Ax Seyed Mahdi B. Azad, Jasper Hoffmann, Iman Nematollahi, Hao Zhu, Abhinav Valada, Joschka Boedecker 5/8/2026

Spectral Alignment in Forward-Backward Representations via Temporal Abstraction

Temporal abstraction method resolving spectral mismatch in forward-backward successor representation learning.

Ax Yuning Huang, Xiaoyu Ji, Joseph Huang, Yichi Zhang, Fengqing Zhu 5/8/2026

Adaptive Greedy Frame Selection for Long Video Understanding

Greedy frame selection algorithm for efficient long-video understanding balancing relevance and temporal coverage.

Ax Tony Mason, Vaastav Anand 5/8/2026

Epistemic Observability in Language Models

Formal proof that LLM confidence inversely correlates with accuracy due to observational constraints, not capability gaps.

Ax Abhinaba Basu, Kumkum Basu, Koushik Deb 5/8/2026

Structural Sensitivity in Compressed Transformers: Relative Error Propagation and Layer Removal

Analysis of error propagation through compressed transformer layers and impact of layer removal on model performance.

Ax Xinyu Lu, Kaiqi Zhang, Jinglin Yang, Boxi Cao, Yaojie Lu, Hongyu Lin, Min He, Xianpei Han, Le Sun 5/8/2026

P^2O: Joint Policy and Prompt Optimization

Joint optimization of LLM policies and prompts via RLVR to address advantage collapse on hard reasoning samples.

Ax Naveen Mysore 5/8/2026

Prediction-Based Markov Violation Scores for Detecting Non-Markovian Observations in Reinforcement Learning

Prediction-based metric for detecting Markov property violations in RL observation streams caused by noise and latency.

Ax Yongzhong Xu 5/8/2026

Spectral Edge Dynamics: An Analytical-Empirical Study of Phase Transitions in Neural Network Training

Spectral edge analysis explaining phase transitions in neural network training through rolling-window Gram matrix dynamics.

Ax Linda Zeng, Steven Y. Feng, Michael C. Frank 5/8/2026

Bringing Up a Bilingual BabyLM: Investigating Multilingual Language Acquisition Using Small-Scale Models

Investigation of multilingual language acquisition patterns using small-scale language models as experimental tool.

Ax Zheng Li, Jerry Cheng, Huanying Helen Gu 5/8/2026

StableTTA: Improving Vision Model Performance by Training-free Test-Time Adaptation Methods

Training-free test-time adaptation method improving vision model performance through ensemble aggregation stability.

Ax Yehui Yang, Zelin Zang, Xienan Zheng, Yuzhe Jia, Changxi Chi, Jingbo Zhou, Chang Yu, Jinlin Wu, Fuji Yang, Jiebo Luo, Zhen Lei, Stan Z. Li 5/8/2026

MAT-Cell: A Multi-Agent Tree-Structured Reasoning Framework for Batch-Level Single-Cell Annotation

Multi-agent tree-structured reasoning framework using LLMs for automated single-cell RNA-seq annotation with prompt optimization.

Ax Md Hasebul Hasan, Krity Haque Charu, Eshwara Prasad Sridhar, Shuchisnigdha Deb, Mohammad A. Islam 5/8/2026

DeEscalWild: A Real-World Benchmark for Automated De-Escalation Training with SLMs

Real-world benchmark for training de-escalation skills using small language models for law enforcement field training.

Ax Jeongjae Lee, Jinho Chang, Jeongsol Kim, Jong Chul Ye 5/8/2026

Reward Score Matching: Unifying Reward-based Fine-tuning for Flow and Diffusion Models

Framework unifying reward-based fine-tuning methods for diffusion and flow models through reward score matching perspective.

Ax Ha Lan N. T, Minh-Anh Nguyen, Dung D. Le 5/8/2026

Latent Abstraction for Retrieval-Augmented Generation

Research proposal for improving RAG systems by using latent representations instead of natural language queries, enabling closer retriever-generator integration.

Ax Andrew Wang, Ellie Pavlick, Ritambhara Singh 5/8/2026

Handling and Interpreting Missing Modalities in Patient Clinical Trajectories via Autoregressive Sequence Modeling

Research on multimodal ML for healthcare addressing missing modalities in clinical data using autoregressive sequence modeling for temporal trajectories.

Ax Yuan Zhuang, Yuexin Bian, Sihong He, Jie Feng, Qing Su, Songyang Han, Jonathan Petit, Shihao Ji, Yuanyuan Shi, Fei Miao 5/8/2026

Low-Rank Adaptation for Critic Learning in Off-Policy Reinforcement Learning

Uses Low-Rank Adaptation as structural regularizer for critic learning in off-policy reinforcement learning to reduce overfitting and instability.

Ax Spyros Galanis 5/8/2026

Information Aggregation with AI Agents

Studies how LLM-based AI agents aggregate dispersed information through prediction market trading and reason about others' knowledge via price signals.

Ax Jo\~ao Mattos, Arlei Silva 5/8/2026

Mochi: Aligning Pre-training and Inference for Efficient Graph Foundation Models via Meta-Learning

Mochi proposes graph foundation model using meta-learning to align pre-training and inference for efficient downstream task performance.

Ax Fang Wan, Guangyi Huang, Tianyu Wu, Zishang Zhang, Bangchao Huang, Haoran Sun, Mingdong Chen, Chaoyang Song 5/8/2026

asRoBallet: Closing the Sim2Real Gap via Friction-Aware Reinforcement Learning for Underactuated Spherical Dynamics

asRoBallet deploys reinforcement learning policy on humanoid ballbot hardware, addressing sim-to-real gap for underactuated spherical dynamics.

Ax Chu-Cheng Lin, Eugene Ie 5/8/2026

How Fast Should a Model Commit to Supervision? Training Reasoning Models on the Tsallis Loss Continuum

Analyzes SFT-then-RLVR training ordering for reasoning models via Tsallis loss family, providing theoretical framework for post-training strategies.

Ax Jun Guo, Qiwei Li, Peiyan Li, Zilong Chen, Nan Sun, Yifei Su, Heyun Wang, Yuan Zhang, Xinghang Li, Huaping Liu 5/8/2026

Unified 4D World Action Modeling from Video Priors with Asynchronous Denoising

X-WAM unifies robotic action execution and 4D world synthesis using video diffusion model priors for real-time robotics applications.

Ax Chengcao Yang 5/8/2026

ANCORA: Learning to Question via Manifold-Anchored Self-Play for Verifiable Reasoning

ANCORA uses self-play curriculum learning where a unified policy alternates between generating verifiable problems and solving them without human annotations.

Ax Jugal Gajjar, Kamalasankari Subramaniakuppusamy 5/8/2026

RSAT: Structured Attribution Makes Small Language Models Faithful Table Reasoners

RSAT trains small language models to produce step-by-step table reasoning with cell-level citations via structured output and reward optimization.

Ax Bingzheng Gan, Tianyi Zhang, Yusu Li, Jing Huang, Wei Shi, Yangkai Ding, Tao Yu 5/8/2026

Caracal: Causal Architecture via Spectral Mixing

Caracal proposes efficient LLM architecture using Fast Fourier Transform for sequence mixing instead of attention, achieving O(L log L) complexity.

Ax Chenyu Huang, Jianghao Lin, Zhengyang Tang, Bo Jiang, Ruoqing Jiang, Benyou Wang, Lai Wei 5/8/2026

InvEvolve: Evolving White-Box Inventory Policies via Large Language Models with Performance Guarantees

InvEvolve uses LLMs with evolutionary search to evolve inventory policies in dynamic, non-stationary environments with performance guarantees.

Ax Cutter Dawes, Aryan Sharma, Angelos Ioannis Lagos, Shivam Raval 5/8/2026

H-Probes: Extracting Hierarchical Structures From Latent Representations of Language Models

H-Probes method extracting hierarchical structure from LLM latent representations via linear probes, analyzing geometric representation of hierarchical reasoning.

Ax Iman Sharifi, Hyeong Tae Kim, Maheed Hatem Ahmed, Mahsa Ghasemi, Peng Wei 5/8/2026

Separation Assurance between Heterogeneous Fleets of Small Unmanned Aerial Systems via Multi-Agent Reinforcement Learning

Multi-agent reinforcement learning for tactical deconfliction of heterogeneous unmanned aerial systems in dense airspace, handling fleet-level policies and constraints.

Ax Minh-Dung Le, Minh-Duc Hoang, Hoang-Vu Truong, Thi-Thu-Hong Phan 5/8/2026

AgriKD: Cross-Architecture Knowledge Distillation for Efficient Leaf Disease Classification

Knowledge distillation from Vision Transformers to efficient models for leaf disease classification on edge devices, balancing accuracy and deployment constraints.

Ax Anamika Paul Rupa, Anietie Andy 5/8/2026

Probe-Geometry Alignment: Erasing the Cross-Sequence Memorization Signature Below Chance

Method for surgically removing memorization traces from unlearned LLMs using leave-one-out cross-sequence probes, eliminating recovery by adversarial attacks.

Ax Javad Forough, Marios Kogias, Hamed Haddadi 5/8/2026

When Agents Handle Secrets: A Survey of Confidential Computing for Agentic AI

Survey of confidential computing approaches for agentic AI systems, addressing threat surface from persistent memory, credentials, and cross-agent protocols like MCP.

Ax Ishrith Gowda (University of California, Berkeley) 5/8/2026

MEMSAD: Gradient-Coupled Anomaly Detection for Memory Poisoning in Retrieval-Augmented Agents

Formalization and defense of memory poisoning attacks on retrieval-augmented LLM agents using gradient-coupled anomaly detection across three attack classes.

Ax Jorge L. Ruiz Williams 5/8/2026

HeadQ: Model-Visible Distortion and Score-Space Correction for KV-Cache Quantization

KV-cache quantization method measuring distortion in model-visible score space rather than storage space, with calibration-learned residual storage for efficiency.

Ax Zhen Liu, Yuhan Liu, Jinjun Wang, Wei Song, Jianyi Liu, Jingwen Fu 5/8/2026

Structured Progressive Knowledge Activation for LLM-Driven Neural Architecture Search

LLM-assisted neural architecture search with progressive knowledge activation, managing architectural priors while exploring new designs under expensive evaluations.

Ax Andreas Pattichis, Constantine Dovrolis 5/8/2026

Continual Knowledge Updating in LLM Systems: Learning Through Multi-Timescale Memory Dynamics

Memini system enabling continuous knowledge updating in deployed LLMs via multi-timescale memory dynamics mimicking biological learning and memory consolidation.

Ax Michael Timothy Bennett 5/8/2026

Are Flat Minima an Illusion?

Analysis showing sharp vs flat minima in neural networks are reparameterization artifacts without causal relationship to generalization, challenging sharpness-aware minimization theory.

Ax Yi Xie, Yangyang Xu, Yi Fan, Bo Liu 5/8/2026

SAT: Sequential Agent Tuning for Coordinator Free Plug and Play Multi-LLM Training with Monotonic Improvement Guarantees

Multi-agent training framework for coordinating multiple smaller LLMs without centralized coordinator, with monotonic improvement guarantees and stability mechanisms.

Ax Reza Pirayeshshirazinezhad 5/8/2026

Physics-Informed Neural Networks with Learnable Loss Balancing and Transfer Learning

Self-supervised physics-informed neural networks with learnable loss weighting for scientific ML under data scarcity, dynamically balancing physics and data supervision.

Ax Gauri Kale, Rahul Vishwakarma, Holly Diamond, Ava Hedayatipour, Amin Rezaei 5/8/2026

Horizon-Constrained Rashomon Sets for Chaotic Forecasting

Framework characterizing model multiplicity in chaotic prediction systems using horizon-constrained Rashomon sets, bridging predictive multiplicity and chaos theory.

Ax Mikhail Shirokikh, Sergey Nikolenko 5/8/2026

Sparse Prefix Caching for Hybrid and Recurrent LLM Serving

Optimization technique for LLM serving combining sparse checkpoint caching with recurrent state models, reducing latency for hybrid architectures beyond dense key-value caching.

Ax Tatiana Gaintseva, Andrew Stepanov, Ziquan Liu, Martin Benning, Gregory Slabaugh, Jiankang Deng, Ismail Elezi 5/8/2026

MidSteer: Optimal Affine Framework for Steering Generative Models

Theoretical framework formalizing concept steering in generative models via affine transformations, enabling controlled post-deployment alignment and safety applications.

Ax Andrew Kiruluta 5/8/2026

Data-Driven Variational Basis Learning Beyond Neural Networks: A Non-Neural Framework for Adaptive Basis Discovery

Non-neural framework for learning adaptive basis representations from data, offering interpretability alternatives to neural networks for high-dimensional data analysis.

Ax Ahmed Abdelmuniem Abdalla Mohammed 5/8/2026

Adaptive Computation Depth via Learned Token Routing in Transformers

Token-Selective Attention mechanism enabling adaptive computation depth via learned per-token routing in transformers.

Ax Yunpeng Zhou 5/8/2026

Structural Instability of Feature Composition

Theoretical analysis of compositional steering in Sparse Autoencoders, examining non-linear interference in feature activation.

Ax Bo Wang, Jia Ni, Mengnan Zhao, Zhan Qin, Kui Ren 5/8/2026

Channel-Level Semantic Perturbations: Unlearnable Examples for Diverse Training Paradigms

Unlearnable examples via semantic perturbations for privacy protection across training paradigms including pretraining-finetuning.

Ax Bo Li, Chuan Wu, shaolin Zhu 5/8/2026

MACS: Modality-Aware Capacity Scaling for Efficient Multimodal MoE Inference

Load balancing technique for multimodal MoE LLMs addressing information heterogeneity and stragglers in expert parallelism inference.

Ax Fei Ding, Yongkang Zhang, Runhao Liu, Yuhao Liao, Zijian Zeng, Sibo wang, Huiming Yang 5/8/2026

Internalizing Outcome Supervision into Process Supervision: A New Paradigm for Reinforcement Learning for Reasoning

Framework converting outcome-level supervision into process-level signals for reasoning tasks via reinforcement learning credit assignment.

Ax Wanru Zhao, Yihong Chen, Yuzhi Tang, Wentao Ma, Shengchao Hu, Shell Xu Hu, Alex Iacob, Abhinav Mehrotra, Nicholas D. Lane 5/8/2026

Rethinking Data Curation in LLM Training: Online Reweighting Offers Better Generalization than Offline Methods

Online data reweighting during LLM training outperforms offline curation methods, improving generalization without preprocessing overhead.

Ax Marcin Pietro\'n 5/8/2026

Evolutionary fine tuning of quantized convolution-based deep learning models

Evolutionary algorithms for fine-tuning quantized convolutional models for IoT and edge device deployment.

Ax Yazheng Liu, Yuxuan Wan, Rui Xu, Xi Zhang, Sihong Xie, Hui Xiong 5/8/2026

Attribution-Guided Continual Learning for Large Language Models

Attribution-based method to mitigate catastrophic forgetting in LLMs during continual learning by selectively updating parameters.

Ax Laurent Guigues 5/8/2026

Graph Normalization: Fast Binarizing Dynamics for Differentiable MWIS

Graph Normalization: differentiable dynamical system for approximating NP-hard Maximum Weight Independent Set with convergence guarantees.

Ax Faris Chaudhry, Keisuke Yano, Anthea Monod 5/8/2026

Feature Starvation as Geometric Instability in Sparse Autoencoders

Analysis of feature starvation in sparse autoencoders for LLM interpretability, proposing geometric solutions to dead neuron problems.