Every paper checked — nothing hidden

All papers

Every paper checked in the week of 12–18 Oct 2026 · AI safety and reliability in deployment — nothing is hidden. Brief papers first.

All3,579 In brief3 More (on-topic)755 Not relevant2,801 Not assessed3 Older63 Could not confirm4 Blogs with no paper8 All blogs8

Paper	Topic	Relevance
Sensitivity Fidelity Reveals Physical Dependencies Beyond Forecast Accuracy in Deep-Learning Earth-System Models The paper proposes sensitivity fidelity as an evaluation criterion to verify physical consistency in deep-learning Earth-system forecasting models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on Earth-system model physics rather than AI safety or reliability.
Beyond Aggregate Drift: Optimal Client-Specific Windows in Federated Learning The paper presents an accelerated federated learning method that assigns client-specific memory windows based on individual noise and drift rates. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with federated learning convergence bounds rather than deployment safety.
Vision-Predictor-Action: Towards Efficient Embodied AI via Neuro-Symbolic JEPA Predictors The paper proposes a robot control framework using predictive architectures and flow matching to generate action chunks without autoregressive decoding. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on inference latency and control rather than safety or reliability guardrails.
UNISON: A Unified Framework for Learning from Scored Bags The paper presents a unified three-stage framework for multiple instance learning and recommendation from scored bag collections. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on recommender system architectures rather than AI safety guardrails.
Latent-Space Importance Weighting under Covariate Shift The paper proposes a latent-space importance weighting algorithm to improve prediction accuracy under covariate shift. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers covariate shift bounds rather than AI safety or reliability.
An Attention-Guided Graph Neural Network Surrogate for Boundary Integral Computation in Physical Field Reconstruction The paper introduces an attention-guided graph neural network surrogate to reconstruct missing boundary information in physical field simulations. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Pertains to physical field engineering computation, not AI safety risks.
Joint Contrastive-Adversarial Representation Learning for Unsupervised Domain Adaptation The paper establishes theoretical risk bounds for a two-stage unsupervised domain adaptation method combining contrastive learning and adversarial alignment. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Analyzes statistical bounds for domain adaptation, not safety or reliability.
ProML: Progressive Multi-label Learning The paper introduces a progressive multi-label learning framework that decouples direct label fitting from predictor optimization via an auxiliary variable. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on mathematical optimization for multi-label tasks, not AI safety.
Adaptive Role Assignment and Controlled Feature Reconstruction for Federated Image Semantic Communication The paper introduces an adaptive role-assignment and feature reconstruction framework to improve hybrid federated image communication under wireless constraints. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with wireless network transmission rather than deployment safety or jailbreaks.
The Price of Precision: A Calibrated Cost Model for Mixed-Precision LLM Serving The paper presents a hardware-calibrated cost model to optimize mixed-precision serving configurations across varying batch sizes and context lengths. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses serving resource constraints rather than AI safety or reliability guardrails.
MERIT: Scaling KG Foundation Models with Relation Transformers The paper presents a transformer-based relation encoder for knowledge graph foundation models to reduce scaling and inference costs. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on link prediction efficiency rather than AI safety or reliability.
Context-Aware Precision Scheduling for Small Language Models The paper proposes a context-aware mixed-precision scheduling method to balance execution latency and accuracy for small language models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on latency and precision tradeoffs rather than safety guardrails or robustness.
Testing Central Structure for Complex Networks The paper proposes a hypothesis testing framework using spectral radius to determine if a complex network contains true central structure. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with graph centrality testing rather than AI safety or reliability.
Drift, Then Spread: Learning One-Step Generative Policies for Online Reinforcement Learning The paper proposes a one-step generative reinforcement learning policy that balances reward improvement and entropy without iterative sampling. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on RL sample and inference speed, not AI safety.
When do explicit penalties improve flow matching? The paper derives geometric and statistical criteria showing when explicit penalties improve sample accuracy in flow matching generative models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Analyzes statistical trade-offs in generative modeling, not deployment safety.
Notes on Uniform Policy Score Bounds and Lipschitz Continuity for Truncated Gaussian, Product Cauchy, Boltzmann, and Softmax Policies The paper establishes uniform policy score bounds and Lipschitz continuity conditions across four common reinforcement learning policy parameterizations. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Mathematical statistics note on RL bounds, not AI safety.
DRanker: A Transformer-Based Multi-Task Ranking Model for the Disney+ Homepage The paper presents an encoder-only transformer ranking model deployed at Disney+ that uses lightweight debiasing to improve recommendations. Peer-reviewed	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers recommendation ranking systems, not AI safety or guardrails.
Adaptive Robust Confidence Sets in High Dimensions This paper analyzes the statistical costs and computational lower bounds of constructing adaptive confidence sets under unknown Huber contamination. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Statistical theory on Huber contamination, not AI deployment safety.
A Dependency-Aware Multi-GPU Training Framework for Temporal Graph Neural Networks This paper presents a multi-GPU training framework for temporal graph neural networks that relaxes event dependencies to reduce synchronization. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on GPU training throughput, not AI safety or reliability.
Inference on the Value of Policy Improvement in Adaptive Experiments This paper develops an asymptotically normal estimator to measure the value of policy improvement in adaptive experimental settings. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Statistical theory for adaptive experiments, not AI safety or guardrails.
Benchmarking Deep Learning Models for Anomaly Detection in Real-World Cloud Telemetry This paper benchmarks deep learning architectures for anomaly detection on real-world cloud telemetry using timing-aware evaluation metrics. Peer-reviewed	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on IT cloud telemetry monitoring, not AI model safety.
Guarded Block Transfers for Polytope Distance: Linear Convergence with Explicit Constants and Gap-Certified Screening for SVM Training This paper presents guarded block transfers for polytope distance computation to accelerate support vector machine training. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on convex optimization for SVMs, not AI safety.
How Shared Representations Become Specialized: Formation, Maintenance, and Latent Susceptibility This paper analyzes how shared representations specialize among multiple targets and how optimizer state influences this specialization. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Basic science study of representation drift, not deployment AI safety.
GuidanceNFT: NFT-Based Reward Fine-Tuning of Guided Diffusion Models This paper introduces an online flow matching method to fine-tune guided diffusion models toward specific rewards without losing prompt alignment. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant About diffusion alignment with prompts, not system deployment safety guardrails.
A Hybrid Gaussian Process-Extreme Gradient Boosting Machine Learning Framework for Robust Knockdown Factor Prediction and Reliability-Based Design of Thin Spherical Shell Caps This paper combines Gaussian process regression and gradient boosting to predict buckling knockdown factors for thin spherical shell caps. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Mechanical buckling prediction, not AI safety or deployment reliability.
Entropy-Regularized Mean-Field Games with Average-Reward Criterion: Span-Seminorm Contractions and Model-Free Learning This paper establishes contraction properties and finite-sample bounds for model-free learning in average-reward entropy-regularized mean-field games. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Mathematical game theory paper, not AI safety or guardrails.
Minimax Rates and Improved Spectral Convergence for Multi-Manifold Clustering This paper establishes minimax rates and improved spectral convergence bounds for multi-manifold clustering using Laplace-Beltrami operator estimators. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical clustering analysis, not AI safety or deployment robustness.
Lightweight Scheduling Policies for Continuously Batched LLM Serving Across GPU Architectures This paper evaluates lightweight scheduling policies for continuous batching in large language model serving across different GPU architectures. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on GPU serving throughput and latency, not safety risks.
Two Linear Programs are Better than One This paper proves that virtual extension complexity for polytopes can be exponentially smaller than standard extension complexity. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Abstract complexity theory paper, not AI safety or reliability.
Sample-Optimal Multi-Group Learning with an ERM Oracle This paper introduces an empirical risk minimization oracle algorithm that achieves optimal sample complexity for multi-group learning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical multi-group learning paper, not deployment safety or robustness guardrails.
Fast and Accurate: Improving Personalized LLM Performance by Semantic Concentration Inspired Workload Split between RAG and Fine-Tuning This paper splits personalized language model workloads between parameter-efficient fine-tuning and retrieval based on historical data semantic concentration. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses latency and accuracy via workload partitioning, not safety or guardrails.
Mesh-Free Convolution: Learned Spectral Attenuation and Transport This paper proposes continuous-space mesh-free convolutions to improve the stability of learned operators for partial differential equations across resolutions. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with numerical PDE stability, not AI safety or deployment guardrails.
Error-Based Bidirectional Predictive Coding Enables Deep, 4-Bit Predictive Coding Networks This paper combines bidirectional and error-based predictive coding to enable deep, 4-bit quantized predictive coding networks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses low-precision neural architectures, not AI deployment safety or guardrails.
Physics-guided machine learning for saturation control across variable hydraulic conditions in heap leaching This paper develops a physics-guided machine learning surrogate model to predict and control hydraulic saturation profiles in heap leaching. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Applies machine learning to hydraulic control, not AI deployment safety.
Small area learning This paper develops a semi-supervised learning framework for small area estimation that accommodates covariate and conditional distribution shifts. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with survey statistics and area estimation, not AI deployment safety.
Instance-Adaptive Online Multiclass Multicalibration This paper provides an online multiclass multicalibration algorithm with error bounds that adapt to the geometric dimension of outcome means. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical multicalibration analysis, not practical deployment safety or guardrails.
Harmonic Representations in Symbolic Music Transformers are Localized and Controllable This paper shows that harmonic representations in symbolic music transformers are localized across middle layers and can guide music generation. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies harmonic steering in music models, not AI deployment safety or jailbreaks.
Using Line Segment Tracking to enable ML-based Track Reconstruction in CMS This paper introduces a transformer-based track reconstruction algorithm using line segment tracking objects for high-pileup particle collisions. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Applies machine learning to physics detectors, not AI deployment safety.
Softmax-Free Attention in Spiking Transformers as the Zero-Temperature Limit of Two-Level Systems This paper provides a thermodynamic framework explaining energy efficiency and optimization issues in softmax-free spiking transformers. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on theoretical attention thermodynamics, not AI deployment safety or guardrails.
Probabilistic Forecasting of Wildfire Size Using Conditional Flow Matching This paper proposes a conditional flow matching model to predict the temporal evolution of wildfire acreage from weather and terrain data. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Applies machine learning to wildfire prediction, not AI deployment safety.
Dissipative Lagrangian Neural Networks for the Interpretable Modelling of Passive Mechanical Systems This paper presents a neural network method that separately learns conservative and dissipative dynamics to model passive mechanical systems. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on mechanical system modeling, not AI deployment safety or robustness.
A 4-Qubit Variational Quantum Classifier for Sequence-Specific Classification of DNMT3A and DNMT3B Flanking Methylation Preferences This paper applies a simulated four-qubit variational quantum classifier to distinguish sequence-specific DNA methylation preferences. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses biological sequence classification, not AI deployment safety.
Limit Theory for Online Expectile Estimation under Short and Long Range Dependence This paper develops asymptotic theory and convergence limits for online expectile estimation under dependent time series data. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on statistical risk estimation theory, not AI safety or guardrails.
A Nonlinear Time-Series Theory for Constant-step SGD with Momentum This paper provides theoretical convergence and distributional limits for constant-step stochastic gradient descent with momentum viewed as a time series. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on optimization convergence limits, not AI safety or system reliability.
CLNO: Conservative Local Neural Operator for Compressible Flow Prediction This paper proposes a local neural operator for compressible flow prediction that enforces discrete physical conservation across subdomains. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Applies neural operators to compressible flows, not deployment safety guardrails.
How the Classical Baseline Shapes Machine-Learning Speed Claims in Topology Optimization This paper audits speedup claims of machine learning models in structural topology optimization against modern classical solvers. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Investigates speed comparison claims in physics engineering, not deployment safety or jailbreaks.
Interpretable Graph Partitioning for Spatial Prediction This paper develops an interpretable graph partitioning approach to create simplified surrogate models for spatial point predictions. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Discusses spatial model interpretability, not AI safety or deployment reliability guardrails.
Memorization in image classification: test set behavior and scaling laws This paper studies the behavior, scaling laws, and predictive variance of image classification networks trained to memorize random labels. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on CNN memorization theory, not deployment safety or hallucination guardrails.
CloneOT: predicting clone-specific perturbation responses with lineage-supervised neural inverse optimal transport This paper introduces a neural inverse optimal transport framework that uses cell lineage tracking to learn perturbation response costs. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses computational biology perturbation responses, not AI safety or deployment robustness.
SelectTTRL: Representation-Guided Data Selection for Efficient Test-Time Reinforcement Learning This paper presents a method using internal representation similarity to identify which test problems benefit from test-time reinforcement learning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with sample efficiency during test-time training, not safety or guardrails.
Silence and Noise: Resolving Suppression and Overvaluation in Multimodal Entity Alignment This paper identifies modality imbalance failure modes in multimodal knowledge graph alignment and proposes methods to address them. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers multimodal knowledge graphs rather than AI deployment safety or jailbreaks.
XForecast: The KDD 2026 Finance Day Challenge on News-Grounded Financial Forecasting This paper introduces a financial forecasting benchmark and competition that evaluates four-week stock movement predictions using both prices and news. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on financial stock forecasting rather than AI safety, red-teaming, or guardrails.
Distributional Regression for Arbitrary Response Variables Using ReLU Neural Networks This paper proposes a distributional regression framework using ReLU neural networks to model conditional event probabilities on arbitrary response spaces. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses mathematical regression theory rather than deployment safety or robustness.
BARTER: Compact Dynamic Sparse Training with Standard Dense Kernels This paper presents a dynamic sparse training method that reduces compute overhead by maintaining compact tensors with standard dense kernels. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses training efficiency rather than deployment safety, hallucinations, or guardrails.
Equilibrium Learning in Unbounded Stochastic Games with Gaussian Transitions This paper studies equilibrium learning in unbounded stochastic games using continuous trajectories to fit Gaussian process dynamics. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies theoretical equilibrium learning rather than AI safety or reliability.
Exploration strategies for Gaussian process bandit optimization with inference regret guarantees This paper introduces exploration queries for Gaussian process Bayesian optimization to provide theoretical guarantees on the recommended final solution. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers theoretical bandit regret rather than deployment safety or jailbreak prevention.
MSC: Multi-Stage Caster-aware Control Framework with Uncertainty-Aware NMPC and RL-based Execution Layer This paper proposes a hierarchical control framework combining model predictive control and reinforcement learning to stabilize autonomous mobile robots with swiveling casters. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on mechanical caster control rather than AI safety or alignment.
When Participant Disjoint Validation Is Not Enough to Establish Generalization in Parkinson's Gait Classification This paper examines how site-dependent data structures can produce misleadingly high performance estimates in Parkinson's gait classification models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Examines medical data confounding rather than AI deployment safety or jailbreaks.
NetRL: Powering Network Decisions via NetGRPO-Driven Reinforcement Learning This paper introduces a two-stage reinforcement learning framework using language models to optimize complex network control tasks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses network resource allocation rather than AI safety or system guardrails.
From Simulated Fluids to Living Tissue: Zero-Shot Foundation Model Transfer to Perturbed Embryos This paper adapts a simulated fluid foundation model to forecast living embryo tissue dynamics under genetic and temperature perturbations. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses biological video dynamics, not deployment safety or reliability risks.

1,861–1,920 of 3,579← Previous Next →