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.
New research proposes Sequential Adaptive Rollout Allocation (SARA) to improve compute efficiency in Reinforcement Learning with Verifiable Rewards (RLVR).
Research paper introduces Top-k Pareto Bandits, a multi-objective bandit algorithm for selecting a slate of k actions that approximate a Pareto frontier.
FloDR, a new invertible dimensionality reduction method based on a normalising flow, addresses limitations of t-SNE and UMAP in preserving data structure.
Researchers built and evaluated a self-serve entity resolution system across six datasets, identifying three key lessons for practical deployment.
New research proposes Implicit Causal World Models to learn environmental dynamics and agent intents from multi-agent demonstrations, addressing distribution shifts.
RAGuard proposes a layered defense framework using adversarial fine-tuning for dense retrievers to protect RAG systems from corpus data poisoning attacks.
Research identifies a structural issue where permutation-equivariant GNN explainers produce non-canonical top-k explanations due to arbitrary ordering.
Research introduces a method for high-order Markov blanket discovery, relaxing the faithfulness assumption common in causal and graphical model learning.
New research proposes a game-theoretic approach to fine-tuning large language models, balancing performance improvement with drift control via KL regularization.
New research proposes a nongradient vector flow method for learning flow maps in diffusion and consistency models, aiming to reduce inference overhead.
New research proposes an Existence-Field Diffusion Model to generate spatial point processes with variable cardinality, overcoming limitations of existing methods.
New research proposes Dense Hierarchical Rewards and Curriculum Learning (DHRCL) to improve code LLM training using execution, unit test, and structural analysis.
Research introduces conformal changepoint localization for robust root cause analysis in engineered systems, providing statistical reliability guarantees.
Researchers propose HiFloat4 for the first end-to-end 4-bit precision Reinforcement Learning (RL) post-training of large language models.
AgentGFM introduces a Graph Foundation Model with node-agent information-flow control, addressing variability in local graph structural patterns.
Research presents an analytical method to estimate energy consumption of LLM inference on GPUs, aiming to simplify sustainability reporting and design.
New research proposes a method for online conformal prediction that simultaneously guarantees coverage and efficiency, addressing limitations of existing adaptive conformal inference techniques.
Research proposes an uncertainty-guided LLM semantic augmentation method to improve heterogeneous treatment effect estimation in personalized interventions.
Research investigates how statistical similarity in train-test splitting impacts AutoML model evaluation, especially with imbalanced datasets.
Research investigates context sampling choices for TabPFN on small tabular datasets, impacting prediction stability, accuracy, and selection cost.
Researchers propose AIGen, a framework automating AI Bill of Materials generation using SPDX 3.0 standards within hybrid MLOps pipelines.
Researchers propose a framework to forecast safety risks in multi-turn AI agent interactions where malicious intent is spread across turns.
Researchers propose a cost-calibrated search framework for LLMs that balances compute-token expenditures against marginal quality gains.
Researchers analyzed how open-weight models with inference-time thinking handle uncertainty and information-seeking in bandit trials.
Researchers propose ReCo to fix a flaw in GRPO reinforcement learning where models over-concentrate on high-probability reasoning paths.
Researchers introduced OPAB, a testing framework to detect outcome performativity where predictive models causally alter their target outcomes.
A research paper demonstrates that simple two-call self-refinement outperforms complex five-role multi-agent architectures on local LLMs.
Researchers benchmarked vision foundation models using linear probing for face presentation attack detection to address cross-dataset domain shifts.
Researchers introduce BayesAME, a Bayesian method to estimate LLM benchmark performance using dynamic, optimized subsets of evaluation data.
Academic paper proposes a cost-aware stopping framework for LLM agents to optimize the number of external tools and database queries used.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion