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
Hybrid Parallel Bessel KAN Model: a Kolmogorov Arnold Network: integrated framework for developing a novel architecture to reconstruct the fused-basis approach via a parameter free quadratic interaction The paper proposes a Kolmogorov-Arnold Network architecture using parallel Bessel branches and quadratic interactions to reduce parameters and conditioning problems. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Develops a specialized neural network architecture rather than methods for small, efficient language models.
Follow the Perturbed Leader for Linear Bandits: $\sqrt{T}$ Regret over General Convex Sets The paper provides an adversarial linear bandit algorithm achieving sublinear regret over general convex action sets using follow-the-perturbed-leader. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Analyzes theoretical bandit regret bounds rather than language model evaluation frameworks.
Follow the Perturbed Leader for Linear Bandits: $\sqrt{T}$ Regret over General Convex Sets The paper provides an adversarial linear bandit algorithm achieving sublinear regret over general convex action sets using follow-the-perturbed-leader. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Covers theoretical online optimization rather than small, efficient model architectures or compression.
Follow the Perturbed Leader for Linear Bandits: $\sqrt{T}$ Regret over General Convex Sets The paper provides an adversarial linear bandit algorithm achieving sublinear regret over general convex action sets using follow-the-perturbed-leader. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on theoretical linear bandits rather than deployment safety, guardrails, or hallucination.
GeoNIRS: Geometry-Aware Pretraining for Heterogeneous fNIRS Representation Learning The paper presents a geometry-aware pretraining framework that aligns heterogeneous probe layouts for functional near-infrared spectroscopy representation learning. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on brain signal pretraining rather than compact or efficient model architectures.
GeoNIRS: Geometry-Aware Pretraining for Heterogeneous fNIRS Representation Learning The paper presents a geometry-aware pretraining framework that aligns heterogeneous probe layouts for functional near-infrared spectroscopy representation learning. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies fNIRS neuroimaging representations rather than language model benchmark design.
GeoNIRS: Geometry-Aware Pretraining for Heterogeneous fNIRS Representation Learning The paper presents a geometry-aware pretraining framework that aligns heterogeneous probe layouts for functional near-infrared spectroscopy representation learning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers hemodynamic representation learning rather than AI safety or deployment guardrails.
Persistence Positional Encoding: A Minimal Transformer Recipe for Persistence Diagrams The paper proposes a minimal positional encoding method using random Fourier features to feed persistence diagrams directly into standard Transformers. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses topological data representations, not deployment safety or reliability risks.
Persistence Positional Encoding: A Minimal Transformer Recipe for Persistence Diagrams The paper proposes a minimal positional encoding method using random Fourier features to feed persistence diagrams directly into standard Transformers. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Investigates persistence diagram encodings rather than model compression or efficient serving.
Persistence Positional Encoding: A Minimal Transformer Recipe for Persistence Diagrams The paper proposes a minimal positional encoding method using random Fourier features to feed persistence diagrams directly into standard Transformers. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Evaluates topological representation methods on graph tasks rather than LLM benchmarking.
Searching for What Matters: Open-Ended Discovery of Multiple Hypotheses with Agentic Data Exploration The paper introduces an open-ended benchmark for evaluating agentic systems on discovering multiple data-supported hypotheses from tabular datasets. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies scientific hypothesis generation rather than safety, guardrails, or deployment risks.
Searching for What Matters: Open-Ended Discovery of Multiple Hypotheses with Agentic Data Exploration The paper introduces an open-ended benchmark for evaluating agentic systems on discovering multiple data-supported hypotheses from tabular datasets. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on exploratory hypothesis discovery rather than model compression or inference efficiency.
Deep Mesoscopic Dynamics The paper develops a neural network approach to learn effective mesoscopic field equations directly from microscopic trajectory data. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses physical system simulation, not AI deployment safety or reliability.
Deep Mesoscopic Dynamics The paper develops a neural network approach to learn effective mesoscopic field equations directly from microscopic trajectory data. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies mesoscopic physical dynamics rather than small, efficient model compression or serving.
Deep Mesoscopic Dynamics The paper develops a neural network approach to learn effective mesoscopic field equations directly from microscopic trajectory data. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on physics PDE surrogate modeling rather than language model evaluation.
FlashMask-3: Efficient and Expressive Mask-Aware Distributed Attention The paper presents an expressive and distributed attention kernel that compresses structured attention masks to enable efficient context parallelism on Hopper GPUs. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers GPU kernel and communication optimizations, not AI safety or guardrails.
FlashMask-3: Efficient and Expressive Mask-Aware Distributed Attention The paper presents an expressive and distributed attention kernel that compresses structured attention masks to enable efficient context parallelism on Hopper GPUs. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Introduces an attention kernel optimization, not an evaluation benchmark.
Stable-Plastic Subspace Construction for Differentially Private Continual LoRA The paper presents a subspace construction method for continual low-rank adaptation to mitigate forgetting under differential privacy constraints. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Addresses private continual fine-tuning methods, not evaluation benchmarks.
Stable-Plastic Subspace Construction for Differentially Private Continual LoRA The paper presents a subspace construction method for continual low-rank adaptation to mitigate forgetting under differential privacy constraints. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on privacy guarantees during adaptation, not model inference efficiency.
NBCLT: Parameter-Agnostic Norm-Ball Clipped Local Training The paper introduces a unified local training framework using norm-ball clipping to improve communication efficiency in distributed and federated learning. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on distributed optimization bounds, not small models or serving.
NBCLT: Parameter-Agnostic Norm-Ball Clipped Local Training The paper introduces a unified local training framework using norm-ball clipping to improve communication efficiency in distributed and federated learning. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies federated learning optimization algorithms, not LLM evaluation.
NBCLT: Parameter-Agnostic Norm-Ball Clipped Local Training The paper introduces a unified local training framework using norm-ball clipping to improve communication efficiency in distributed and federated learning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical federated learning convergence, not safety guardrails or deployment risks.
Stein Variational Distributional Reinforcement Learning: Score-Guided Particles for Multivariate Return Distributions The paper introduces a score-guided particle framework using Stein variational gradient descent to model multivariate return distributions in reinforcement learning. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Covers distributional reinforcement learning algorithms, not model compression or efficiency.
Stein Variational Distributional Reinforcement Learning: Score-Guided Particles for Multivariate Return Distributions The paper introduces a score-guided particle framework using Stein variational gradient descent to model multivariate return distributions in reinforcement learning. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on reinforcement learning algorithms, not LLM evaluation benchmarks.
Stein Variational Distributional Reinforcement Learning: Score-Guided Particles for Multivariate Return Distributions The paper introduces a score-guided particle framework using Stein variational gradient descent to model multivariate return distributions in reinforcement learning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses theoretical RL particle representation, not deployment safety or alignment.
Learning Distributions from Samples and Qualities This paper establishes accelerated parameter estimation rates when continuous data samples are accompanied by binary quality labels or pairwise preferences. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Covers continuous parameter estimation theory, not efficient AI deployment.
Learning Distributions from Samples and Qualities This paper establishes accelerated parameter estimation rates when continuous data samples are accompanied by binary quality labels or pairwise preferences. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical statistical estimation paper, not LLM evaluation.
Learning Distributions from Samples and Qualities This paper establishes accelerated parameter estimation rates when continuous data samples are accompanied by binary quality labels or pairwise preferences. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with theoretical sample distribution estimation, not AI safety or reliability.
CALM-5G: Event-Level Conformal Risk Control for Predictive Service-Assurance Alarms This paper proposes a conformal risk control layer that certifies miss rates and reduces alarm duty cycles in predictive cellular network operations. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on 5G network telemetry forecasting, not language model evaluation.
CALM-5G: Event-Level Conformal Risk Control for Predictive Service-Assurance Alarms This paper proposes a conformal risk control layer that certifies miss rates and reduces alarm duty cycles in predictive cellular network operations. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Applies conformal prediction to telecommunications alarms, not model efficiency.
Neural Dimensional Analysis: Foundations of Dimensionless and Scale-Consistent Computation This paper introduces a dimensional analysis framework and diagnostic metrics to maintain scale consistency across Transformer components. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Addresses model architectural scaling principles rather than evaluation benchmarks.
Neural Dimensional Analysis: Foundations of Dimensionless and Scale-Consistent Computation This paper introduces a dimensional analysis framework and diagnostic metrics to maintain scale consistency across Transformer components. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on training stability via dimensional analysis, not deployment safety or reliability.
Neural Dimensional Analysis: Foundations of Dimensionless and Scale-Consistent Computation This paper introduces a dimensional analysis framework and diagnostic metrics to maintain scale consistency across Transformer components. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Deals with training stability theory across scales, not small models or inference compression.
AlignPred: Error-Controlled KV Cache Compression via Alignment and Predictive Coding This paper presents an error-controlled compression codec for key-value caches using coordinate alignment and predictive coding. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses serving efficiency and memory compression, not AI safety or guardrails.
AlignPred: Error-Controlled KV Cache Compression via Alignment and Predictive Coding This paper presents an error-controlled compression codec for key-value caches using coordinate alignment and predictive coding. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on KV cache compression techniques rather than general LLM evaluation.
Rarity Determines Minimax Rates of Semantic Entropy From Weak Pairwise Judgments With a Known Channel This paper establishes minimax rates for estimating language model semantic entropy using verdicts from weak pairwise equivalence judges. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Theoretical information-theoretic analysis of entropy estimation, not model efficiency.
All-electron machine-learned orbital-free density functional for periodic CHNO systems This paper presents an equivariant neural network functional to perform orbital-free density functional theory for periodic CHNO systems. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Addresses molecular orbital modeling, not small language models or efficient inference.
All-electron machine-learned orbital-free density functional for periodic CHNO systems This paper presents an equivariant neural network functional to perform orbital-free density functional theory for periodic CHNO systems. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Concerns computational chemistry and density functional theory, not LLMs.
All-electron machine-learned orbital-free density functional for periodic CHNO systems This paper presents an equivariant neural network functional to perform orbital-free density functional theory for periodic CHNO systems. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on quantum chemistry calculations, not AI deployment safety.
Exact power indices for plurality-voting ensembles This paper introduces an exact method to compute Banzhaf and Shapley power indices for base classifiers in plurality-voting ensembles. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on game-theoretic power indices, not efficient model architectures.
Exact power indices for plurality-voting ensembles This paper introduces an exact method to compute Banzhaf and Shapley power indices for base classifiers in plurality-voting ensembles. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies voting ensemble attribution, not AI safety or deployment robustness.
Exact power indices for plurality-voting ensembles This paper introduces an exact method to compute Banzhaf and Shapley power indices for base classifiers in plurality-voting ensembles. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical attribution work on ensemble voting, not LLM evaluation.
Counterfactual Certification for Sequential Unlearning with Long-Run Performance Guarantees This paper proposes a counterfactual certification framework to bound performance loss during sequential machine unlearning requests. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Deals with certified machine unlearning, not LLM evaluation.
Counterfactual Certification for Sequential Unlearning with Long-Run Performance Guarantees This paper proposes a counterfactual certification framework to bound performance loss during sequential machine unlearning requests. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on theoretical unlearning guarantees, not model compression or efficiency.
Improving Neural Symbolic Regression with a Diverse Prior and Post-training Alignment This paper improves neural symbolic regression by using a tabular prior and reinforcement learning alignment to fit noisy data. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on symbolic regression algorithms rather than LLM evaluation or benchmarks.
Improving Neural Symbolic Regression with a Diverse Prior and Post-training Alignment This paper improves neural symbolic regression by using a tabular prior and reinforcement learning alignment to fit noisy data. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Addresses neural symbolic regression training recipes, not model compression or inference efficiency.
Improving Neural Symbolic Regression with a Diverse Prior and Post-training Alignment This paper improves neural symbolic regression by using a tabular prior and reinforcement learning alignment to fit noisy data. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on symbolic mathematical regression, not deployment safety or jailbreak prevention.
FRACTION: Fraud Explanation Generation via Agentic Graph Context Navigation This paper presents an agentic framework and an evaluation rubric for generating explanations of risk labels in graph fraud detection. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Presents an agentic tool for fraud explanation rather than safety or alignment guardrails.
FRACTION: Fraud Explanation Generation via Agentic Graph Context Navigation This paper presents an agentic framework and an evaluation rubric for generating explanations of risk labels in graph fraud detection. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on graph context navigation for fraud explanations, not model compression.
Weighting-Based Strategies for Classification: Analysis and Local Peak Dynamics with CART Trees This paper theoretically and empirically compares weighting-based rebalancing strategies to unweighted methods in imbalanced tabular classification tasks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies class-imbalance weighting in decision trees, not AI safety or deployment guardrails.
Weighting-Based Strategies for Classification: Analysis and Local Peak Dynamics with CART Trees This paper theoretically and empirically compares weighting-based rebalancing strategies to unweighted methods in imbalanced tabular classification tasks. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Analyzes class weighting in CART trees, not small or efficient models.
Weighting-Based Strategies for Classification: Analysis and Local Peak Dynamics with CART Trees This paper theoretically and empirically compares weighting-based rebalancing strategies to unweighted methods in imbalanced tabular classification tasks. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical tabular classification analysis, not language model evaluation.
LADDER: A Long-Horizon Robot Manipulation Benchmark This paper introduces a robot manipulation benchmark to isolate failures across scaling subgoal horizons and orchestration methods. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies long-horizon robot policies, not model compression or efficient serving.
LADDER: A Long-Horizon Robot Manipulation Benchmark This paper introduces a robot manipulation benchmark to isolate failures across scaling subgoal horizons and orchestration methods. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Evaluates robotics manipulation policies, not AI deployment safety or jailbreak risks.
LADDER: A Long-Horizon Robot Manipulation Benchmark This paper introduces a robot manipulation benchmark to isolate failures across scaling subgoal horizons and orchestration methods. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:18	Not relevant Covers general robot manipulation, not industrial inspection or defect detection.
LADDER: A Long-Horizon Robot Manipulation Benchmark This paper introduces a robot manipulation benchmark to isolate failures across scaling subgoal horizons and orchestration methods. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Introduces a robotic manipulation benchmark rather than an LLM evaluation benchmark.
Reliable Scale-Free Bandits via Empirical Likelihood This paper develops an empirical-likelihood bandit policy that provides asymptotic regret guarantees without knowledge of reward scales. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical study of multi-armed bandits, not LLM benchmarking.
Reliable Scale-Free Bandits via Empirical Likelihood This paper develops an empirical-likelihood bandit policy that provides asymptotic regret guarantees without knowledge of reward scales. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies stochastic bandit regret bounds, not model efficiency or serving.
Reliable Scale-Free Bandits via Empirical Likelihood This paper develops an empirical-likelihood bandit policy that provides asymptotic regret guarantees without knowledge of reward scales. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on theoretical bandit regret bounds rather than AI safety or guardrails.
Causal Discovery Under Hard Selection Bias: A New Robust Score-Matching Approach This paper proposes a score-matching approach for causal discovery in additive noise models under hard selection bias and truncation. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical causal discovery work, not language model evaluation.

9,361–9,420 of 17,313← Previous Next →