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 GeoDetect, a new method for detecting adversarial attacks on Vision-Language Pre-trained models (VLPs) by analyzing their embedding space geometry.
New research proposes a novel 'diffusion distance' measure to quantify discrepancies between probability distributions on a graph, using Metropolis-Hastings.
Research addresses domain adaptation challenges for plug-and-play image reconstruction denoisers used outside their training domains, impacting convergence.
Research explores optimal self-distillation for rectified flow generative models, proving a student can improve a suboptimal teacher.
Researchers introduced cGAP, a new visualization method using HOMALS-Guided Heatmaps for high-dimensional categorical data analysis, enhancing interpretability.
Research details parameter-efficient prompt tuning of vision foundation models using adaptive focal loss for interpretable MCI screening from drawing tests.
DriftWorld introduces a new generative model for faster world modeling, enabling quicker rollouts for action planning in robotic control.
AlphaWiSE, a new weight interpolation method, addresses catastrophic forgetting in multimodal models like CLIP by adapting to new data without losing prior alignments.
NeuronSoup introduces an asynchronous, shared-neuron temporal graph architecture for neural computation, replacing synchronous layer processing.
Research introduces MeanFlowNFT, a forward-process reinforcement learning framework for efficient average-velocity generative models, enhancing alignment.
Research explores human-in-the-loop ML for safe autonomous vehicles, addressing challenges in complex scenarios and data labeling bottlenecks.
New SNS-MDP framework models environments with switching dynamics governed by a latent Markov chain, showing equivalence to stationary dynamics.
Researchers propose a novel primal-dual algorithm for contextual stochastic combinatorial optimization using neural networks with combinatorial layers.
Research proposes Generalized Fisher-Weighted SVD to efficiently compress large language models by approximating full Fisher information.
Researchers introduced SOReL, a fully offline Bayesian model-based Reinforcement Learning method for hyperparameter tuning and performance estimation.
SafeOR-Gym introduces a new benchmark suite for safe reinforcement learning algorithms focused on operations research problems, unlike prior robotics benchmarks.
Research investigates the impact of stochasticity in score-based diffusion models on generation quality using KL divergence analysis.
Researchers introduced Similarity as Reward Alignment (SARA), a contrastive framework for preference-based reinforcement learning robust to labeler errors.
Research introduces a Dynamic, Automatic, and Systematic (DAS) red-teaming framework to continuously stress-test LLMs for health, addressing static benchmark obsolescence.
Research questions the reliability of temporal graph learning benchmarks, finding simple heuristics often competitive with complex models.
Research proposes an adaptive layer-freezing strategy for energy-efficient federated learning in medical image conversion to reduce computational load.
CluCERT proposes a clustering-guided denoising smoothing method to certify Large Language Model robustness against adversarial text attacks like synonym substitutions.
Research details a statistical and economic framework for determining when to switch from an incumbent production model to a challenger model leveraging new data sources.
Research explores Federated Continual Learning (FCL) for streaming data without task IDs or consistent class distribution, aiming for better knowledge retention.
Research proves 'grokking'—generalization long after overfitting—in ridge regression using gradient descent with weight decay.
Research empirically compares popular hyperparameter optimization methods for tree-boosting, a technique widely used for tabular data.
Research demonstrates stable pretraining of Large Language Models using exclusively low-rank weights, matching dense model performance while reducing compute.
Research introduces flow-based generative models with native extrapolation awareness to prevent silent failures in safety-critical prediction tasks.
ARTIST, a new method for adaptive time series reasoning, selects relevant segments rather than encoding entire sequences, improving efficiency.
InfoFlow KV proposes an information-flow-aware KV recomputation method to optimize inference-time prefilling for RAG in long-context LLMs.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion