Probabilistic Residual Learning for Online Recommendations
Research proposes Probabilistic Residual Learning (PRL), a causal Bayesian approach for online recommender systems to enhance explainability.
arXiv cs.LG — Machine Learning·24 Jul 2026
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 Probabilistic Residual Learning (PRL), a causal Bayesian approach for online recommender systems to enhance explainability.
RadioTrace, a new diffusion model, estimates wireless signal distributions from sparse data without deployment-time fine-tuning, improving spectrum management.
Research explores Inductive Logic Programming (ILP) for explaining weather bulletins, leveraging the FastLAS2 framework.
Adaptive Depth Sparse Framework (AdaDSF) proposes converting off-the-shelf pre-trained LLMs into depth-sparse models to reduce inference cost.
Research explores the scaling behavior and generalization of deep parameterized quantum circuits (PQCs), a key challenge in quantum machine learning.
Research characterizes Chain-of-Thought reasoning non-convergence when models exhaust token budgets, showing a 90.3% accuracy drop.
Research proposes an interpretable causal-ensemble framework for preventive maintenance and fault prediction in climate-stressed EV charging infrastructure.
Research disentangles how code language models represent grammatical concepts, comparing Python/Rust with Qwen2.5-Coder-7B and DeepSeek-Coder-V1-6.7B.
Research explores multimodal LLMs for generalizable speech-based cognitive impairment detection, leveraging linguistic and acoustic markers.
Research paper proposes a theory that human participation in AI systems is not merely a temporary limitation but a persistent, conceptually necessary feature.
Research identifies vision-language models struggle with visual reasoning, often failing on diagrams but succeeding on text for the same problem.
Research details a synthetic data generation framework using deep learning to automate quality control in gravure printing, addressing real-world data scarcity.
Research demonstrates the Barzilai-Borwein optimization method fails superlinear convergence for an open set of quadratic problems in dimensions n ≥ 4.
Research explores localizing and modifying LLM representations for safety alignment, aiming to ensure interventions faithfully erase unwanted behaviors.
New research proposes a federated learning client-selection algorithm to mitigate data heterogeneity by forming client coalitions and selecting representatives.
Research paper explores communication-efficient distributed training algorithms for LLMs in bandwidth-constrained settings, focusing on methods like DiLoCo.
Research identifies reward hacking vulnerability in Group Relative Policy Optimization (GRPO) for multi-objective problems and proposes a mitigation.
Researchers propose Amortized Group Relative Policy Optimization, a reinforcement learning technique for Diffusion Large Language Models (dLLMs).
Researchers developed ActVAE, a deep conditional generative model for predicting human activity schedules based on individual and household data.
CLOAK introduces a latent diffusion model for time-series data obfuscation to enhance privacy against attribute inference attacks while preserving data utility.
NeuraLSP proposes a neural spectral preconditioner to accelerate solving sparse linear systems from PDEs, improving on traditional multigrid methods.
Physics-Informed Learning via Diffusion (PILD) is a new framework that integrates physical constraints into generative diffusion models using a probabilistic residual formulation.
Variational Speculative Decoding (VSD) is a new method for training draft models in speculative decoding to maximize target-model acceptance.
Research describes a 'self-evolving' recommendation system using LLM agents for autonomous model optimization, reducing manual iteration.
Research proposes reusing internal audio language model representations for temporal localization to improve speed and accuracy over token generation.
Research paper retracts prior claims of improved long-context retrieval for LLMs using Surprisal-Aware Residual Test-Time Training (SR-TTT), citing evaluation artifacts.
PISmith, a reinforcement learning framework, is proposed to red team and systematically assess prompt injection defenses for LLM applications.
Research paper proposes Bethe free energy minimization as an alternative training method for Bayesian neural networks, aiming for exact predictive density scoring.
Research details MuonQ, an optimization method to improve low-bit quantization for the Muon optimizer in large language models by preserving directional fidelity.
Research explores methods to accelerate the training of Masked Diffusion Language Models (MDMs), which are slower to learn than autoregressive models (ARMs).
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion