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
TRACER: Trace-Regularized Training Against Indirect Prompt Injection The paper introduces an adversarial training framework that hardens models against indirect prompt injection across retrieval and tool interaction channels. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Deals with adversarial training against attacks, not model efficiency.
Student Copula Markov Networks The paper extends undirected graphical models to non-Gaussian data using copula densities and provides an estimation algorithm. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Deals with statistical copula networks, not small language models.
Student Copula Markov Networks The paper extends undirected graphical models to non-Gaussian data using copula densities and provides an estimation algorithm. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Theoretical copula graphical model paper, not an LLM benchmark.
Student Copula Markov Networks The paper extends undirected graphical models to non-Gaussian data using copula densities and provides an estimation algorithm. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on probabilistic graphical models, not AI safety deployments.
Online Tokenization with Optimal Delay in Linear Time The paper introduces an online tokenization algorithm that processes text streams in linear time with minimal algorithmic delay. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on streaming tokenization algorithms, not AI safety risks.
Online Tokenization with Optimal Delay in Linear Time The paper introduces an online tokenization algorithm that processes text streams in linear time with minimal algorithmic delay. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Presents a tokenization algorithm, not an LLM benchmark.
Towards Training and Inference of Graph Neural Networks in Graph Databases The paper shows how graph databases can execute graph neural network sampling, retrieval, and inference directly without separate in-memory pipelines. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Studies GNNs in graph databases, not small language models.
Towards Training and Inference of Graph Neural Networks in Graph Databases The paper shows how graph databases can execute graph neural network sampling, retrieval, and inference directly without separate in-memory pipelines. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on GNN database systems, not LLM evaluation benchmarks.
Towards Training and Inference of Graph Neural Networks in Graph Databases The paper shows how graph databases can execute graph neural network sampling, retrieval, and inference directly without separate in-memory pipelines. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Deals with GNN training infrastructure, not AI safety or hallucination.
Demystifying On-Policy Distillation: Where Does the Teacher's Advantage Live? The paper examines teacher interventions along student generation paths to identify how teacher guidance drives performance gains in on-policy distillation. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Investigates distillation mechanics rather than LLM benchmarking suites.
Demystifying On-Policy Distillation: Where Does the Teacher's Advantage Live? The paper examines teacher interventions along student generation paths to identify how teacher guidance drives performance gains in on-policy distillation. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on distillation dynamics rather than safety guardrails or reliability.
Learning in Strategic Queuing with Geometric Recharge: From Bandits to Systems The paper analyzes packet routing throughput when selfish queue agents use no-regret learning algorithms to select buffered servers. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on packet routing queues, not efficient model inference.
Learning in Strategic Queuing with Geometric Recharge: From Bandits to Systems The paper analyzes packet routing throughput when selfish queue agents use no-regret learning algorithms to select buffered servers. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Queuing bandit problem unrelated to LLM deployment safety.
Learning in Strategic Queuing with Geometric Recharge: From Bandits to Systems The paper analyzes packet routing throughput when selfish queue agents use no-regret learning algorithms to select buffered servers. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Analyzes multi-agent queuing theory, not LLM benchmarking.
Structure-aware Bayesian full waveform inversion via diffusion-regularized Stein variational gradient descent The paper proposes a Bayesian full waveform inversion framework that uses learned diffusion models as regularizers for geological reconstruction. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Pertains to geophysics waveform inversion, not LLM evaluation.
Structure-aware Bayesian full waveform inversion via diffusion-regularized Stein variational gradient descent The paper proposes a Bayesian full waveform inversion framework that uses learned diffusion models as regularizers for geological reconstruction. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on geophysical inverse modeling, not small model efficiency.
Structure-aware Bayesian full waveform inversion via diffusion-regularized Stein variational gradient descent The paper proposes a Bayesian full waveform inversion framework that uses learned diffusion models as regularizers for geological reconstruction. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Addresses geophysical inversion rather than AI safety or reliability.
Adaptive Computation with Conformal Certification The paper develops an adaptive computation framework offering conformal uncertainty guarantees for processing large scientific and medical imagery efficiently. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Deals with conformal risk certification for imaging, not LLM benchmarks.
Adaptive Computation with Conformal Certification The paper develops an adaptive computation framework offering conformal uncertainty guarantees for processing large scientific and medical imagery efficiently. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on scientific imaging compute adaptation, not small language models.
FireGraphNet: Sub-daily wildfire spread prediction with directed graph neural networks The paper proposes a directed graph neural network approach to forecast twelve-hour wildfire spread using spatial and environmental attributes. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Addresses environmental hazard modeling rather than AI deployment safety.
FireGraphNet: Sub-daily wildfire spread prediction with directed graph neural networks The paper proposes a directed graph neural network approach to forecast twelve-hour wildfire spread using spatial and environmental attributes. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Wildfire spread prediction with GNNs, not LLM evaluation.
FireGraphNet: Sub-daily wildfire spread prediction with directed graph neural networks The paper proposes a directed graph neural network approach to forecast twelve-hour wildfire spread using spatial and environmental attributes. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on geospatial graph models rather than model compression.
Spectral-MoE: Heterogeneous Per-Layer Expert Utilization in Dense-to-MoE Conversion The paper introduces a method to convert dense models into mixture-of-experts by dynamically allocating different numbers of active experts across layers. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on MoE model architecture optimization, not LLM benchmarking.
Spectral-MoE: Heterogeneous Per-Layer Expert Utilization in Dense-to-MoE Conversion The paper introduces a method to convert dense models into mixture-of-experts by dynamically allocating different numbers of active experts across layers. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on inference acceleration and MoE conversion, not safety.
From Detection To Anticipation: Continuous Onset Forecasting For Stuttered Speech The paper develops a continuous forecasting method using speech representations to anticipate stuttering events shortly before they occur. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Applies speech models to stuttering, unrelated to AI deployment safety.
From Detection To Anticipation: Continuous Onset Forecasting For Stuttered Speech The paper develops a continuous forecasting method using speech representations to anticipate stuttering events shortly before they occur. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Studies speech dysfluency forecasting, not LLM benchmark evaluation.
From Detection To Anticipation: Continuous Onset Forecasting For Stuttered Speech The paper develops a continuous forecasting method using speech representations to anticipate stuttering events shortly before they occur. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on speech onset forecasting rather than LLM efficiency.
LLMs program using Reusable and Composable Abstract Representations The paper demonstrates that language models use modular abstract representations to generalize code generation across different programming languages and strategies. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Examines internal representations of code, not safety or guardrails.
LLMs program using Reusable and Composable Abstract Representations The paper demonstrates that language models use modular abstract representations to generalize code generation across different programming languages and strategies. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Studies activation geometry in code LLMs, not model compression.
LLMs program using Reusable and Composable Abstract Representations The paper demonstrates that language models use modular abstract representations to generalize code generation across different programming languages and strategies. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on mechanistic representation analysis rather than LLM benchmarking methods.
FinGREAT: Evaluating LLMs on Financial Graph-guided Retrieval And Typing The paper benchmarks language models on extracting and categorizing double-entry accounting relationships from corporate financial disclosure filings. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on benchmarking financial extraction rather than safety or guardrails.
FinGREAT: Evaluating LLMs on Financial Graph-guided Retrieval And Typing The paper benchmarks language models on extracting and categorizing double-entry accounting relationships from corporate financial disclosure filings. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Introduces a financial benchmark rather than model compression techniques.
Multi-Sensitive Distributional Fairness: Joint Wasserstein Audits and Shapley Allocation The paper presents a framework using Wasserstein geometry and Shapley values to audit and decompose intersectional fairness disparities across multiple attributes. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on statistical fairness metrics rather than LLM evaluation benchmarks.
Multi-Sensitive Distributional Fairness: Joint Wasserstein Audits and Shapley Allocation The paper presents a framework using Wasserstein geometry and Shapley values to audit and decompose intersectional fairness disparities across multiple attributes. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Theoretical fairness decomposition, unrelated to efficient model inference.
Beyond Time Induction: Structured Conditional Neural Solvers for Parametric Stochastic-Volatility Pricing The paper introduces a neural solver architecture to accurately compute solutions and sensitivities for parametric stochastic volatility partial differential equations. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Deals with financial PDE solving rather than AI deployment safety.
Beyond Time Induction: Structured Conditional Neural Solvers for Parametric Stochastic-Volatility Pricing The paper introduces a neural solver architecture to accurately compute solutions and sensitivities for parametric stochastic volatility partial differential equations. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on neural PDE solvers rather than LLM efficiency.
Beyond Time Induction: Structured Conditional Neural Solvers for Parametric Stochastic-Volatility Pricing The paper introduces a neural solver architecture to accurately compute solutions and sensitivities for parametric stochastic volatility partial differential equations. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Applies neural solvers to stochastic PDEs, not LLM evaluation.
Outcome-Routed Distillation and Reinforcement Learning for Diffusion Language Models The paper presents an outcome-routed distillation and reinforcement learning method tailored for training mixture-of-experts diffusion language models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Develops a distillation training pipeline, not an evaluation framework.
Outcome-Routed Distillation and Reinforcement Learning for Diffusion Language Models The paper presents an outcome-routed distillation and reinforcement learning method tailored for training mixture-of-experts diffusion language models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on training distillation algorithms rather than safety guardrails.
Exact Scaling Laws for a Random-Features Bigram Model The paper derives theoretical scaling laws for prediction regret in a capacity-constrained bigram model with power-law token frequencies. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Theoretical scaling work without AI safety or reliability focus.
Exact Scaling Laws for a Random-Features Bigram Model The paper derives theoretical scaling laws for prediction regret in a capacity-constrained bigram model with power-law token frequencies. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Theoretical derivation of scaling laws, not empirical benchmark evaluation.
Exact Scaling Laws for a Random-Features Bigram Model The paper derives theoretical scaling laws for prediction regret in a capacity-constrained bigram model with power-law token frequencies. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Analyzes bigram capacity limits rather than small model efficiency techniques.
Verifier-Grounded Agentic RL for Provably Correct Code The paper introduces an agentic reinforcement learning approach that uses formal program verifier diagnostics as intermediate rewards to produce verified code. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Does not address model compression, pruning, or efficient inference.
Verifier-Grounded Agentic RL for Provably Correct Code The paper introduces an agentic reinforcement learning approach that uses formal program verifier diagnostics as intermediate rewards to produce verified code. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes an RL training environment rather than LLM benchmarking methods.
Historical-Feature Residual Adaptation for Sequential Screening The paper proposes an acquisition-time residual adaptation method that uses historical features to improve hit rates in sequential screening. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Deals with sequential screening policies, not LLM benchmark evaluation.
Historical-Feature Residual Adaptation for Sequential Screening The paper proposes an acquisition-time residual adaptation method that uses historical features to improve hit rates in sequential screening. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Does not investigate model compression or efficient model serving.
Historical-Feature Residual Adaptation for Sequential Screening The paper proposes an acquisition-time residual adaptation method that uses historical features to improve hit rates in sequential screening. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on screening policies rather than AI safety or reliability.
Why don't Graph Transformers attend to the same token? The paper examines why molecular graph transformers avoid attention sink patterns seen in language models and highlights attention design impacts. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Studies attention sink mechanisms rather than model compression or efficiency.
Why don't Graph Transformers attend to the same token? The paper examines why molecular graph transformers avoid attention sink patterns seen in language models and highlights attention design impacts. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Investigates attention mechanisms, not AI safety or deployment reliability.
Why don't Graph Transformers attend to the same token? The paper examines why molecular graph transformers avoid attention sink patterns seen in language models and highlights attention design impacts. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Analyzes internal attention mechanisms in graph transformers, not benchmark evaluation.
BFHD: Bidirectional Feature Harmonization Decomposition for Heterogeneous Clinical Assessments The paper introduces a framework to translate and decompose clinical measurement records across heterogeneous assessment systems within predefined error tolerances. Peer-reviewed	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on clinical measurement harmonization, not LLM evaluation.
BFHD: Bidirectional Feature Harmonization Decomposition for Heterogeneous Clinical Assessments The paper introduces a framework to translate and decompose clinical measurement records across heterogeneous assessment systems within predefined error tolerances. Peer-reviewed	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Does not focus on model compression or efficient inference.
BFHD: Bidirectional Feature Harmonization Decomposition for Heterogeneous Clinical Assessments The paper introduces a framework to translate and decompose clinical measurement records across heterogeneous assessment systems within predefined error tolerances. Peer-reviewed	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Addresses clinical tabular data harmonization, not LLM deployment safety.
Dynamic Graph World Modelling: Learning Future Dynamics from Temporal Context The paper presents a dynamic graph framework to jointly forecast evolving node and relation representations from historical graph contexts. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on graph modeling rather than small efficient architectures.
Dynamic Graph World Modelling: Learning Future Dynamics from Temporal Context The paper presents a dynamic graph framework to jointly forecast evolving node and relation representations from historical graph contexts. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Does not address AI safety, jailbreaks, or deployment guardrails.
Dynamic Graph World Modelling: Learning Future Dynamics from Temporal Context The paper presents a dynamic graph framework to jointly forecast evolving node and relation representations from historical graph contexts. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes a dynamic graph architecture, not LLM evaluation benchmarks.
One-Step Prediction Fidelity Is a Poor Proxy for Planning Performance in JEPA World Models The paper shows that one-step prediction accuracy in joint-embedding world models is an unreliable indicator of downstream planning performance. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Studies prediction fidelity vs planning metrics, not model efficiency.
One-Step Prediction Fidelity Is a Poor Proxy for Planning Performance in JEPA World Models The paper shows that one-step prediction accuracy in joint-embedding world models is an unreliable indicator of downstream planning performance. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Does not focus on safety, guardrails, or deployment reliability.
RA$^2$: Retain-Anchored Attraction Preserves Forget-Adjacent Utility in LLM Unlearning The paper proposes a latent-space unlearning method that preserves model performance on benign tasks conceptually adjacent to forgotten domains. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Does not address model efficiency, compression, or fast inference.
RA$^2$: Retain-Anchored Attraction Preserves Forget-Adjacent Utility in LLM Unlearning The paper proposes a latent-space unlearning method that preserves model performance on benign tasks conceptually adjacent to forgotten domains. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on machine unlearning methodology, not LLM evaluation frameworks or benchmarks.

6,001–6,060 of 17,313← Previous Next →