Every paper checked — nothing hidden

All papers

Every paper checked for brief #4 of 12–19 Oct 2026 — nothing is hidden. Brief papers first.

All17,313 In brief24 More (on-topic)3,333 Not relevant13,754 Not assessed87 Older352 Could not confirm93 Blogs with no paper266 All blogs284

Paper	Topic	Relevance
Why Heavy-Tailed Weight Spectra Predict Model Quality Provides a probabilistic marginal likelihood derivation explaining why heavy-tailed weight spectral metrics correlate with deep network generalization quality. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Studies statistical justification for model quality metrics, not AI safety or deployment guardrails.
Why Heavy-Tailed Weight Spectra Predict Model Quality Provides a probabilistic marginal likelihood derivation explaining why heavy-tailed weight spectral metrics correlate with deep network generalization quality. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Investigates theoretical quality metrics from weight spectra, not model compression or efficiency.
Fast Gradient-Based Approximation of Continuous Dynamic Time Warping Introduces fast gradient-based approximation schemes for continuous dynamic time warping to enable scalable, differentiable time series alignment. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on time-series dynamic time warping optimization, unrelated to AI safety or reliability.
Fast Gradient-Based Approximation of Continuous Dynamic Time Warping Introduces fast gradient-based approximation schemes for continuous dynamic time warping to enable scalable, differentiable time series alignment. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Proposes a fast time-series alignment solver, not small or efficient language model methods.
Fast Gradient-Based Approximation of Continuous Dynamic Time Warping Introduces fast gradient-based approximation schemes for continuous dynamic time warping to enable scalable, differentiable time series alignment. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Presents an approximation algorithm for Dynamic Time Warping, not LLM evaluation benchmarks.
WaSA: Wavelet Scattering Autoencoders for Unsupervised Offline Detection of Malicious Clients in Federated Learning Uses wavelet scattering autoencoders to detect malicious client data updates offline in federated learning without requiring attack labels. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on federated learning poisoning defense, not evaluating or benchmarking language models.
WaSA: Wavelet Scattering Autoencoders for Unsupervised Offline Detection of Malicious Clients in Federated Learning Uses wavelet scattering autoencoders to detect malicious client data updates offline in federated learning without requiring attack labels. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on malicious client detection in federated learning, not model compression or efficient serving.
WaSA: Wavelet Scattering Autoencoders for Unsupervised Offline Detection of Malicious Clients in Federated Learning Uses wavelet scattering autoencoders to detect malicious client data updates offline in federated learning without requiring attack labels. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses federated learning data poisoning in IoT, not LLM safety, guardrails, or alignment.
Prediction-Informed Landing Window Planning for UAV Landing on USV in High Sea States Develops a motion-prediction hierarchical control framework to plan safe touchdown windows for autonomous drone landings on surface vessels. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Develops robotic landing control algorithms, completely unrelated to LLM evaluation benchmarks.
Prediction-Informed Landing Window Planning for UAV Landing on USV in High Sea States Develops a motion-prediction hierarchical control framework to plan safe touchdown windows for autonomous drone landings on surface vessels. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:29	Not relevant Focuses on UAV flight trajectory planning and control, not computer vision defect inspection or perception.
Prediction-Informed Landing Window Planning for UAV Landing on USV in High Sea States Develops a motion-prediction hierarchical control framework to plan safe touchdown windows for autonomous drone landings on surface vessels. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Addresses UAV/USV maritime control systems, not small language models or efficient AI serving.
Prediction-Informed Landing Window Planning for UAV Landing on USV in High Sea States Develops a motion-prediction hierarchical control framework to plan safe touchdown windows for autonomous drone landings on surface vessels. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on autonomous vehicle control dynamics rather than AI safety, red-teaming, or LLM reliability.
Rethinking Relative Policy Differences Transfer of On-Policy Distillation Formulates relative policy difference transfer as an implicit reward in KL-regularized optimization for flexible on-policy knowledge distillation. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Studies distillation dynamics, not AI safety, jailbreaking, hallucination, or deployment guardrails.
Rethinking Relative Policy Differences Transfer of On-Policy Distillation Formulates relative policy difference transfer as an implicit reward in KL-regularized optimization for flexible on-policy knowledge distillation. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Investigates distillation mechanics and dynamics, not LLM evaluation benchmarks or methodologies.
How Much Computation Does Every Forecast Need? Structured Block Skipping in Two Time-Series Foundation Models Evaluates static Transformer block skipping in time-series foundation models to reduce inference latency with minimal forecasting degradation. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Evaluates layer skipping in time-series models, not LLM evaluation or benchmarking methodology.
How Much Computation Does Every Forecast Need? Structured Block Skipping in Two Time-Series Foundation Models Evaluates static Transformer block skipping in time-series foundation models to reduce inference latency with minimal forecasting degradation. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on dynamic computation skipping, not AI safety, alignment, or guardrails.
Verifiable Sampling for Delegated Generative Quantum Machine Learning Proposes a blind and verifiable sampling protocol for securely delegating quantum generative machine learning inference to remote servers. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on quantum generative model protocols, not small language models or efficient LLM serving.
Verifiable Sampling for Delegated Generative Quantum Machine Learning Proposes a blind and verifiable sampling protocol for securely delegating quantum generative machine learning inference to remote servers. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Examines cryptographic verification for quantum computing protocols, not standard AI safety or guardrails.
Verifiable Sampling for Delegated Generative Quantum Machine Learning Proposes a blind and verifiable sampling protocol for securely delegating quantum generative machine learning inference to remote servers. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies verifiable protocols for quantum computing, not LLM evaluation benchmarks.
Monarch-MoE: Structured Experts for Multimodal Continual Instruction Tuning Replaces dense LoRA experts with block-diagonal factorizations to isolate gradients and prevent catastrophic forgetting during continual multimodal tuning. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses catastrophic forgetting in parameter-efficient tuning, not AI safety, red-teaming, or guardrails.
Monarch-MoE: Structured Experts for Multimodal Continual Instruction Tuning Replaces dense LoRA experts with block-diagonal factorizations to isolate gradients and prevent catastrophic forgetting during continual multimodal tuning. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Presents parameter-efficient continual tuning architectures, not evaluation benchmarks or metrics.
Tensor-based Fusion of Multi-modal Data for Predicting Cancer Risk Levels Applies tensor decomposition and clustering to fuse heterogeneous clinical and genomic datasets for unsupervised cancer risk prediction. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Applies tensor decomposition to cancer risk prediction data, not LLM evaluation.
Tensor-based Fusion of Multi-modal Data for Predicting Cancer Risk Levels Applies tensor decomposition and clustering to fuse heterogeneous clinical and genomic datasets for unsupervised cancer risk prediction. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on clinical data clustering rather than small language models or LLM efficiency.
Tensor-based Fusion of Multi-modal Data for Predicting Cancer Risk Levels Applies tensor decomposition and clustering to fuse heterogeneous clinical and genomic datasets for unsupervised cancer risk prediction. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses clinical risk modeling, unrelated to AI safety, hallucination, or deployment security.
Learning Schwarzschild Geodesic Motion with Lagrangian Neural Networks Introduces coordinate normalization and energy-constrained sampling to train Lagrangian neural networks on complex Schwarzschild geodesic dynamics. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Investigates neural networks for relativistic physics, not language model evaluation.
Learning Schwarzschild Geodesic Motion with Lagrangian Neural Networks Introduces coordinate normalization and energy-constrained sampling to train Lagrangian neural networks on complex Schwarzschild geodesic dynamics. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses computational physics equations, not AI deployment safety or reliability guardrails.
Learning Schwarzschild Geodesic Motion with Lagrangian Neural Networks Introduces coordinate normalization and energy-constrained sampling to train Lagrangian neural networks on complex Schwarzschild geodesic dynamics. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on relativistic geodesic physics modeling, not small language models or model compression.
State-OPD: State-Grounded On-Policy Distillation for Multi-Turn Agents Investigates on-policy distillation for multi-turn agents by decoupling objective interaction states from internal student reasoning trajectories. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on distillation performance rather than AI safety, guardrails, or alignment risks.
State-OPD: State-Grounded On-Policy Distillation for Multi-Turn Agents Investigates on-policy distillation for multi-turn agents by decoupling objective interaction states from internal student reasoning trajectories. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on knowledge distillation techniques for agent policies, not evaluating or benchmarking LLMs.
Neural Surrogate Modeling of Ultrasonic Wave Propagation in Heterogeneous Media Combines Fourier neural operators with local-integral resampling in a U-Net to model high-frequency ultrasonic waves in heterogeneous media. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Presents surrogate models for ultrasonic wave simulation, not LLM evaluation.
Neural Surrogate Modeling of Ultrasonic Wave Propagation in Heterogeneous Media Combines Fourier neural operators with local-integral resampling in a U-Net to model high-frequency ultrasonic waves in heterogeneous media. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Addresses PDE wave equation surrogates, not small language models or efficient LLM inference.
Neural Surrogate Modeling of Ultrasonic Wave Propagation in Heterogeneous Media Combines Fourier neural operators with local-integral resampling in a U-Net to model high-frequency ultrasonic waves in heterogeneous media. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with ultrasonic physics simulation, having no relation to AI safety or reliability.
Data-Adaptive Mahalanobis Metric Learning for Cross-Head Attention in Transformers Generalizes multi-head attention by learning data-adaptive, full cross-head Mahalanobis metric matrices to capture inter-head feature interactions. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on attention architecture formulation rather than AI safety, red-teaming, or guardrails.
Data-Adaptive Mahalanobis Metric Learning for Cross-Head Attention in Transformers Generalizes multi-head attention by learning data-adaptive, full cross-head Mahalanobis metric matrices to capture inter-head feature interactions. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Introduces a general attention mechanism for model expressiveness, not model compression or efficiency serving.
Data-Adaptive Mahalanobis Metric Learning for Cross-Head Attention in Transformers Generalizes multi-head attention by learning data-adaptive, full cross-head Mahalanobis metric matrices to capture inter-head feature interactions. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Proposes an attention variant for Transformers, unrelated to LLM evaluation or benchmarks.
CopRoute: Dependence-Aware Conformal Routing for Language Models Proposes a conformal routing method that accounts for model failure dependencies to optimize multi-model query coverage under risk budgets. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies routing algorithms to minimize LLM inference calls, not evaluation benchmarks or metrics.
MHR-Router: Multimodal Heterogeneous Retrieval-Enhanced Adaptive Routing for VLM Inference Develops a retrieval-enhanced routing framework that dispatches multimodal queries to specialized vision-language models based on local failure patterns. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Optimizes expert routing for performance rather than addressing AI safety, hallucination, or red-teaming.
MHR-Router: Multimodal Heterogeneous Retrieval-Enhanced Adaptive Routing for VLM Inference Develops a retrieval-enhanced routing framework that dispatches multimodal queries to specialized vision-language models based on local failure patterns. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Proposes an adaptive query router for VLM inference, not an evaluation framework or benchmark.
Recovering Irreducible Three-Body Interactions with Lagrangian Neural Networks Applies Lagrangian neural networks with targeted sampling to recover non-pairwise three-body dispersion interactions directly from trajectory data. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies physics-informed neural networks for molecular physics, not LLM evaluation or benchmarks.
Recovering Irreducible Three-Body Interactions with Lagrangian Neural Networks Applies Lagrangian neural networks with targeted sampling to recover non-pairwise three-body dispersion interactions directly from trajectory data. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Investigates Lagrangian neural networks for three-body physics, not model compression or small language models.
Recovering Irreducible Three-Body Interactions with Lagrangian Neural Networks Applies Lagrangian neural networks with targeted sampling to recover non-pairwise three-body dispersion interactions directly from trajectory data. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with physical dynamic modeling, unrelated to deployment safety, alignment, or LLM guardrails.
DAG Structure Learning with Node-Varying Heavy-Tailed Disturbances Presents a DAG structure learning method that jointly estimates graph structure and node-varying heavy-tailed disturbance parameters. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Investigates causal DAG structure learning under heavy-tailed noise, not LLM benchmark evaluation.
DAG Structure Learning with Node-Varying Heavy-Tailed Disturbances Presents a DAG structure learning method that jointly estimates graph structure and node-varying heavy-tailed disturbance parameters. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on graphical model estimation, not AI deployment safety, hallucination, or jailbreaking.
DAG Structure Learning with Node-Varying Heavy-Tailed Disturbances Presents a DAG structure learning method that jointly estimates graph structure and node-varying heavy-tailed disturbance parameters. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on statistical graph learning algorithms, unrelated to small or efficient language models.
MORAD: Reinforcement Learning with Multi-Objective Rewards for RNA Inverse Design Introduces an online reinforcement learning framework optimizing multi-objective rewards across diverse target backbones for RNA inverse design. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Studies biological RNA sequence design, having no connection to AI safety or LLM guardrails.
MORAD: Reinforcement Learning with Multi-Objective Rewards for RNA Inverse Design Introduces an online reinforcement learning framework optimizing multi-objective rewards across diverse target backbones for RNA inverse design. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Deals with reinforcement learning for RNA sequence design, not LLM evaluation or benchmarks.
MORAD: Reinforcement Learning with Multi-Objective Rewards for RNA Inverse Design Introduces an online reinforcement learning framework optimizing multi-objective rewards across diverse target backbones for RNA inverse design. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on biological molecular generative modeling rather than efficient or small language models.
Structure-Aware Numerical Stabilization for Half-Precision Proximal Policy Optimization Proposes stabilization modules to enable half-precision FP16 proximal policy optimization training on commodity GPUs without numerical divergence. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on numerical stabilization for half-precision PPO reinforcement learning, not LLM benchmarking.
Structure-Aware Numerical Stabilization for Half-Precision Proximal Policy Optimization Proposes stabilization modules to enable half-precision FP16 proximal policy optimization training on commodity GPUs without numerical divergence. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses RL numerical optimization details rather than AI alignment, safety, or deployment guardrails.
Sustainable Cloud-Fog-Edge Systems for Smart Buildings: Algorithmic Adaptation under Memory Constraints and Dynamic Orchestration Strategies Presents an edge-computing architecture using low-cost hardware and adaptive algorithms to enable autonomous data management in smart buildings. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on IoT building orchestration, not AI safety, hallucination, or LLM guardrails.
Sustainable Cloud-Fog-Edge Systems for Smart Buildings: Algorithmic Adaptation under Memory Constraints and Dynamic Orchestration Strategies Presents an edge-computing architecture using low-cost hardware and adaptive algorithms to enable autonomous data management in smart buildings. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Covers IoT and smart building edge computing systems, not LLM evaluation methodology.
Sustainable Cloud-Fog-Edge Systems for Smart Buildings: Algorithmic Adaptation under Memory Constraints and Dynamic Orchestration Strategies Presents an edge-computing architecture using low-cost hardware and adaptive algorithms to enable autonomous data management in smart buildings. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on general edge IoT computing architectures rather than small language models or LLM compression.
JEPA-kMMVAE: CSI Prediction for Time-Varying Wireless Channels Combines joint-embedding predictive architecture with a Markov variational autoencoder to predict wireless channel state information under noisy observations. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with signal processing in wireless channels, not AI safety or deployment reliability.
JEPA-kMMVAE: CSI Prediction for Time-Varying Wireless Channels Combines joint-embedding predictive architecture with a Markov variational autoencoder to predict wireless channel state information under noisy observations. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Applies VAEs to telecommunications channel forecasting rather than efficient language model compression or serving.
JEPA-kMMVAE: CSI Prediction for Time-Varying Wireless Channels Combines joint-embedding predictive architecture with a Markov variational autoencoder to predict wireless channel state information under noisy observations. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on wireless channel state information prediction, completely unrelated to LLM evaluation.
SBNO : Schrödinger Bridge Neural Operator for the Forward and Inverse Problems under Missing Information Introduces a Schrödinger bridge diffusion matching method to solve forward and inverse partial differential equation problems under sparse observations. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies surrogate neural operators for PDEs, not LLM benchmarks or evaluation methods.
SBNO : Schrödinger Bridge Neural Operator for the Forward and Inverse Problems under Missing Information Introduces a Schrödinger bridge diffusion matching method to solve forward and inverse partial differential equation problems under sparse observations. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses scientific PDE modeling, not AI safety, alignment, or LLM deployment guardrails.
SBNO : Schrödinger Bridge Neural Operator for the Forward and Inverse Problems under Missing Information Introduces a Schrödinger bridge diffusion matching method to solve forward and inverse partial differential equation problems under sparse observations. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on solving partial differential equations with diffusion bridges, not small language models or LLM compression.
DuoRate: Two-Rate Per-Expert Precision Allocation for Mixture-of-Experts Serving Allocates precision across MoE experts by separately handling resident storage footprint and active batch decode traffic via dynamic optimization. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Investigates inference efficiency and quantization, not AI safety, jailbreaks, or deployment reliability risks.
DuoRate: Two-Rate Per-Expert Precision Allocation for Mixture-of-Experts Serving Allocates precision across MoE experts by separately handling resident storage footprint and active batch decode traffic via dynamic optimization. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on precision allocation and serving efficiency for MoE, not LLM evaluation benchmarks.

11,521–11,580 of 13,754← Previous Next →