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
Compiling Multi-Threshold Operators for Quantized Non-Monotonic and Fine-Granular Activations in Custom-Tailored DNN Accelerators The paper proposes an ONNX compiler method to map quantized non-monotonic and fine-granular activations into multi-threshold operators for hardware accelerators. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on FPGA hardware quantization, not safety or guardrails.
Compiling Multi-Threshold Operators for Quantized Non-Monotonic and Fine-Granular Activations in Custom-Tailored DNN Accelerators The paper proposes an ONNX compiler method to map quantized non-monotonic and fine-granular activations into multi-threshold operators for hardware accelerators. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on hardware operator compilation, not LLM evaluation or benchmarks.
Local Epochs as Optimal Stopping: Theory and Evidence in Federated Learning The paper frames client local training in federated learning as an optimal stopping problem to balance local updates against drift. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on federated learning stopping rules, not LLM evaluation benchmarks.
Local Epochs as Optimal Stopping: Theory and Evidence in Federated Learning The paper frames client local training in federated learning as an optimal stopping problem to balance local updates against drift. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Deals with federated learning local epochs rather than model efficiency methods.
Local Epochs as Optimal Stopping: Theory and Evidence in Federated Learning The paper frames client local training in federated learning as an optimal stopping problem to balance local updates against drift. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on federated optimization, not AI safety or deployment guardrails.
Physics-Aware Distillation of a Grid Foundation Model for Real-Time Edge Inference The paper distils a power-grid foundation model into a compact student network for real-time edge inference on power systems. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on model distillation, not deployment safety or guardrails.
Physics-Aware Distillation of a Grid Foundation Model for Real-Time Edge Inference The paper distils a power-grid foundation model into a compact student network for real-time edge inference on power systems. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on power-flow models, not language model evaluation.
Lethe: Link Inference Attacks For Evaluation of Edge Unlearning Methods This paper presents a benchmark and link inference attack framework to evaluate edge unlearning methods in graph neural networks. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Benchmarking is for graph edge unlearning, not language models.
Lethe: Link Inference Attacks For Evaluation of Edge Unlearning Methods This paper presents a benchmark and link inference attack framework to evaluate edge unlearning methods in graph neural networks. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not study model compression, distillation, or efficient inference.
Learning Lean Feedback for Efficient Proof-Candidate Evaluation on GPUs This paper trains GPU-based neural evaluators to predict Lean 4 compiler verdicts and feedback, accelerating proof-candidate filtering. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Develops neural evaluators for Lean proofs, not general LLM benchmarks.
Learning Lean Feedback for Efficient Proof-Candidate Evaluation on GPUs This paper trains GPU-based neural evaluators to predict Lean 4 compiler verdicts and feedback, accelerating proof-candidate filtering. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on training throughput for theorem provers, not safety.
Graph Dirichlet Energy for Fitting-Free Bivariate Causal Inference This paper proposes a Dirichlet energy criterion on kernel similarity graphs for bivariate causal inference without regression fitting or independence testing. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on bivariate causal discovery, not LLM evaluation.
Graph Dirichlet Energy for Fitting-Free Bivariate Causal Inference This paper proposes a Dirichlet energy criterion on kernel similarity graphs for bivariate causal inference without regression fitting or independence testing. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not study model compression or efficient inference methods.
Graph Dirichlet Energy for Fitting-Free Bivariate Causal Inference This paper proposes a Dirichlet energy criterion on kernel similarity graphs for bivariate causal inference without regression fitting or independence testing. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Covers statistical causal inference, not AI deployment safety.
Self-Calibrated Subspace Robust PCA for Parameter-Free Removal of Mixed Noise and Line Stripes in Hyperspectral Images This paper presents a self-calibrated robust PCA method for removing mixed noise and line stripes in hyperspectral images without preset parameters. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not cover small models or inference efficiency techniques.
Self-Calibrated Subspace Robust PCA for Parameter-Free Removal of Mixed Noise and Line Stripes in Hyperspectral Images This paper presents a self-calibrated robust PCA method for removing mixed noise and line stripes in hyperspectral images without preset parameters. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on image processing algorithms, not deployment AI safety.
Self-Calibrated Subspace Robust PCA for Parameter-Free Removal of Mixed Noise and Line Stripes in Hyperspectral Images This paper presents a self-calibrated robust PCA method for removing mixed noise and line stripes in hyperspectral images without preset parameters. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Covers hyperspectral image denoising, not LLM evaluation.
The Tilted Sampling Problem: Optimism, Posterior Sampling, and Logarithmic Regret This paper formulates tilted sampling as an online learning problem for sampling from Gibbs distributions and establishes logarithmic regret bounds. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on online learning regret theory, not LLM benchmarks.
The Tilted Sampling Problem: Optimism, Posterior Sampling, and Logarithmic Regret This paper formulates tilted sampling as an online learning problem for sampling from Gibbs distributions and establishes logarithmic regret bounds. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not address model compression or efficient serving.
The Tilted Sampling Problem: Optimism, Posterior Sampling, and Logarithmic Regret This paper formulates tilted sampling as an online learning problem for sampling from Gibbs distributions and establishes logarithmic regret bounds. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Analyzes sampling regret theory, not deployment safety or guardrails.
The Lagrangian Structure of Cost-Penalized Active Feature Acquisition This paper analyzes the theoretical Lagrangian structure of cost-penalized heuristics in active feature acquisition under hard budget constraints. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not address model compression or efficient LLM architectures.
The Lagrangian Structure of Cost-Penalized Active Feature Acquisition This paper analyzes the theoretical Lagrangian structure of cost-penalized heuristics in active feature acquisition under hard budget constraints. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Explores optimization theory for features, not deployment safety.
The Lagrangian Structure of Cost-Penalized Active Feature Acquisition This paper analyzes the theoretical Lagrangian structure of cost-penalized heuristics in active feature acquisition under hard budget constraints. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Studies active feature acquisition theory, not LLM evaluation.
SelectPFN: Amortized Bayesian Variable Selection for Sparse Nonlinear Regression with Prior-Data Fitted Networks This paper introduces a prior-data fitted network that performs simultaneous Bayesian variable selection and nonlinear prediction in a single forward pass. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Covers regression variable selection, not deployment safety risks.
SelectPFN: Amortized Bayesian Variable Selection for Sparse Nonlinear Regression with Prior-Data Fitted Networks This paper introduces a prior-data fitted network that performs simultaneous Bayesian variable selection and nonlinear prediction in a single forward pass. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Examines tabular Bayesian variable selection, not LLM evaluation.
SelectPFN: Amortized Bayesian Variable Selection for Sparse Nonlinear Regression with Prior-Data Fitted Networks This paper introduces a prior-data fitted network that performs simultaneous Bayesian variable selection and nonlinear prediction in a single forward pass. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not investigate model compression or efficient LLM inference.
ChronoSheaf: Joint Spatio-Temporal Spectral Filtering on Cellular Sheaves This paper introduces a spatio-temporal spectral sheaf Laplacian and filter to jointly capture space and time dependencies on graphs. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not cover model quantization, distillation, or serving.
ChronoSheaf: Joint Spatio-Temporal Spectral Filtering on Cellular Sheaves This paper introduces a spatio-temporal spectral sheaf Laplacian and filter to jointly capture space and time dependencies on graphs. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Proposes spatio-temporal sheaf networks, not LLM evaluation.
ChronoSheaf: Joint Spatio-Temporal Spectral Filtering on Cellular Sheaves This paper introduces a spatio-temporal spectral sheaf Laplacian and filter to jointly capture space and time dependencies on graphs. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Presents graph neural network theory, not deployment safety.
Uncertainty-Aware Federated Learning for Joint Communication and Computation in XR Systems This paper presents an uncertainty-aware federated learning method to manage power and computing allocation for extended reality devices under state uncertainties. Peer-reviewed	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Examines wireless communication and computation, not LLM evaluation.
Uncertainty-Aware Federated Learning for Joint Communication and Computation in XR Systems This paper presents an uncertainty-aware federated learning method to manage power and computing allocation for extended reality devices under state uncertainties. Peer-reviewed	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Investigates telecommunications latency under uncertainty, not AI safety guardrails.
Uncertainty-Aware Federated Learning for Joint Communication and Computation in XR Systems This paper presents an uncertainty-aware federated learning method to manage power and computing allocation for extended reality devices under state uncertainties. Peer-reviewed	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on edge resource allocation, not model compression.
Learn Only Where You Can't Compute: Structured Corrections to Gaussian Process Posteriors This paper combines Gaussian process references with targeted diffusion model corrections to infer scientific physical fields from sparse data. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on scientific inference methods, not deployment safety.
Learn Only Where You Can't Compute: Structured Corrections to Gaussian Process Posteriors This paper combines Gaussian process references with targeted diffusion model corrections to infer scientific physical fields from sparse data. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not study small language models or inference efficiency.
Learn Only Where You Can't Compute: Structured Corrections to Gaussian Process Posteriors This paper combines Gaussian process references with targeted diffusion model corrections to infer scientific physical fields from sparse data. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on Gaussian processes and diffusion, not LLM evaluation.
Deep Learning of Topological Equivalence Classes This paper applies deep learning to classify topological equivalence classes of finite group actions on Riemann surfaces. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Addresses pure mathematics, not safety or reliability in deployment.
Deep Learning of Topological Equivalence Classes This paper applies deep learning to classify topological equivalence classes of finite group actions on Riemann surfaces. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Studies Riemann surface group actions, not LLM evaluation.
Deep Learning of Topological Equivalence Classes This paper applies deep learning to classify topological equivalence classes of finite group actions on Riemann surfaces. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not examine model compression or efficient serving.
Interfaces and Modules: A Scalable Representation for Weight-Space Learning across Architectures This paper introduces an interface-based graph representation and encoder for scalable weight-space learning across diverse neural network architectures. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not focus on reducing model footprint or efficient serving.
Interfaces and Modules: A Scalable Representation for Weight-Space Learning across Architectures This paper introduces an interface-based graph representation and encoder for scalable weight-space learning across diverse neural network architectures. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Covers neural architecture representations, not AI safety or guardrails.
Interfaces and Modules: A Scalable Representation for Weight-Space Learning across Architectures This paper introduces an interface-based graph representation and encoder for scalable weight-space learning across diverse neural network architectures. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on weight-space representation learning, not LLM evaluation.
Stateful Testing for Deep Learning Compilers This paper introduces a stateful testing method to detect bugs in deep learning compilers caused by retained state across invocations. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Evaluates compiler correctness rather than model compression or efficiency.
Stateful Testing for Deep Learning Compilers This paper introduces a stateful testing method to detect bugs in deep learning compilers caused by retained state across invocations. Preprint	LLMs for code via arxiv · 17 Oct, 06:18	Not relevant Tests deep learning compilers, not models writing or reviewing code.
Stateful Testing for Deep Learning Compilers This paper introduces a stateful testing method to detect bugs in deep learning compilers caused by retained state across invocations. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on software compiler testing rather than LLM benchmarks.
Close Enough is Nowhere Near: Fidelity-Consumability Decoupling in KV Caches This paper investigates direct key-value cache substitution between frozen language models and finds geometric similarity fails to ensure consumability. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Analyzes KV cache substitution, not evaluation methodology or benchmarks.
Close Enough is Nowhere Near: Fidelity-Consumability Decoupling in KV Caches This paper investigates direct key-value cache substitution between frozen language models and finds geometric similarity fails to ensure consumability. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Explores transformer representations rather than safety or reliability issues.
Not All Unlabeled Data Are Equal: Unlabeled Data Selection for Semi-Supervised QSAR Modeling with Large Chemical Repositories This paper investigates unlabeled data selection strategies for semi-supervised QSAR modeling using large chemical repositories. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Investigates semi-supervised learning for chemistry, not LLM evaluation.
Not All Unlabeled Data Are Equal: Unlabeled Data Selection for Semi-Supervised QSAR Modeling with Large Chemical Repositories This paper investigates unlabeled data selection strategies for semi-supervised QSAR modeling using large chemical repositories. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not study model compression or efficient LLM inference.
Not All Unlabeled Data Are Equal: Unlabeled Data Selection for Semi-Supervised QSAR Modeling with Large Chemical Repositories This paper investigates unlabeled data selection strategies for semi-supervised QSAR modeling using large chemical repositories. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Addresses molecular property prediction, not deployment safety or robustness.
Augment the Cache, Replace the Activation: A 2x2 Study of Internal-State Sharing This paper explores internal-state sharing in multi-agent language models through a central pool rather than pairwise transfer bridges. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Does not investigate model safety, guardrails, or hallucinations.
Augment the Cache, Replace the Activation: A 2x2 Study of Internal-State Sharing This paper explores internal-state sharing in multi-agent language models through a central pool rather than pairwise transfer bridges. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Studies multi-agent internal state communication, not LLM evaluation.
Designing Experiments for External Decision Makers This paper introduces usage-aware Bayesian experimental design to minimize risk when experimental data is processed by unknown or external decision makers. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Covers theoretical Bayesian experimental design, not LLM benchmarks.
Designing Experiments for External Decision Makers This paper introduces usage-aware Bayesian experimental design to minimize risk when experimental data is processed by unknown or external decision makers. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not address model efficiency or compression methods.
Designing Experiments for External Decision Makers This paper introduces usage-aware Bayesian experimental design to minimize risk when experimental data is processed by unknown or external decision makers. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Addresses Bayesian design risk, not AI deployment safety or guardrails.
ETH: Data-Balanced Heterogeneous Pipeline Parallelism for Training Large Models This paper presents a pipeline parallelism training system that balances data workloads across heterogeneous hardware while reducing communication overhead. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Addresses distributed model training, not LLM evaluation or benchmarks.
ETH: Data-Balanced Heterogeneous Pipeline Parallelism for Training Large Models This paper presents a pipeline parallelism training system that balances data workloads across heterogeneous hardware while reducing communication overhead. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on system parallelism rather than safety or reliability.
Missed Detections and False Alarms in a CNN Technosignature Detector on Breakthrough Listen Data This paper tests a convolutional neural network technosignature detector against synthetic injections in real radio data to reveal out-of-distribution failure modes. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on radio technosignature CNN detectors, not LLM evaluation.
Missed Detections and False Alarms in a CNN Technosignature Detector on Breakthrough Listen Data This paper tests a convolutional neural network technosignature detector against synthetic injections in real radio data to reveal out-of-distribution failure modes. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Studies astronomy signal detection rather than deployment safety or guardrails.
Missed Detections and False Alarms in a CNN Technosignature Detector on Breakthrough Listen Data This paper tests a convolutional neural network technosignature detector against synthetic injections in real radio data to reveal out-of-distribution failure modes. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not address model compression, quantization, or efficient inference.
Rethinking Successive Rejects Algorithms This paper optimizes the predefined phase-length schedules of Successive Rejects algorithms to improve error exponents in fixed-budget best-arm identification. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Analyzes best-arm bandit identification, not AI safety guardrails or reliability.

3,961–4,020 of 17,313← Previous Next →