Adaptive Bayesian Online Learning via Expert Aggregation
Research proposes aggregating Bayesian update rules as experts to achieve uncertainty-aware prediction on data streams with adaptive inferential choices.
arXiv cs.LG — Machine Learning·23 Jul
Search signals, briefings, benchmarks and glossary terms.
Search signals, briefings, 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.
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 explores dynamic architectural adaptation via Optimal Brain Damage Masking to enhance deep neural networks' robustness to noisy labels.
Research evaluates nine pre-trained embedding models for gender bias in ML-based recruitment, analyzing inference from unstructured CV text.
Research explores using neural collapse instability to prioritize test cases for Deep Neural Networks, aiming to reduce validation costs in critical domains.
Research introduces an isotropy-preserving spectral cap for Muon, a matrix-sign optimizer, to better understand its impact on LLM weight matrices.
New research proposes a method for learning local causal structures from observational data while addressing latent variables and selection bias.
Research explores autonomous collaborative learning among Tsetlin Machines (TMs) with consensus-based inference, extending distributed TM learning.
Research introduces a unifying framework for post-training methods in time series foundation models (TSFMs) to improve downstream deployment.
Research proposes a multi-stage dynamic selection framework for cross-project defect prediction, addressing distribution shifts.
Research explores expert-guided editing for time-series foundation model forecasts, allowing human feedback to revise model-generated trajectories.
Research proposes a novel clustering method for LLM inference at scale, ensuring per-sample quality control and reducing cost and latency bottlenecks.
Research frames Active Inference (AIF) as a convex Markov Decision Process, simplifying policy optimization for expected free energy minimization.
SCPP is a new open-source Python library that unifies the interface for various soft clustering methods, including fuzzy, probabilistic, and deep learning.
Research finds DreamerV3-family model-based RL agents catastrophically forget tasks, with the 'actor' component, not the 'world model', being the source of forgetting.
Research presents CURED, a web demonstrator unifying ML-based data cleaning and error models for tabular data.
Research proposes machine learning for plausibility-driven prioritization of candidate biomedical annotations to reduce expert review bottlenecks.
Research reviews Multi-Agent Reinforcement Learning (MARL) for adaptability, highlighting real-world violations of design assumptions.
AlphaRoute, a new adaptive search framework using LLMs, optimizes multi-objective routing in VLSI design, addressing NP-hard combinatorial problems.
Research finds optimizing reward model inference speed in RLHF pipelines significantly reduces bottlenecks, suggesting C++ implementations over PyTorch.
OPIUM is a training-free method introduced to mitigate unintended externalities like over-refusal and weakened safety from LLM activation steering.
New research on the Lattice Deduction Transformer (LDT) reveals neural 'solvers' make one-shot predictions rather than iterative deductions.
Research paper proposes advanced AI/ML models for proactive forecasting of time series network utilization KPIs to optimize resource provisioning.
New research proposes an algorithm for reinforcement learning that achieves asymptotically optimal regret without dependence on the task horizon.
Research explored Time Series Foundation Models (TSFMs) like TimesFM, Chronos, and MOIRAI for zero-shot heart rate variability forecasting from consumer wearables.
Research identifies conditions where knowledge graphs enhance reinforcement learning performance, rather than being neutral or detrimental.
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.
A Good Practice Guide for quantifying uncertainties in machine learning models applied to photoplethysmography (PPG) signals from wearables was published.
Researchers propose a neuromorphic-inspired Receptron model for efficient, non-linear classification at the edge, reducing compute and memory.
Research explores Gaussian averaging as a smooth surrogate for quantized neural networks, deriving bounds on local oscillation for discontinuous models.
Koopman Dreamer proposes a spectrally constrained latent dynamics core for world models to improve stability and control in long imagined trajectories.
Research introduces DynamicRubric, a method to co-evolve LLM evaluators and policies, addressing score-gap collapse in policy optimization.
Research explores statistical inference for optimal rank allocation in Low-Rank Adaptation (LoRA) for fine-tuning large language models.
Researchers propose a novel differentially private training framework for deep neural networks that selectively privatizes inputs, not labels.
Research proposes LAARA, a framework for adaptive rank allocation in parameter-efficient fine-tuning (PEFT), addressing the suboptimality of uniform rank.
STN-TGAT proposes a Transformer-Graph Attention Network for stock ranking and portfolio construction, combining temporal dynamics and cross-sectional dependencies.
New research proposes Memory Merge DQN, an alternative target network update for Deep Q-networks to improve stability and preserve value function structure.
New benchmark, CruiseBench, for aero-engine remaining useful life (RUL) prediction uses real-flight profiles, extending N-CMAPSS with more realistic data.
Research introduces confidence intervals for the ℓ2 Expected Calibration Error (ECE), enhancing statistical evaluation of model calibration.
New dataset and benchmark, Air Quality Arena, evaluates time-series foundation models for large-scale, multi-region air quality forecasting.
GeoDES, a new diffusion-based model, aims to improve storm-centered weather prediction by augmenting limited regional historical data.
HeadCast proposes a training-free method to reduce inference costs in autoregressive video generation models by optimizing attention head usage.
Research explores methods for estimating total variation distance between autoregressive model distributions, accounting for inference optimizations.
Researchers propose SynPre-FL, a framework integrating synthetic data pretraining with federated learning to improve privacy-preserving risk prediction.
Research proposes Boundary Embedding Shaping (BES) with adaptive contrastive learning to mitigate graph structural entanglement in GNNs, improving node classification.
Research explores low-rank implicit regularization stability in deep matrix factorization, clarifying conditions for gradient descent to favor low-rank structures.
Research introduces a framework to audit silent safety failures in tool-augmented LLM agents, focusing on malformed tool responses.
New research challenges current machine unlearning evaluation methods, finding they may favor models that retain forgotten data, proposing a restoration-based audit.
Research introduces REGEN, a method for distilling expert LLM knowledge into smaller models using offline reinforcement learning and replay-recycling to reduce compute costs.
New algorithm for linear quadratic systems achieves anytime regret of order $\mathcal{O}(\sqrt{t})$ with computational efficiency and explicit system dimension dependence.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion