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 introduces a causal dictionary learning framework to extract and validate interpretable features within genomic language models, addressing model opacity.
Research identifies variable fault sensitivity across bit positions in DNN inference memory for transformer and CNN models, suggesting unequal error protection.
Research proposes ECRAM, a continual learning method for edge devices, combining previous data summaries with new sensory input to avoid catastrophic forgetting.
FedLSG is a research framework integrating LLMs into federated graph neural networks to defend against backdoor poisoning attacks by improving semantic understanding.
Researchers introduced McQuack, a trainable quantum kernel method for multiclass classification that achieves linear scaling with training set size.
Research identifies "correlated agreement blindness" in multi-agent systems where improving base models converge, creating a safety monitoring blind spot.
Researchers discovered HijackKV, a new security threat in LLM inference that exploits position-independent KV cache reuse for data leakage.
Research paper proposes current LLMs suffer from a 'structural monoculture' and future AI systems will be 'systems of systems' for specialized tasks.
Researchers propose using physical noise as a native regularizer in photonic hybrid quantum neural networks, testing on Iris, Digits, and MNIST datasets.
Researchers introduced Directional Kernel Mean Difference (DKMD), a signed statistic for univariate distribution comparison, preserving directional shifts.
HeadCast proposes a training-free method to reduce inference costs in autoregressive video generation models by optimizing attention head usage.
Research proposes a multi-stage dynamic selection framework for cross-project defect prediction, addressing distribution shifts.
Research proposes machine learning for plausibility-driven prioritization of candidate biomedical annotations to reduce expert review bottlenecks.
Research explores statistical inference for optimal rank allocation in Low-Rank Adaptation (LoRA) for fine-tuning large language models.
Research proposes aggregating Bayesian update rules as experts to achieve uncertainty-aware prediction on data streams with adaptive inferential choices.
Research explores deep neural network estimators for nonparametric quantile and Huber regression under covariate shift and dependent data.
Research proposes training Attention Graph Neural Networks without labeled data by minimizing finite-volume method residuals for thermo-fluid field prediction.
Researchers propose Differentially Private Decoupled Training (DP-DT) to improve utility in differentially private neural network training by decoupling representation learning from privacy enforcement.
Research identifies a fundamental 'theory-to-practice gap' in learning with ReLU neural networks and neural operators, related to sampling complexity.
Research introduces a new method for membership inference attacks (MIAs) that identifies if data, even from unseen categories, was used in model training.
NeuCoReClass AD proposes a self-supervised time series anomaly detection method using multiple proxy tasks to improve pattern capture.
Research explores dynamic architectural adaptation via Optimal Brain Damage Masking to enhance deep neural networks' robustness to noisy labels.
Research applies Self-Explaining Neural Networks (SENNs) to a PPO reinforcement learning agent for mobile network resource allocation, generating intrinsic local and aggregated global explanations.
Research explores information separability in pixel-space diffusion models, finding most information is for small-scale perceptual details.
Research on 'Always-On Inter-Cluster Repulsion in Replay-Based Continual Learning' found that geometric separation in feature space may not improve retention.
Research identifies a "Dominant-vs-Dominated" (DvD) imbalance in text-to-image diffusion models where one concept dominates multi-concept generation.
ArenaRL proposes a tournament-based relative ranking system for training open-ended LLM agents, moving beyond scalar reward models for complex tasks.
Research proposes a generative AI-enhanced probabilistic multi-fidelity surrogate modeling framework using transfer learning to address data scarcity.
Research proposes an "In-Run" data attribution method for the Adam optimizer, addressing limitations of prior methods restricted to SGD.
Research explores a geometric perspective on intervention learning, characterizing intervention avoidance in autonomous systems by constraining policies.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion