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.
STRATA is a new research model for inferring race and ethnicity from names and geography, aiming to improve fair lending compliance under ECOA and HMDA.
Research proposes enhancing LLM clinical reasoning using real-world data from a sepsis registry, addressing privacy concerns in training.
A systematic survey of data balancing strategies, including resampling and augmentation methods, addresses challenges of imbalanced datasets in ML.
Research introduces Chameleon, a multiplier-free temporal convolutional network accelerator for low-power, on-device few-shot and continual learning.
Research identifies widespread misuse of t-SNE and UMAP in visual analytics, particularly for interpreting inter-cluster relationships due to projection distortions.
SATQuest is a new verifier tool for evaluating and fine-tuning LLMs on logical reasoning tasks using SAT-based problems and objective checking.
Research proposes a one-shot, online-learned preconditioner to accelerate transient PDE solvers without extensive classical simulation data for training.
A research paper proposes a Representation-Aware Distributionally Robust Optimization framework to improve model robustness against data shifts.
OmniX introduces a panorama-based 2D lifting technique for automatic 3D scene construction, generating graphics-ready scenes for PBR, relighting, and simulation.
A research paper proposes an 'AI epidemiology' framework to standardize expert-AI interaction data for prospective risk detection in deployed AI systems.
Research proposes Singular Bayesian Neural Networks to provide calibrated uncertainty with lower parameter cost than standard BNNs, optimizing for weight matrix sparsity.
Hippasus, a new research approach, proposes an automated feature augmentation method for ML tasks on relational data across complex multi-table schemas.
Research introduces a method for decomposing uncertainty in Bayes-filtered transformers, separating aleatoric from epistemic uncertainty for better risk assessment.
Researchers developed a Bayesian framework, "diffusion-within-Gibbs sampling," for improved signal component decomposition in noisy data.
Research explores Thompson Sampling performance under model misspecification in dynamic decision-making, common in economic environments.
A research paper proposes a 'Brouwerian assertibility constraint' for responsible AI, requiring verifiable certificates for high-stakes assertions.
Research proposes Segment-Wise CoT Compression with Answer Alignment (SCA) to reduce LLM inference costs without compromising answer quality.
Research introduces a perturbation-based method for LLM extrapolation, reinterpreting models through pre-post-additive noise for next-token prediction.
FAME, a new Failure-Aware Mixture-of-Experts model, improves log anomaly detection by identifying specific anomalous messages, not just log groups.
Research proposes diffusion-guided estimators for counterfactual distribution learning, improving stability for high-dimensional outcomes concentrated on lower-dimensional structures.
SEC-bench Pro is a new benchmark evaluating frontier LLMs' ability to find real software vulnerabilities and generate proof-of-concept exploits.
Research paper benchmarks classical and quantum machine learning models for image recognition on MNIST dataset, exploring quantum as a new paradigm.
Research explores Bayesian reinforcement learning to manage epistemic uncertainty in offline policy optimization through posterior hybrid belief.
Research introduces a three-level audit protocol to test for systematic asset-specific preferences and biases in financial large language models.
CyberGym-E2E is a new large-scale, real-world benchmark for evaluating AI agents' end-to-end cybersecurity vulnerability discovery and remediation.
Research introduces Spectral Adaptive Conformal Prediction for robust prediction intervals in structured non-exchangeable time-indexed data.
Research introduces a model-free temporal-switch framework for neuromorphic computing to mitigate device variations and improve transferability.
Research proposes a method to synthesize parameters for nonlinear systems using gradient-based optimization and Signal Temporal Logic (STL) to satisfy robust constraints for uncertain initial conditions.
Research explores mixed-timescale differential coding for federated learning model broadcast to reduce communication load in wireless environments.
Research proposes multi-level context modeling to improve expert selection consistency in Mixture-of-Experts (MoE) models, addressing unstable routing.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion