zeroth-Order (ZO) on-device training on INT8 NPUs, reducing memory needs.
New research introduces Posterior Prefix Tuning (PPT), a method to steer transformer behavior by optimizing the latent posterior without backpropagation.
Research applies machine learning super-resolution to downscale atmospheric data for better detection and forecasting of extreme weather events.
Research paper proposes Utility-Augmented Transformer (UAT) to address 'feedback-blind' retrieval in sequential decision-making models, enhancing adaptation.
New research argues that 'circuit extraction' methods for identifying AI model mechanisms are under-determined, with findings depending on reporting and comparison choices.
Research explores decoding visual perception from brain activity using ECoG and deep learning to predict visual categories from videos.
Research explores the functional equivalence and geometric diversity of neural network approximations, addressing identifiability concerns beyond the Universal Approximation Theorem.
Researchers propose SFGA, a statistics-first gating architecture for evaluating the quality of supervised fine-tuning (SFT) data based on diversity, utility, and redundancy.
Research proposes variational meta-learning for neural system identification to improve data efficiency and uncertainty quantification in low-data regimes.
Research proposes unsupervised multi-kernel learning for automated algorithm selection in black-box optimization, reducing reliance on costly supervised models.
Research explores using reinforcement learning with verifiable rewards for molecular generation, improving LLM-based chemical and drug design.
Research incorporates physics principles into machine learning for electric truck energy prediction, improving reliability over standard methods.
Research explores Parallel Noising in Neural Markov Logic Networks (NMLNs) to improve generative modeling performance on larger relational structures.
Research presents a generative state-space model learning directly from sparse, noisy observations to improve ocean modeling beyond complete datasets.
Research explores specialized multi-generator adversarial learning to detect rare, non-homogeneous failures in predictive maintenance using imbalanced datasets.
Neural Kolmogorov Equations (NKEs) are proposed for learning stochastic dynamics with parallelizable training and support for general noise types.
New research explores Conservative Query and Adaptive Regularization for offline reinforcement learning to improve policy performance from static datasets using expert feedback.
New research introduces S3, a Hierarchical Reinforcement Learning approach using uncertainty-constrained subgoal selection to improve long-horizon task solving.
Research combines graph features, TreeSHAP explanations, and an LLM agent for auditable fraud detection on the PaySim dataset.
Research explores network-agnostic feature initialization for Graph Neural Networks to improve spatial transferability in traffic assignment models.
Research explores shallow recurrent decoder networks for real-time optimal control in computationally demanding dynamical systems.
New research proposes an optimization stack for Reinforcement Learning with Verifiable Rewards (RLVR), focusing on reusing base model weight spectra.
Research introduces \pddim, a diffusion-based posterior sampling algorithm for linear inverse problems, claiming improved efficiency and theoretical guarantees.
Research explores using schema-derived constrained decoding for MLIR code generation, improving generalization across dialects without retraining.
Research explores 'solver-in-the-loop' interactive optimization modeling using natural language, moving beyond one-shot LLM approaches.
Research finds aggregating judgments from multiple LLMs can outperform individual models but risks 'contamination' from poor performers.
Research introduces Polynomial-Sensitivity-Aware Pruning (PSAP) to improve reliability of pruned homomorphically encrypted neural networks for efficient inference.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion