AI papers — 2026-08-17

The Morning Preprint — this day as one edition: the day's papers, the five things worth knowing, and the episodes behind them.

Today's papers

One Frozen Simulator Is Not Enough: Simulator Collapse in Multi-Agent RL. The paper addresses a critical failure mode in multi-agent reinforcement... [paper] [episode]
PEER: Unified Process-Outcome Reinforcement Learning for Structured Empathetic Reasoning. The paper introduces structured empathetic reasoning, a framework that... [paper] [episode]
Limits to scalable evaluation at the frontier: LLM as Judge won't beat twice the data. Limits to scalable evaluation at the frontier: LLM as Judge won't beat twice... [paper] [episode]
mR AG: Multimodal Retrieval-Reflection-Augmented Generation for Knowledge-Based VQA. The paper proposes a novel generalized framework called multimodal...
QA-Merging: Query-Adaptive Reasoning via Layer Selective Model Merging. The paper introduces Reasoning Pattern Alignment Merging (RPAM), a layer-wise...
Towards Understanding Linear Word Analogies. This paper provides a formal explanation for why word analogies can be solved... [paper] [episode]
DiffImaginE: Imagine to Verify Entity Types with Diffusion. DiffImaginE is a method for multimodal named entity recognition (MNER) that... [paper] [episode]
Improving Influence-based Instruction Tuning Data Selection for Balanced Learning of Diverse Capabilities. Selecting appropriate training data is crucial for effective instruction... [paper] [episode]
Thinking Outside the (Gray) Box: A Context-Based Score for Assessing Value and Originality in Neural Text Generation. Despite the increasing use of large language models for creative tasks, their... [paper] [episode]
Online Regularized Statistical Learning in Reproducing Kernel Hilbert Space With Non-Stationary Data. This paper investigates the convergence properties of a recursive regularized... [paper] [episode]
DiagnosisArena: Benchmarking Diagnostic Reasoning for Large Language Models. This paper introduces DiagnosisArena, a comprehensive and challenging benchmark... [paper] [episode]
Multi-Bin Batching for Increasing LLM Inference Throughput. This paper introduces Multi-Bin Batching, a novel control policy designed to... [paper] [episode]
MedMCP-Calc: Benchmarking LLMs for Realistic Medical Calculator Scenarios via MCP Integration. MedMCP-Calc is the first benchmark designed to evaluate large language models... [paper] [episode]
Robust Privacy: Inference-Stage Privacy through Certified Robustness. This paper introduces Robust Privacy (RP), an inference-stage privacy notion... [paper] [episode]
Doubly robust nearest neighbors in factor models. Problem Setting The paper addresses matrix completion with missing data under a... [paper] [episode]
jina-vlm: Small Multilingual Vision Language Model. The paper presents jina-vlm, a token-efficient 2.4B parameter vision-language... [paper] [episode]
A Generative Deep Learning Workflow for Inverse Molecular Design of Fuels. This paper presents a generative deep learning framework for the inverse design...
Decode-Branch Transformers: Decoupling the Primary Prefill Path from Additional Decode Computation. The paper introduces the Dual-Flow Transformer, an architecture that decouples... [paper] [episode]
DR.GAP: Mitigating Bias in Large Language Models using Gender-Aware Prompting with Decoupled Reasoning. The paper proposes DR.GAP (Demonstration and Reasoning for Gender-Aware...
Budget-Aware Tool Use Enables Effective Agent Scaling. The paper introduces a systematic study of budget-constrained tool-use agents... [paper] [episode]
Efficient Code Embeddings from Code Generation Models. The paper introduces jina-code-embeddings, a novel code embedding model suite... [paper] [episode]
PhyxMamba: Chaotic System Reconstruction from Short Context Observations with Generative State-Space Models. Problem Statement Long-term forecasting of chaotic systems remains a...
Learning Multi-Timescale Interventions under Safety and Resource Constraints. The paper introduces the Multi-Timescale Intervention Markov decision process... [paper] [episode]
A Physics-Informed Hybrid Neural Operator for Transient Magnetization Prediction in Power Magnetics. This paper proposes the Physics-Informed Hybrid Neural Operator (PI-HNO), a... [paper] [episode]
DirMixE: Harnessing Test Agnostic Long-tail Recognition with Hierarchical Label Variations. This paper explores test-agnostic long-tail recognition, a challenging...
SPOT: Sparse Probing and Outcome Calibration for On-Policy Distillation. The paper introduces SPOT (Sparse Probing and Outcome-calibrated Targets for... [paper] [episode]
HYGENE: A Diffusion-based Hypergraph Generation Method. This paper introduces HYGENE, a diffusion-based method for hypergraph... [paper] [episode]
Hybrid quantum recurrent neural network for remaining useful life prediction of turbofan engines. This paper introduces a Hybrid Quantum Recurrent Neural Network (HQRNN)... [paper] [episode]
Exact Attention Sensitivity and the Geometry of Transformer Stability. The paper develops a stability theory for transformers that explains why... [paper] [episode]
Geometric Self-Supervised Pre-training for Neural Combinatorial Optimization. This paper introduces a geometric self-supervised pre-training framework... [paper] [episode]
Bactrainus: Optimizing Large Language Models for Multi-hop Complex Question Answering Tasks. This paper introduces Bactrainus, a two-stage selector-reader architecture... [paper] [episode]
SMA: Who Said That? Auditing Membership Leakage in Semi-Black-box RAG Controlling. This paper introduces the Source-aware Membership Audit (SMA) framework, the... [paper] [episode]
LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning. The paper introduces the LLM-jury, a training-free verification method for... [paper] [episode]
On Fibonacci Ensembles: An Alternative Approach to Ensemble Learning Inspired by the Timeless Architecture of the Golden Ratio. The paper introduces Fibonacci Ensembles, a mathematically principled framework... [paper] [episode]
ContextClaim: A Context-Driven Paradigm for Verifiable Claim Detection. The paper proposes Context-Driven Claim Detection (ContextClaim), a paradigm... [paper] [episode]
Shorter, but Still Trustworthy? An Empirical Study of Chain-of-Thought Compression. This paper presents the first systematic empirical study of how... [paper] [episode]
BiAxisBias: Evaluating LLM Bias Beyond a Single Prompt and a Single Explanation. The paper introduces BiAxisAudit, a novel framework designed to evaluate bias...
One-shot Robust Federated Learning of Independent Component Analysis. This paper investigates a general robust one-shot aggregation framework for...
Train Yourself as an LLM: Exploring Effects of AI Literacy on Persuasion via Role-playing LLM Training. Train Yourself as an LLM: Exploring Effects of AI Literacy on Persuasion via... [paper] [episode]
MCBench: A Multicontext Safety Assessment Benchmark for Omni Large Language Models. The paper introduces MCBench, a new benchmark designed to evaluate the... [paper] [episode]
A Machine-Learned Comorbidity Index. The paper introduces a Machine-Learned Comorbidity Index (MLCI) that "maps... [paper] [episode]
Closing the Loop: A Control-Theoretic Framework for Provably Stable Time Series Forecasting with LLMs. Problem Statement The paper identifies a critical theoretical flaw in applying... [paper] [episode]
Differentiable Thermodynamic Phase-Equilibria for Machine Learning. The paper introduces DISCOMAX, a differentiable algorithm for phase-equilibrium... [paper] [episode]
Measuring the Prevalence of Policy Violating Content with ML Assisted Sampling and LLM Labeling. This paper presents a production system at Pinterest for measuring the daily... [paper] [episode]
Decentralized Federated Learning by Partial Message Exchange. The paper introduces PaME (DFL by Partial Message Exchange), a novel... [paper] [episode]
Speculative Rollback Correction for Quality-Diverse Web Agent Imitation. The paper introduces Speculative Rollback Correction (SRC), a branch-level... [paper] [episode]
Free-Flow Class-Incremental Learning: Towards Robust CIL under Variable Class Arrivals. This paper introduces and formalizes a new, more realistic setting for...
A Theoretical Framework for Statistical Evaluability of Generative Models. A Theoretical Framework for Statistical Evaluability of Generative Models... [paper] [episode]
From Uniform to Learned Knots: A Study of Spline-Based Numerical Encodings for Tabular Deep Learning. The paper "From Uniform to Learned Knots: A Study of Spline-Based Numerical... [paper] [episode]
The Optimal Sample Complexity of Multiclass and List Learning. This paper resolves a longstanding open conjecture in learning theory, thereby... [paper] [episode]
Learning Koopman operators for coupled systems via information on governing equations of subsystems. This paper proposes a method to learn the Koopman operator for coupled... [paper] [episode]
From Lab to Reality: A Practical Evaluation of Deep Learning Models and LLMs for Vulnerability Detection. This paper presents a systematic, deployment-oriented evaluation of deep... [paper] [episode]
Multilook Coherent Imaging: Theoretical Guarantees and Algorithms. Multilook Coherent Imaging: Theoretical Guarantees and Algorithms Problem... [paper] [episode]
Spectral Certificates and Projection-DPP Rounding for Determinantal MAP Selection. Spectral DPPs via NEPv: A Scalable Continuous Relaxation of Determinantal MAP...
Feature-Aware (Hyper)graph Generation via Next-Scale Prediction. The paper introduces FAHNES (feature-aware (hyper)graph generation via... [paper] [episode]
Hierarchical Copula-Gumbel-Top-K Routing: Two-Sided Dependence Control for Frozen Mixture-of-Experts at Fixed Per-Token Routing Laws. This paper introduces Hierarchical Copula-Gumbel-Top-K (H-CGA), a method for... [paper] [episode]
Neurai-VN Benchmark: Standardized Machine Learning Models for Multimodal Digital Phenotyping in Mental Health Classification. The paper introduces a reproducible benchmark for mental health classification... [paper] [episode]
RT-Lynx: Putting GEMM Sparsity in the Right Place for Diffusion Models. The paper "RT-Lynx: Putting GEMM Sparsity in the Right Place for Diffusion...
jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers. The paper introduces GELATO (Geometry-preserving Embeddings via Locked Aligned... [paper] [episode]
Delay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVR. This paper investigates the impact of systematic verification errors on... [paper] [episode]
Annealed Softmax Greedy in Many-Armed Bayesian Bandits. This paper investigates whether annealed softmax (Boltzmann) policies, which... [paper] [episode]
Red-Teaming the Agentic Red-Team. This paper presents the first in-depth security analysis of agentic systems... [paper] [episode]
ROC-n-reroll: How verifier imperfection affects test-time scaling. This paper theoretically and empirically analyzes how imperfections in... [paper] [episode]
HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam. The paper introduces HLE-Verified, a verified and revised version of the... [paper] [episode]
PolyWorkBench: Benchmarking LLM Agents for Cross-Lingual Long-Horizon Workflows. The paper introduces PolyWorkBench, a benchmark for evaluating LLM agents on...
Multimodal Language Models Benchmarked Against the NRC Reactor Operator Licensing Examination: Fine-Tuning and Retrieval Strategies. This paper presents a systematic benchmark of a 31-billion-parameter...
Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models. Vision-language models (VLMs) excel at many tasks, yet continue to struggle... [paper] [episode]
Evaluating the trustworthiness of the Fréchet Inception Distance with stochastic embedding representations. This paper investigates whether uncertainty quantification (UQ) techniques,...
PRO-Bid: Pareto-Prioritized Regret Optimization for Constraint-Aware Generative Auto-Bidding. This paper introduces PRO-Bid, a constraint-aware generative auto-bidding...
In-Context Source and Channel Coding. The paper, titled "In-Context Source and Channel Coding," proposes a... [paper] [episode]
Not All Neighbors Matter: Understanding the Impact of Graph Sparsification on GNN Pipelines. The paper addresses the fundamental question of "how much of the graph... [paper] [episode]
MoRFI: Monotonic Sparse Autoencoder Feature Identification. MoRFI: Monotonic Sparse Autoencoder Feature Identification Abstract Large... [paper] [episode]
Negative Ontology of True Target for Machine Learning: Towards Recognition, Evaluation and Learning under Democratic Supervision. This paper philosophically examines how shifts in assumptions regarding the... [paper] [episode]
A Quantitative Approximation Framework for Flow Distillation in Diffusion Models. This paper develops a quantitative approximation framework for diffusion... [paper] [episode]
Shorthand for Thought: Compressing LLM Reasoning via Entropy-Guided Supertokens. The paper "Shorthand for Thought: Compressing LLM Reasoning via Entropy-Guided... [paper] [episode]
SymbolicLight V1: Spike-Gated Dual-Path Language Modeling at High Activation Sparsity. and Core Contribution The paper presents SymbolicLight V1, "a spike-gated...

The papers

One Frozen Simulator Is Not Enough: Simulator Collapse in Multi-Agent RL — The paper addresses a critical failure mode in multi-agent reinforcement learning for human-AI interaction. Multi-agent reinforcement learning for human-AI interaction typically relies on a single large language model to simulate user behavior. [episode]
LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning — The paper introduces the LLM-jury, a training-free verification method for selecting correct answers from a pool of candidate reasoning chains in large language models. [episode]
Decentralized Federated Learning by Partial Message Exchange — The paper introduces PaME (DFL by Partial Message Exchange), a novel decentralized federated learning (DFL) algorithm designed to improve the trade-off among communication efficiency, privacy preservation, and model accuracy. [episode]
HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam — Summary The paper introduces HLE-Verified, a verified and revised version of the Humanity’s Last Exam (HLE) benchmark, constructed via a transparent, component-wise verification protocol and a fine-grained error taxonomy. [episode]
Constraint-Aware Generative Auto-bidding via Pareto-Prioritized Regret Optimization — Summary This paper introduces PRO-Bid, a constraint-aware generative auto-bidding framework designed to address two fundamental limitations of existing Decision Transformer (DT) approaches in the context of hard ratio constraints in online advertising. [episode]
BiAxisAudit: A Novel Framework to Evaluate LLM Bias Across Prompt Sensitivity and Response-Layer Divergence — Summary The paper introduces BiAxisAudit, a novel framework designed to evaluate bias in large language models (LLMs) by addressing two structural blind spots in existing bias benchmarks: prompt sensitivity and response-layer divergence.
MoRFI: Monotonic Sparse Autoencoder Feature Identification — MoRFI: Monotonic Sparse Autoencoder Feature Identification Abstract Large language models (LLMs) acquire most of their factual knowledge during the pre-training stage, through next token prediction. [episode]
RT-Lynx: Putting the GEMM Sparsity In a Right Way for Diffusion Models — Summary The paper "RT-Lynx: Putting GEMM Sparsity in the Right Place for Diffusion Models" addresses the inference cost bottleneck of Diffusion Transformers (DiT) in image generation. [episode]
Shorter, but Still Trustworthy? An Empirical Study of Chain-of-Thought Compression — Summary This paper presents the first systematic empirical study of how chain-of-thought (CoT) compression affects model trustworthiness, moving beyond the standard evaluation focus on task accuracy and token savings. [episode]
Red-Teaming the Agentic Red-Team — Summary This paper presents the first in-depth security analysis of agentic systems designed for offensive security operations, termed "agentic-red-teams." The authors analyze 12 open-source tools of this class and demonstrate that they introduce new attack vectors against the or [episode]
Multilook Coherent Imaging: Theoretical Guarantees and Algorithms — The paper addresses the problem of signal recovery from multilook coherent imaging systems in the presence of speckle noise. [episode]
In-Context Source and Channel Coding — Summary The paper, titled "In-Context Source and Channel Coding," proposes a receiver-side framework called In-Context Decoding (ICD) to enhance the robustness of Separate Source–Channel Coding (SSCC) systems, particularly in low Signal-to-Noise Ratio (SNR) regimes. [episode]
Shorthand for Thought: Compressing LLM Reasoning via Entropy-Guided Supertokens — The paper "Shorthand for Thought: Compressing LLM Reasoning via Entropy-Guided Supertokens" proposes a vocabulary-level approach to compressing chain-of-thought (CoT) reasoning traces in Large Language Models. [episode]
Closing the Loop: A Control-Theoretic Framework for Provably Stable Time Series Forecasting with LLMs — The paper identifies a critical theoretical flaw in applying Large Language Models (LLMs) to time series forecasting. [episode]
Annealed Softmax Greedy in Many-Armed Bayesian Bandits — Summary This paper investigates whether annealed softmax (Boltzmann) policies, which are agnostic to epistemic uncertainty, can achieve near-optimal Bayes regret in the many-armed Bayesian Bernoulli bandit setting. [episode]
Feature-Aware (Hyper)graph Generation via Next-Scale Prediction — The paper introduces FAHNES (feature-aware (hyper)graph generation via next-scale prediction), a hierarchical generative framework that jointly models topology and features for both graphs and hypergraphs. [episode]
Not All Neighbors Matter: Understanding the Impact of Graph Sparsification on GNN Pipelines — The paper addresses the fundamental question of "how much of the graph structure is actually necessary for effective learning" in Graph Neural Networks (GNNs). [episode]
Delay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVR — Summary This paper investigates the impact of systematic verification errors on Reinforcement Learning with Verifiable Rewards (RLVR), a technique used to improve the reasoning capabilities of large language models (LLMs). [episode]
Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models — Vision-language models (VLMs) excel at many tasks, yet continue to struggle with spatial reasoning—problems where the key information is not directly observable in the input. [episode]
Neurai-VN Benchmark: Standardized Machine Learning Models for Multimodal Digital Phenotyping in Mental Health Classification — Summary The paper introduces a reproducible benchmark for mental health classification using digital phenotyping (DP) data, built upon the NEURAI-VN dataset. [episode]
A Quantitative Approximation Framework for Flow Distillation in Diffusion Models — Summary This paper develops a quantitative approximation framework for diffusion distillation, viewing few-step sampling as error propagation under compositions of learned flow maps. [episode]
Speculative Rollback Correction for Quality-Diverse Web Agent Imitation — Summary The paper introduces Speculative Rollback Correction (SRC), a branch-level imitation learning framework designed for training interactive web and GUI agents in resettable environments. [episode]
jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers — The paper introduces GELATO (Geometry-preserving Embeddings via Locked Aligned TOwers), a novel approach to constructing multimodal embedding models. [episode]
PolyWorkBench: Benchmarking Multilingual Long-Horizon LLM Agents — The paper introduces PolyWorkBench, a benchmark for evaluating LLM agents on multilingual long-horizon workplace workflows. [episode]
Measuring the Prevalence of Policy Violating Content with ML Assisted Sampling and LLM Labeling — Summary This paper presents a production system at Pinterest for measuring the daily prevalence of policy-violating content on the platform. [episode]
Federated Independent Component Analysis via Spectral Alignment and Robust Aggregation — This paper investigates a general robust one-shot aggregation framework for distributed and federated Independent Component Analysis (ICA) problem.
On Fibonacci Ensembles: An Alternative Approach to Ensemble Learning Inspired by the Timeless Architecture of the Golden Ratio — The paper introduces Fibonacci Ensembles, a mathematically principled framework for ensemble learning inspired by the Fibonacci sequence and the golden ratio. [episode]
MCBench: A Multicontext Safety Assessment Benchmark for Omni Large Language Models — Summary The paper introduces MCBench, a new benchmark designed to evaluate the safety-awareness of Omni Large Language Models (LLMs) that process vision, audio, and text simultaneously. [episode]
Learning Koopman operators for coupled systems via information on governing equations of subsystems — This paper proposes a method to learn the Koopman operator for coupled nonlinear dynamical systems by leveraging prior knowledge of the differential equations governing each subsystem, combined with online Extended Dynamic Mode Decomposition (EDMD). [episode]
ContextClaim: A Context-Driven Paradigm for Verifiable Claim Detection — The paper proposes Context-Driven Claim Detection (ContextClaim), a paradigm that augments verifiable claim detection with retrieved background context. [episode]
From Lab to Reality: A Practical Evaluation of Deep Learning Models and LLMs for Vulnerability Detection — Summary This paper presents a systematic, deployment-oriented evaluation of deep learning (DL) models and large language models (LLMs) for software vulnerability detection, aiming to bridge the gap between academic benchmark performance and real-world applicability. [episode]
SymbolicLight V1: Spike-Gated Dual-Path Language Modeling at High Encoder Spike Sparsity — The paper presents SymbolicLight V1, "a spike-gated dual-path language model that combines binary Leaky Integrate-and-Fire (LIF) spike dynamics with a continuous residual stream." The architecture "replaces dense self-attention with Dual-Path SparseTCAM: an exponential-decay aggr [episode]
From Uniform to Learned Knots: A Study of Spline-Based Numerical Encodings for Tabular Deep Learning — The paper "From Uniform to Learned Knots: A Study of Spline-Based Numerical Encodings for Tabular Deep Learning" by Manish Kumar, Anton Frederik Thielmann, Christoph Weisser, and Benjamin Säfken investigates the role of numerical preprocessing in tabular deep learning, with a pa [episode]
Differentiable Thermodynamic Phase-Equilibria for Machine Learning — Summary The paper introduces DISCOMAX, a differentiable algorithm for phase-equilibrium calculation that guarantees thermodynamic consistency at both training and inference, only subject to a user-specified discretization. [episode]
ROC-n-reroll: How verifier imperfection affects test-time scaling — This paper theoretically and empirically analyzes how imperfections in verifiers affect the performance of two test-time scaling methods: Rejection Sampling (RS) and Best-of-N (BoN). [episode]
Investigation into using stochastic embedding representations for evaluating the trustworthiness of the Fr' e chet Inception Distance — This paper investigates whether uncertainty quantification (UQ) techniques, specifically Monte Carlo dropout (MCD), can be used to evaluate the trustworthiness of the Fréchet Inception Distance (FID) when applied to out-of-distribution (OOD) data, particularly medical images. [episode]
The Optimal Sample Complexity of Multiclass and List Learning — Summary This paper resolves a longstanding open conjecture in learning theory, thereby determining the optimal sample complexity of multiclass and list learning. Main Result The central contribution is a positive resolution of a conjecture by Daniely and Shalev-Shwartz (2014). [episode]
A Machine-Learned Comorbidity Index — The paper introduces a Machine-Learned Comorbidity Index (MLCI) that "maps diagnosis codes to a single scalar by maximizing the normalized Hilbert–Schmidt Independence Criterion (nHSIC) between the learned score and multiple clinical outcomes." The authors state that "MLCI capt [episode]
Spectral DPPs via NEPv: A Scalable Continuous Relaxation of Determinantal MAP for Diversity-Aware Data Selection — Authors: Richard Yi Da Xu (Hong Kong Baptist University, TadReamk Limited) arXiv: 2606.19411v2 [cs.LG], 22 Jul 2026 --- The paper addresses the computational intractability of Determinantal Point Process (DPP) MAP inference for large-scale data selection tasks. [episode]
Towards Realistic Class-Incremental Learning with Free-Flow Increments — Summary This paper introduces and formalizes a new, more realistic setting for Class-Incremental Learning (CIL) termed Free-Flow Class-Incremental Learning (FFCIL). [episode]
A Theoretical Framework for Statistical Evaluability of Generative Models — Authors: Shashaank Aiyer, Yishay Mansour, Shay Moran, and Han Shao arXiv: 2604.05324v2 [cs.LG], June 2026 --- The paper addresses a fundamental question in generative model evaluation: "Can the performance of generative models be evaluated statistically from finite samples?" The [episode]
Benchmarking Fine-tuning and Retrieval Strategies for a Multimodal Language Model on the NRC Reactor Operator Licensing Examination — This paper presents a systematic benchmark of a 31-billion-parameter open-weight multimodal model (Gemma 4 31B-IT) on the U.S. Nuclear Regulatory Commission (NRC) Reactor Operator licensing examination. [episode]
Train Yourself as an LLM: Exploring Effects of AI Literacy on Persuasion via Role-playing LLM Training — Authors: Qihui Fan, Min Ge, Chenyan Jia, Weiyan Shi (Northeastern University) Publication: arXiv:2604.02637v1 [cs.CL], April 3, 2026 --- As large language models (LLMs) become increasingly persuasive, there is concern that people's opinions and decisions may be influenced across [episode]
Limits to scalable evaluation at the frontier: LLM as Judge won't beat twice the data — Authors: Florian E. Dorner, Vivian Y. [episode]
PEER: Unified Process-Outcome Reinforcement Learning for Structured Empathetic Reasoning — Summary The paper introduces structured empathetic reasoning, a framework that decomposes emotional support into three steps prior to generating a final reply: "1) Conversation history analysis to identify the seeker’s issue and the current support stage, 2) Multimodal emotiona [episode]
Reasoning Pattern Alignment Merging for Adaptive Reasoning — The paper introduces Reasoning Pattern Alignment Merging (RPAM), a layer-wise model merging framework designed to achieve query-adaptive reasoning by combining a Long-CoT reasoning model with a Short-CoT instruction model. [episode]
Towards Understanding Linear Word Analogies — This paper provides a formal explanation for why word analogies can be solved using vector arithmetic in non-linear embedding models such as skip-gram with negative sampling (SGNS) and GloVe, without making the strong assumptions of prior theories. [episode]
mR squared AG: Multimodal Retrieval-Reflection-Augmented Generation for Knowledge-Based VQA — The paper proposes a novel generalized framework called multimodal Retrieval-Reflection-Augmented Generation (mR2 AG) to address challenges in Knowledge-based Visual Question Answering (VQA) tasks. [episode]
DiffImaginE: Imagine to Verify Entity Types with Diffusion — Summary DiffImaginE is a method for multimodal named entity recognition (MNER) that reformulates type verification as conditional latent diffusion inference. The paper states: "DiffImaginE, which formulates MNER type verification as conditional latent diffusion inference. [episode]
Improving Influence-based Instruction Tuning Data Selection for Balanced Learning of Diverse Capabilities — Selecting appropriate training data is crucial for effective instruction fine-tuning of large language models (LLMs), which aims to (1) elicit strong capabilities, and (2) achieve balanced performance across a diverse range of tasks. [episode]
Online Regularized Statistical Learning in Reproducing Kernel Hilbert Space With Non-Stationary Data — Summary This paper investigates the convergence properties of a recursive regularized learning algorithm within a reproducing kernel Hilbert space (RKHS) when the online data streams are non-stationary, meaning they are neither independent nor identically distributed. [episode]
Thinking Outside the (Gray) Box: A Context-Based Score for Assessing Value and Originality in Neural Text Generation — Despite the increasing use of large language models for creative tasks, their outputs often lack diversity. Common solutions, such as sampling at higher temperatures, can compromise the quality of the results. [episode]
DiagnosisArena: Benchmarking Diagnostic Reasoning for Large Language Models — Summary This paper introduces DiagnosisArena, a comprehensive and challenging benchmark designed to rigorously assess the professional-level diagnostic competence of large language models (LLMs) in complex clinical scenarios. [episode]
Multi-Bin Batching for Increasing LLM Inference Throughput — Summary This paper introduces Multi-Bin Batching, a novel control policy designed to increase the throughput of Large Language Model (LLM) inference systems. [episode]
Generative Deep Learning Framework for Inverse Design of Fuels — Summary This paper presents a generative deep learning framework for the inverse design of fuels, combining a Co-optimized Variational Autoencoder (Co-VAE) with quantitative structure-property relationship (QSPR) techniques to enable accelerated discovery of fuel molecules with h [episode]
jina-vlm: Small Multilingual Vision Language Model — Summary The paper presents jina-vlm, a token-efficient 2.4B parameter vision-language model (VLM) that achieves state-of-the-art multilingual VQA performance among open 2B-scale VLMs. [episode]
Doubly robust nearest neighbors in factor models — The paper addresses matrix completion with missing data under a latent factor model. [episode]
MedMCP-Calc: Benchmarking LLMs for Realistic Medical Calculator Scenarios via MCP Integration — MedMCP-Calc is the first benchmark designed to evaluate large language models (LLMs) on realistic medical calculator workflows through Model Context Protocol (MCP) integration. [episode]
Decode-Branch Transformers: Decoupling the Primary Prefill Path from Additional Decode Computation — The paper introduces the Dual-Flow Transformer, an architecture that decouples prefill and decode computation in large language models. [episode]
Robust Privacy: Inference-Stage Privacy through Certified Robustness — Summary This paper introduces Robust Privacy (RP), an inference-stage privacy notion inspired by certified robustness, and its attribute-level projection, Robust Attribute Privacy (RAP). [episode]
DR.GAP: Mitigating Bias in Large Language Models using Gender-Aware Prompting with Demonstration and Reasoning — The paper proposes DR.GAP (Demonstration and Reasoning for Gender-Aware Prompting), an automated and model-agnostic approach to mitigate gender bias in Large Language Models (LLMs) while preserving model performance. [episode]
Budget-Aware Tool-Use Enables Effective Agent Scaling — The paper introduces a systematic study of budget-constrained tool-use agents and their test-time scaling behaviors, focusing on web search agents. The authors identify a critical bottleneck: "standard agents lack inherent budget awareness. [episode]
Negative Ontology of True Target for Machine Learning: Towards Recognition, Evaluation and Learning under Democratic Supervision — Future work under Negative Ontology as an Expansion Principle emphasizes empirical validation, integration with theoretical tools, and applied system development. [episode]
Hierarchical Copula-Gumbel-Top- K Routing: Two-Sided Dependence Control for Frozen Mixture-of-Experts at Fixed Per-Token Routing Laws — Summary This paper introduces Hierarchical Copula-Gumbel-Top-K (H-CGA), a method for controlling the *joint* distribution of routing choices across tokens in a frozen mixture-of-experts (MoE) model, while holding each individual token’s *marginal* routing law exactly fixed. [episode]
Efficient Code Embeddings from Code Generation Models — The paper introduces jina-code-embeddings, a novel code embedding model suite designed to retrieve code from natural language queries, perform technical question-answering, and identify semantically similar code snippets across programming languages. [episode]
Learning Multi-Timescale Interventions under Safety and Resource Constraints — Summary The paper introduces the Multi-Timescale Intervention Markov decision process (MTI-MDP) and a corresponding reinforcement learning framework called MINT (Multi-timescale Intervention Network Training) to address sequential decision-making problems where actions have effec [episode]
Mamba Integrated with Physics Principles Masters Long-term Chaotic System Forecasting — Long-term forecasting of chaotic systems remains a fundamental challenge due to the intrinsic sensitivity to initial conditions and the complex geometry of strange attractors. [episode]
A Physics-Informed Hybrid Neural Operator for Transient Magnetization Prediction in Power Magnetics — Summary This paper proposes the Physics-Informed Hybrid Neural Operator (PI-HNO), a compact, material-specific neural model designed for core-loss-oriented transient magnetization prediction in power magnetics. [episode]
DirMixE: Harnessing Test Agnostic Long-tail Recognition with Hierarchical Label Vartiations — This paper explores test-agnostic long-tail recognition, a challenging long-tail task where the test label distributions are unknown and arbitrarily imbalanced. [episode]
HYGENE: A Diffusion-based Hypergraph Generation Method — Summary This paper introduces HYGENE, a diffusion-based method for hypergraph generation, which the authors claim is "the first attempt to employ diffusion models for hypergraph generation." The core problem addressed is the difficulty of generating realistic and diverse hypergra [episode]
SPOT: Sparse Probing and Outcome Calibration for On-Policy Distillation — Summary The paper introduces SPOT (Sparse Probing and Outcome-calibrated Targets for On-Policy Distillation), a method for improving on-policy distillation (OPD) of large language models. [episode]
Hybrid quantum recurrent neural network for remaining useful life prediction — Summary This paper introduces a Hybrid Quantum Recurrent Neural Network (HQRNN) framework for predicting the Remaining Useful Life (RUL) of jet engines, evaluated on the NASA Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset, specifically the "FD001" subset. [episode]
Exact Attention Sensitivity and the Geometry of Transformer Stability — The paper develops a stability theory for transformers that explains why pre-LayerNorm works, why DeepNorm uses N−1/4 scaling, and why warmup is necessary, all from first principles. [episode]
Geometric Self-Supervised Pre-training for Neural Combinatorial Optimization — Summary This paper introduces a geometric self-supervised pre-training framework designed to improve the generalization capabilities of Neural Combinatorial Optimization (NCO) models, specifically for routing problems such as the Traveling Salesman Problem (TSP). [episode]
Bactrainus: Optimizing Large Language Models for Multi-hop Complex Question Answering Tasks — Summary This paper introduces Bactrainus, a two-stage selector-reader architecture designed to optimize large language models (LLMs) for multi-hop question answering (MHQA) tasks, evaluated on the HotpotQA dataset under the distractor setting. [episode]
SMA: Who Said That? Auditing Membership Leakage in Semi-Black-box RAG Controlling — Summary This paper introduces the Source-aware Membership Audit (SMA) framework, the first membership inference method designed for fine-grained source attribution in Retrieval-Augmented Generation (RAG) and Multimodal Retrieval-Augmented Generation (MRAG) systems. [episode]