Search signals, briefings, company results, benchmarks and glossary terms.
Search signals, briefings, company results, benchmarks and glossary terms.
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.
DIB-OD proposes a novel Graph Neural Network pretraining method to improve generalization across heterogeneous graph domains by disentangling invariant knowledge.
New research introduces Rooted absorbed prefix Trajectory Balance (RapTB) to improve Generative Flow Networks (GFlowNets) by reducing mode collapse during LLM fine-tuning.
Research shows current benchmarks for malicious prompt classifiers overestimate generalization due to distribution shift, proposing a LODO evaluation method.
Research explores composing Generative Flow Networks (GFlowNets) for multi-objective optimization, aiming to sample diverse candidates based on multiple reward functions.
ProtoTSNet introduces an interpretable approach for multivariate time series classification using prototypical parts, designed for critical domains.
Research explores how Large Language Models can be more effective at automated algorithm design by incorporating strong priors, moving beyond adaptive prompt designs.
Research proposes mHC-GNN, a Graph Neural Network architecture adapting Manifold-Constrained Hyper-Connections to mitigate over-smoothing and enhance expressiveness.
New neural approach for global optimization from noisy samples shows iterative refinement can find global minima, addressing multi-modal function challenges.
Research introduces 'SkillRouter' for LLM agent skill routing, addressing scalability challenges in large skill ecosystems.
IoUCert introduces a new formal verification framework for robustness in anchor-based object detectors, overcoming challenges with IoU metrics.
Research paper L2GTX proposes a new method for generating global, class-level explanations for deep learning time series models, addressing current XAI limitations.
Research addresses the "factorization barrier" in diffusion language models, aiming to improve parallel token generation efficiency and coherence.
Research finds that larger language models exhibit more robust in-context learning under adversarial distribution shifts, offering worst-case guarantees.
Research proposes adapting tabular foundation models for survival analysis by reframing time-to-event outcomes as binary classification problems.
Research proposes 'Epiplexity per Joule' and 'Empowerment per Joule' metrics to connect AI's physical efficiency with intelligence.
Researchers propose SILVER with RL-guided labeling, an extension to the SILVER framework for interpreting deep reinforcement learning policies in complex environments.
A research paper details the development and real-world deployment of an integrated financial prediction system for accounts receivable and cash flow forecasting in SMEs.
Research proposes ProDER, a continual learning framework for fault prediction in smart grids to adapt to evolving operational environments.
NIRVANA, a new structured pruning framework, aims to accelerate LLM inference while maintaining zero-shot performance and reducing retraining needs.
A research survey on "unlearnable data" (ULD) explores methods to prevent ML models from learning patterns for privacy and security through data perturbations.
Research identifies specific reasons why physics-informed machine learning (PIML) fails in traffic flow modeling, underperforming data-driven and physics-based baselines.
Research benchmarks eight attention mechanisms' energy usage and hardware resource demands, addressing high memory and time complexity in LLMs and VLMs.
Research introduces a geometry-aware deep learning framework for enhanced adversarial robustness by sculpting internal representations for better class separation.
Research proposes Variational Autoencoders (VAEs) using hyperspherical coordinates to improve anomaly detection by mitigating hypervolume growth in high-dimensional latent spaces.
New research proposes statistical evaluation of LLM mathematical reasoning using pairwise comparison signals, addressing limitations of current benchmarks.
Researchers propose UnMaskFork (UMF), a framework using deterministic action branching to improve test-time scaling for Masked Diffusion Language Models (MDLMs).
Research introduces gate-based quantum reservoir computing for multivariate time series forecasting, optimized for current NISQ hardware constraints.
Research introduces Sparsity-Aware Low-Rank Representation (SALoR) to reduce fine-tuning and deployment costs for large language models.
Research indicates LLM layer pruning, while boosting efficiency, significantly degrades long-chain reasoning, a critical capability for complex tasks.
Research systematically investigates Reinforcement Learning (RL) based jailbreaking techniques in LLMs, highlighting threats to safe model deployment.
RoVE (Rotary Value Embeddings) is a parameter-free modification to make attention values position-sensitive, building on Rotary Position Embeddings (RoPE).
Research surveys methods for representing black-box single-objective continuous optimization problems using features for ML tasks like algorithm selection.
New research proposes "singularity-aware optimization" to address non-smooth loss landscapes in deep learning, aiming for stable convergence.
Researchers propose Overlapping Schwarz Attention, a hierarchical attention mechanism for LLMs inspired by domain decomposition methods, tested on operator learning.
FlashMemory-DeepSeek-V4 proposes Lookahead Sparse Attention, a new inference method to reduce GPU memory for ultra-long context LLMs.
Research identifies commonalities between neural scaling laws and the Vendi Score, showing both are submodular and a special case of matrix spectral functions.
Research proves a 'Behavioral Credibility Trilemma' for confidence-gated autonomous agents: maximum helpfulness, optimal calibration, and full autonomy are mutually exclusive.
New research introduces 'Unpack,' a backward recursion method to decompose credit through transformer sublayers for mechanistic interpretability.
Research introduces AOE, a method improving out-of-distribution detection for ML models by recalibrating outlier labels during training.
Research identifies a root cause for performance stagnation in PPO reinforcement learning, demonstrating that scaling to 1M parallel environments can prevent it.
New research proposes Counterfactual Shapley Credit Assignment, a principled method to isolate policy skill from environmental stochasticity in RL agents.
Research proposes ADWM, an autoregressive diffusion world model, to evaluate LLM agent policies off-policy using pre-collected trajectories.
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 proposes repurposing edge AI inference accelerators for on-device model adaptation, addressing compute and memory limits for training.
Researchers propose Similarity-based Generative Network (SGN) for synthetic data generation robust to distribution shifts, reducing the need for target-domain optimization.
Research finds low-rank optimizers like GaLore, used for memory-efficient LLM training, lack a stable, trackable subspace beyond a small core.
Research introduces MTSSL, a unified theoretical framework for semi-supervised learning that explains the role of the pseudo-labeling threshold τ.
Research introduces ECO, an efficient learning framework for Neural Combinatorial Optimization using batched preference optimization and a Mamba backbone.
Researchers introduced Building2Building (B2B), a large-scale benchmark for generalizable real-world reinforcement learning.
Researchers introduced SurvCF(t), a framework to generate counterfactual explanations for deep survival models used in predictive maintenance.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion