Ax Bole Ma, Jan Eitzinger, Harald K\"ostler 5/8/2026

Irminsul: MLA-Native Position-Independent Caching for Agentic LLM Serving

Position-independent caching system (Irminsul) for agentic LLM serving that prevents cache invalidation from token shifts during multi-turn interactions.

Ax Huchen Yang, Xinghao Dong, Dan Negrut, Jin-Long Wu 5/8/2026

Active Learning for Communication Structure Optimization in LLM-Based Multi-Agent Systems

Active learning method for optimizing communication structures in LLM multi-agent systems to reduce token usage and improve performance.

Ax Anh H. Vo, Sungyo Lee, Phil-Joong Kim, Soo-Mi Choi, Yong-Guk Kim 5/8/2026

Closing the Loop: Unified 3D Scene Generation and Immersive Interaction via LLM-RL Coupling

Framework combining LLMs with reinforcement learning for unified 3D scene generation and interactive user interaction in multimedia systems.

Ax Ming Liu 5/8/2026

Decodable but Not Corrected by Fixed Residual-Stream Linear Steering: Evidence from Medical LLM Failure Regimes

Research on linear decodability versus correction of medical LLM failure modes, showing limitations of fixed residual-stream steering.

Ax Joseph D. Ramsey 5/8/2026

Fourier Feature Methods for Nonlinear Causal Discovery: FFML Scoring and FFCI Testing in Mixed Data

Research on Fourier feature methods for nonlinear causal discovery in mixed data combining scoring and constraint-based approaches.

Ax Dongjin Kim, Daniel S. King, Yoonjae Park, Roya Savoj, Sebastien Hamel, Xiaoyu Wang, Bingqing Cheng 5/8/2026

Polarizable atomic multipoles for learning long-range electrostatics

Research on machine learning interatomic potentials with polarizable atomic multipoles for modeling long-range electrostatics.

Ax Haodong Liang, Lifeng Lai 5/8/2026

Transformers Provably Implement In-Context Reinforcement Learning with Policy Improvement

Theoretical work proving transformers can implement in-context reinforcement learning with policy improvement via explicit constructions.

Ax Yuqian Cheng, Zhuo Chen, Qian Lin 5/8/2026

Optimal Confidence Band for Kernel Gradient Flow Estimator

Research on supremum-norm generalization error and uniform inference bounds for kernel gradient flow methods.

Ax Lena Helgerth, Andreas Christmann 5/8/2026

Ratio-based Loss Functions

Survey of ratio-based loss functions for supervised and unsupervised learning algorithms.

Ax Hirofumi Ota, Naoto Iwase, Yuki Ichihara, Junpei Komiyama, Masaaki Imaizumi 5/8/2026

CITE: Anytime-Valid Statistical Inference in LLM Self-Consistency

Research on anytime-valid statistical inference for controlling error in LLM self-consistency aggregation methods.

Ax Filip Edstr\"om, Guilherme W. F. Barros, Tetiana Gorbach, Xavier de Luna 5/8/2026

Tuning Derivatives for Causal Fairness in Machine Learning

Research on causal fairness methods for reducing bias in AI systems while respecting mediating variables.

Ax Sankarshana Venugopal (Seoul National University), Mohammad Mostafavi (Seoul National University), Jonghyun Choi (Seoul National University) 5/8/2026

DBMSolver: A Training-free Diffusion Bridge Sampler for High-Quality Image-to-Image Translation

Training-free diffusion sampler for image-to-image translation reducing function evaluations via exponential integrators.

Ax Bogdan Alexandru Bercean, Florinel Alin Croitoru, Vlad Hondru, Ciprian Mihai Ceausescu, Andreea Iuliana Ionescu, Radu Tudor Ionescu 5/8/2026

MTL-MAD: Multi-Task Learners are Effective Medical Anomaly Detectors

Research using multi-task learning with mixture-of-experts for medical image anomaly detection from self-supervised and pseudo-labeling tasks.

Ax Zehao Jin, Ruixuan Deng, Junran Wang, Xinjie Shen, Chao Zhang 5/8/2026

Beyond Steering Vector: Flow-based Activation Steering for Inference-Time Intervention

Research on flow-based activation steering for inference-time control of language model behavior without parameter updates.

Ax Millend Roy, Agostino Capponi, Vineet Goyal 5/8/2026

PREFER: Personalized Review Summarization with Online Preference Learning

Research on personalized review summarization systems that adapt to individual user preferences in e-commerce using online learning.

Ax Borja Aizpurua, Sukhbinder Singh, Augustine Kshetrimayum, Saeed S. Jahromi, Roman Orus 5/8/2026

Quantum-enhanced Large Language Models on Quantum Hardware via Cayley Unitary Adapters

Research on integrating quantum computing with large language models using Cayley unitary adapter parameters on quantum hardware.

Ax Michael Poppel, David Bucher, Maximilian Zorn, Markus Baumann, Sebastian W\"olckert, Claudia Linnhoff-Popien, Philipp Altmann, Jonas Stein 5/8/2026

Architecture Shape Governs QNN Trainability: Jacobian Null Space Growth and Parameter Efficiency

Research on quantum neural network trainability showing architecture shape affects parameter efficiency and gradient flow.

Ax Xinghao Wu, Jianwei Niu, Guogang Zhu, Xuefeng Liu, Shaojie Tang, Jiayuan Zhang 5/8/2026

From Coordinate Matching to Structural Alignment: Rethinking Prototype Alignment in Heterogeneous Federated Learning

Research on prototype alignment methods for heterogeneous federated learning with different data distributions and model architectures.

Ax Yang Xu, Jiefu Zhang, Haixiang Sun, Zihan Zhou, Tianyu Cao, Vaneet Aggarwal 5/8/2026

Towards Reliable LLM Evaluation: Correcting the Winner's Curse in Adaptive Benchmarking

SIREN protocol addresses selection bias in adaptive LLM evaluation benchmarking through selection-aware repeated-split reporting.

Ax Geping Chen, Chunlin Li, Tianzhong Yang, Zhengyuan Zhu, Jing Zhou 5/8/2026

TabCF: Distributional Control Function Estimation with Tabular Foundation Models

TabCF method uses tabular foundation models for control function estimation in causal inference with unmeasured confounding.

Ax Tianlun Hu, Tiancheng Hu, Shengsheng Litang, Sheng Wang, Xiaoming Bao, Yuxing Li, Wei Wang, Zhongzhe Hu, Lijun Li, Hongwei Sun, Jingbin Zhou\\ 5/8/2026

Relay Buffer Independent Communication over Pooled HBM for Efficient MoE Inference on Ascend

Optimization technique for MoE inference communication bottlenecks using pooled HBM relay buffer approach on Ascend hardware.

Ax Rappy Saha, Jude Haris, Nicolas Bohm Agostini, David Kaeli, Jos\'e Cano 5/8/2026

PoTAcc: A Pipeline for End-to-End Acceleration of Power-of-Two Quantized DNNs

Pipeline for end-to-end acceleration of power-of-two quantized DNNs on resource-constrained edge devices.

Ax Bowen Zheng, Weijian Luo, Guang Yang, Colin Zhang, Tianyang Hu 5/8/2026

Autoregressive Visual Generation Needs a Prologue

Prologue approach prepends learnable tokens to visual sequences to bridge reconstruction-generation gap in autoregressive image generation.

Ax Bowen Zheng, Yihong Luo, Tianyang Hu 5/8/2026

Learning Discrete Autoregressive Priors with Wasserstein Gradient Flow

Method for jointly training discrete image tokenizers and autoregressive priors using Wasserstein gradient flow.

Ax Kossi Amouzouvi, Robert Wardenga, Jens Lehmann, Sahar Vahdati 5/8/2026

Graphlets as Building Blocks for Structural Vocabulary in Knowledge Graph Foundation Models

Proposes graphlet-based structural vocabulary for knowledge graph foundation models to enable discrete symbolic representations.

Ax Suoxin Zhang, Run He, Di Fang, Xiang Tan, Kaixuan Chen, Huiping Zhuang 5/8/2026

Rethinking Adapter Placement: A Dominant Adaptation Module Perspective

PAGE method optimizes LoRA adapter placement in low-rank adaptation for efficient fine-tuning of pretrained models.

Ax Jonas Bayer, Stefan Zetzsche, Olivier Bouissou, Remi Delmas, Michael Tautschnig, Soonho Kong 5/8/2026

Teaching LLMs Program Semantics via Symbolic Execution Traces

Evaluation framework testing 14 LLM families on 500 C verification tasks, analyzing program semantics learning via symbolic execution traces.

Ax Paul Valery Nguezet, Elie Tagne Fute, Yusuf Brima, Benoit Martin Azanguezet, Marcellin Atemkeng 5/8/2026

Bridging visual saliency and large language models for explainable deep learning in medical imaging

Multimodal framework combining CNNs and LLMs to generate clinically interpretable diagnostic narratives for brain tumor classification.

Ax Bowen Zheng, Weijian Luo, Guang Yang, Colin Zhang, Tianyang Hu 5/8/2026

Taming the Entropy Cliff: Variable Codebook Size Quantization for Autoregressive Visual Generation

Research on variable codebook size quantization for discrete visual tokenizers to overcome information-theoretic limits in autoregressive image generation.

Ax Ajay Jaiswal, Lauren Hannah, Han-Byul Kim, Duc Hoang, Mehrdad Farajtabar, Minsik Cho 5/8/2026

TIDE: Every Layer Knows the Token Beneath the Context

TIDE revisits single-injection token embedding in LLMs; proposes per-layer token lookups to address rare token under-training and improve efficiency.

Ax Yijia Zheng, Marcel Worring 5/8/2026

LatentRAG: Latent Reasoning and Retrieval for Efficient Agentic RAG

LatentRAG improves agentic RAG by having LLM generate latent reasoning and subqueries in lower-dimensional space for efficient multi-step retrieval.

Ax Heegeon Yoon, Heeyoung Kim 5/8/2026

Multimodal Deep Generative Model for Semi-Supervised Learning under Class Imbalance

Multimodal deep generative model for semi-supervised learning handling class imbalance via addressing bias in pseudo-labels.

Ax Tom Kempton, Viktor Drobnyi, Maeve Madigan, Stuart Burrell 5/8/2026

Log-Likelihood, Simpson's Paradox, and the Detection of Machine-Generated Text

Demonstrates log-likelihood signals for detecting machine-generated text are non-uniform across hidden space, revealing Simpson's paradox in token-level analysis.

Ax Carles Balsells-Rodas, Zhengrui Xiang, Xavier Sumba, Yingzhen Li 5/8/2026

End-to-End Identifiable and Consistent Recurrent Switching Dynamical Systems

Identifiable and consistent learning of recurrent switching dynamical systems using non-variational estimators for sequential data with regime changes.

Ax Florian A. D. Burnat, Brittany I. Davidson 5/8/2026

Gaming the Metric, Not the Harm: Certifying Safety Audits against Strategic Platform Manipulation

Framework to certify safety audits against strategic platform manipulation via semantically equivalent content routing under online safety regulations.

Ax Florian A. D. Burnat, Brittany I. Davidson 5/8/2026

Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity

Paired-prompt protocol measures evaluation-context divergence in open-weight LLMs, revealing alignment-pipeline-specific behavioral differences.

Ax Shouvik Sardar, Sourish Das 5/8/2026

TinyBayes: Closed-Form Bayesian Inference via Jacobi Prior for Real-Time Image Classification on Edge Devices

TinyBayes applies closed-form Bayesian inference with Jacobi priors for real-time image classification on edge devices, tested on plant disease detection.

Ax Ashwani Anand, Ivi Chatzi, Ritam Raha, Anne-Kathrin Schmuck 5/8/2026

MANTRA: Synthesizing SMT-Validated Compliance Benchmarks for Tool-Using LLM Agents

MANTRA synthesizes SMT-validated compliance benchmarks for tool-using LLM agents from natural language procedural manuals to test agent behavior.

Ax Florian A. D. Burnat, Brittany I. Davidson 5/8/2026

A Benchmark for Strategic Auditee Gaming Under Continuous Compliance Monitoring

Benchmark for strategic gaming in continuous compliance audits under regulations like EU AI Act, formalizing delayed reporting and metric cherry-picking.

Ax Flavio Nicoletti, Chenxiao Ma, Enrico Ventura, Luca Saglietti, Stefano Sarao Mannelli 5/8/2026

The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models

Analyzes how class imbalance and data heterogeneity affect training dynamics and memorization in diffusion models.

Ax Paulo Mario P. Medina, Jose Marie Antonio Mi\~noza, Sebastian C. Iba\~nez 5/8/2026

eXplaining to Learn (eX2L): Regularization Using Contrastive Visual Explanation Pairs for Distribution Shifts

eX2L framework uses visual explanations as regularization to address distribution shifts and spurious correlations with improved interpretability.

Ax Leon Halgryn (University of Twente), Sophie Langer (Ruhr-Universit\"at Bochum), Janusz M. Meylahn (University of Twente), E. Moritz Hahn (University of Twente) 5/8/2026

Beyond the Independence Assumption: Finite-Sample Guarantees for Deep Q-Learning under $\tau$-Mixing

Finite-sample analysis of Deep Q-Learning under τ-mixing to handle temporal dependencies in replayed data instead of independence assumption.

Ax Philip Jordan, Maryam Kamgarpour 5/8/2026

Independent Learning of Nash Equilibria in Partially Observable Markov Potential Games with Decoupled Dynamics

Nash equilibrium learning in partially observable Markov games with decoupled dynamics for multi-agent reinforcement learning without centralization.

Ax Nilaksh, Saurav Jha, Artem Zholus, Sarath Chandar 5/8/2026

Reconstruction or Semantics? What Makes a Latent Space Useful for Robotic World Models

Compares latent space choices (VAE vs semantic) for robotic world models using action-conditioned video diffusion for policy evaluation.

Ax Canyu Zhao, Hao Chen, Yunze Tong, Yu Qiao, Jiacheng Li, Chunhua Shen 5/8/2026

MARBLE: Multi-Aspect Reward Balance for Diffusion RL

Reinforcement learning method for fine-tuning diffusion models to balance multiple reward criteria simultaneously.

Ax Peiying Zhu, Sidi Chang 5/8/2026

Market-Alignment Risk in Pricing Agents: Trace Diagnostics and Trace-Prior RL under Hidden Competitor State

Study of pricing agents in revenue management showing how standard agents can achieve high returns while failing at market-like behavior.

Ax Maria Ana Cardei, Matthew Landers, Afsaneh Doryab 5/8/2026

Coordination Matters: Evaluation of Cooperative Multi-Agent Reinforcement Learning

Evaluation framework for cooperative multi-agent RL emphasizing coordination diagnostics beyond aggregate outcome metrics.

Ax Shuyang Jiang, Nan Yu, Yiming Zhang, Zenghui Ding, Zhenyu Wu 5/8/2026

DINORANKCLIP: DINOv3 Distillation and Injection for Vision-Language Pretraining with High-Order Ranking Consistency

Vision-language pretraining method combining DINO distillation with ranking-consistency loss for improved CLIP performance.

Ax David M\"uller, Agon Serifi, Sammy Christen, Ruben Grandia, Espen Knoop, Moritz B\"acher 5/8/2026

ReActor: Reinforcement Learning for Physics-Aware Motion Retargeting

Bilevel optimization framework for robot motion retargeting using reinforcement learning to ensure physical feasibility.

Ax Shuo Liu, Xinzichen Li, Christopher Amato 5/8/2026

Cross-Modal Navigation with Multi-Agent Reinforcement Learning

Multi-agent reinforcement learning approach for cross-modal embodied navigation using modality-specialized lightweight agents.