Ax Hongmin Li 7/1/2026

The Calibration Turn in AI-Assisted Research: A Conceptual and Methodological Framework for Evidence-Licensed Claims

Framework for evaluating calibration of scientific claims made by AI-assisted research systems to underlying evidence.

Ax Ari Pakman, Lior Kreimer, Yakir Berchenko 7/1/2026

Revisiting the Volume Hypothesis

Investigation of volume hypothesis explaining neural network generalization through loss-landscape basin geometry and implicit bias of SGD.

Ax Onkar Jadhav, Tim French, Matthew Rayson, Nicole L. Jones 7/1/2026

Patch-PODiff-ViT: Structured Latent Diffusion with Patchwise POD for Super-Resolution and Uncertainty Quantification

Structured latent diffusion framework using patchwise POD for super-resolution with uncertainty quantification.

Ax Alexander Fabisch, Melvin Laux, Mariela De Lucas \'Alvarez, Edoardo Caroselli, Julian Theis 7/1/2026

Deep Reinforcement Learning for Spacecraft Attitude Control During Atmospheric Re-Entry

Deep reinforcement learning applied to spacecraft attitude control during atmospheric re-entry.

Ax Hongpeng Cao, Liqun Zhao, Yuliang Gu, Naira Hovakimyan, Lui Sha, Marco Caccamo 7/1/2026

Safe Online Learning via Smooth Safety-Structured Policy Composition

Safety-aware policy composition architecture for safe online reinforcement learning with smooth optimization dynamics.

Ax Mohamad Mestoukirdi, Vincent Corlay 7/1/2026

Expected Gain-based Escalation in Vertical Federated Learning

Method for reducing communication overhead in vertical federated learning through selective collaborative inference.

Ax Zewen Liu 7/1/2026

Calibrating the Evaluator: Does Probability Calibration Mitigate Preference Coupling in LLM Agent Feedback Loops?

Investigation of probability calibration to mitigate evaluator preference coupling in LLM agent feedback loops.

Ax Jisung Park, Seohyeon Kang, Daeun Yoo, Eunsu Lee, Seoin Cho, Wooyeop Choi, Ian Choi, James R. Evan, Daesoo Kim, Sonia Gandhi, Minee L. Choi 7/1/2026

Resolving superposition in AI for interpretability and cross-modal alignment in patient-neuronal images

Study of neural network superposition in high-dimensional biological data using sparse autoencoders for interpretability.

Ax Duc Anh Nguyen, Tien Ngoc Luu, Tung Pham, Toan Tran 7/1/2026

Mixture-of-Control: State-Aware Fine-Tuning for Transformer-based Models

State-based fine-tuning method for transformers with mixture-of-control approach for efficient parameter adaptation.

Ax Hyunho Lee, Kyomin Hwang, Hyeonjin Kim, Suyoung Kim, Sunghyun Wee, Nojun Kwak 7/1/2026

Zero-Shot Quantization for Object Detectors using Off-the-Shelf Generative Models

Zero-shot quantization for object detection models using generative models for training data synthesis without original data access.

Ax Dariush Wahdany, Matthew Jagielski, Jesse C. Cresswell, Adam Dziedzic, Franziska Boenisch 7/1/2026

TabPATE: Differentially Private Tabular In-Context Learning Without Public Data

Differentially private PATE-style defense for tabular in-context learning that protects private records without public data.

Ax Zena Al-Khalili, Rafi Hakim, Dietrich Klakow, Ji-Ung Lee 7/1/2026

Fork-Think with Confidence

Decide-first-then-think method for improving LLM reasoning tasks by identifying high-confidence decision points before generation.

Ax Wenhao Li, Jinhao Dong, Hailin Zhang, Wenhang Shi, Wei Lu, Xiaoyong Du 7/1/2026

RaBitQCache: Rotated Binary Quantization for KVCache in Long Context LLM Inference

Binary quantization method for KV cache optimization in long-context LLM inference using rotated binary quantization.

Ax Xudong Wu, Pangpang Liu, Vaneet Aggarwal, Jiayu Chen 7/1/2026

On the Convergence of Self-Improving Online LLM Alignment

Theoretical convergence analysis of Self-Improving Alignment algorithm for handling distribution shift in LLM alignment.

Ax Kung-Ming Lan 7/1/2026

Beyond the Expressivity-Trainability Paradox: A Dynamical Lie Algebra Perspective on Navigating Barren Plateaus in Quantum Machine Learning

Analysis of barren plateaus in quantum machine learning using Lie algebra perspective to address expressivity-trainability tradeoffs.

Ax Ole Winther, Paul Jeha, Sander Dieleman, Andriy Mnih, Manfred Opper, Andrea Dittadi 7/1/2026

Introduction to Stochastic Differential Equations for Generative Machine Learning: A Variational Perspective

Educational introduction to stochastic differential equations for generative modeling with applications to image, video, and biomolecule generation.

Ax Jason R. Brown, Patrick Leask, Lev McKinney 7/1/2026

Evil Spectra: How Optimisers can Amplify or Suppress Emergent Misalignment

Study of emergent misalignment in LLMs during fine-tuning, analyzing how optimizers and training choices affect the phenomenon across multiple models.

Ax Jian Xu, Delu Zeng, John Paisley, Qibin Zhao 7/1/2026

Calibration, Not Compilation: Detecting and Repairing Misspecified Probabilistic Programs Written by Language Models

Research paper on detecting and repairing statistical errors in probabilistic programs generated by LLMs. Uses Bayesian workflow for verification beyond compilation.

Ax Zijun Xie, Binbin Zheng, Enlei Gong, Jihua Liu, Yuyang You, Lingfeng Liu, Jiayao Tang, Guanqun Zhao, Aoqi Hu, Zeyu Chen 7/1/2026

ECHO: Prune to act, trace to learn with selective turn memory in agentic RL

Proposes ECHO for managing context in long-horizon language agents via selective memory pruning and turn tracing under bounded context windows.

Ax Matteo Melis, Jesus Martinez Del Rincon, Vishal Sharma 7/1/2026

Improving Certified Robustness via Adversarial Distillation

Combines certified training and adversarial distillation to improve robustness-accuracy trade-offs in neural networks.

Ax Yutaro Yamada, Luca Grillotti, Rujikorn Charakorn, Sebastian Risi, David Ha, Robert Tjarko Lange 7/1/2026

Diffusing Blame: Task-Dependent Credit Assignment in Biologically Plausible Dual-Stream Networks

Studies credit assignment in biologically plausible neural networks respecting Dale's principle for learning rules.

Ax Shuai Yuan, Sudong Cai, Bingzhi Chen, Shuyuan Zheng, Chuan Xiao, Makoto Onizuka, Rui Mao 7/1/2026

Nonlinearity-Aware LoRA: Structured Gate Adaptation under Low-Rank Constraints

Analyzes low-rank adaptation in gated transformer networks, identifying selection misalignment issues and proposing structured gate adaptation.

Ax Maximilian Andreas Hoefler, Karsten Mueller, Wojciech Samek 7/1/2026

FedXDS: Leveraging Model Attribution Methods to counteract Data Heterogeneity in Federated Learning

Applies explainable AI attribution methods to improve federated learning under data heterogeneity.

Ax Taeyoun Kim, Aviral Kumar 7/1/2026

Addressing Over-Refusal in LLMs with Competing Rewards

Addresses over-refusal in LLM safety training using reinforcement learning with competing rewards to balance safety and helpfulness.

Ax Mingyi Li, Taira Tsuchiya, Kenji Yamanishi 7/1/2026

Policy Optimization Achieves Data-Dependent Regret Bounds in MDPs with Unknown Transitions

Theoretical analysis of policy optimization for online tabular MDPs with unknown transitions, achieving data-dependent regret bounds.

Ax Ying Fan, Anej Svete, Kangwook Lee 7/1/2026

Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers

Proposes looped transformers to bridge latent and explicit chain-of-thought reasoning, addressing performance gaps in language models at scale.

Ax Keivan Faghih Niresi, Alice Cicirello, Olga Fink 7/1/2026

Relational and Sequential Conformal Inference for Energy Time Series over Graphs via Foundation Models

Applies foundation models and graph neural networks to energy demand forecasting with conformal inference for uncertainty quantification.

Ax Ruijia Zhang, Jiacheng Zhu, Hanqing Zhu, Laixi Shi 7/1/2026

Geometry-Preserving Orthonormal Initialization for Low-Rank Adaptation in RLVR

Studies low-rank adaptation (LoRA) variants under reinforcement learning with verifiable rewards, showing PiSSA and MiLoRA underperform standard LoRA in this setting.

Ax Junyu Ren, Lek-Heng Lim 7/1/2026

Low-dimensional topology of deep neural networks

Studies topological properties of layered neural networks including Transformers by constraining representations to low-dimensional spaces.

Ax Kyle Kramer 7/1/2026

Review Residuals: Update-Conditioned Residual Gating for Transformers

Review Residuals introduces learned, input-dependent gating of residual updates in Transformers based on current state and proposed update reliability.

Ax John Sweeney 7/1/2026

Signed-Permutation Coordinate Transport for RMSNorm Transformers

Develops coordinate transport methods for RMSNorm Transformers to properly handle steering vectors and interpretability tools across model checkpoints.

Ax Wojciech {\L}apacz, Stanis{\l}aw Pawlak 7/1/2026

Amplifying Membership Signal Through Chained Regeneration

MADreMIA framework enhances membership inference attacks on generative models through chained regeneration for improved privacy auditing.

Ax Srijan Tiwari, Aditya Chauhan, Manjot Singh 7/1/2026

Radial Suppression Accelerates Algorithmic Generalization: A Geometric Analysis of Delayed Generalization

Geometric analysis of memorization-generalization delay in neural networks, showing radial inflation drives delayed generalization on algorithmic tasks.

Ax Philippe Chlenski, Zachariah Carmichael, Ayush Warikoo, Chia-Tse Shao, Yingxiao Ye, Aobo Yang, Vivek Miglani, Nehal Bandi 7/1/2026

Surrogate Fidelity: When Can Open LLMs Explain Closed Ones?

Studies when open-source LLMs can explain closed models through mechanistic interpretability, evaluating surrogate fidelity across prediction and representation levels.

Ax Sanghyuk Chun, William Yang, Amaya Dharmasiri, Olga Russakovsky 7/1/2026

CoMet: Context and Multiplicity Decomposition for Multimodal Uncertainty Estimation

CoMet addresses uncertainty estimation in multimodal LLMs by decomposing uncertainty sources into context and multiplicity components.

Ax Zekai Chen, Kairui Yang, Xuaner Chen, Xunkai Li, Xun Wu, Rong-Hua Li, Guoren Wang 7/1/2026

FedLAB: Traceable Semantic Codebooks for Federated Multimodal Graph Foundation Learning

FedLAB develops federated learning for multimodal graph foundation models with privacy-preserving semantic codebooks for distributed graph data.

Ax Yuanda Xu, Zhengze Zhou, Hejian Sang, Xiaomin Li, Jiaxin Zhang, Xinchen Du, Zhipeng Wang, Alborz Geramifard 7/1/2026

TRIAGE: Role-Typed Credit Assignment for Agentic Reinforcement Learning

TRIAGE proposes role-typed credit assignment for agentic RL, improving upon GRPO by differentiating credit across action types in agent-environment interactions.

Ax Jian Gu, Aldeida Aleti, Chunyang Chen, Hongyu Zhang 7/1/2026

SemRF: A Semantic Reference Frame for Residual-Stream Dynamics in Language Models

SemRF: semantic reference frame formalism for analyzing residual-stream dynamics in language models.

Ax Sergio Hern\'andez-Guti\'errez, Matteo Merler, Ilze Amanda Auzina, Joschka Str\"uber, Ameya Prabhu, Matthias Bethge 7/1/2026

QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agents

QVal: method for efficiently evaluating dense supervision signals for long-horizon LLM agents.

Ax Shi Bo, Debarghya Mukherjee, AmirEmad Ghassami 7/1/2026

MediEncoder: Nonlinear Representation Learning for High-Dimensional Causal Mediation Analysis

MediEncoder: nonlinear representation learning for causal mediation analysis with high-dimensional data.

Ax Xinyu Qin, Vicky Ye, Ruiheng Yu, Lu Wang 7/1/2026

Can Physician Expertise Improve Machine Learning Identification of Delirium?

User-centered interactive machine learning framework for physician-guided delirium detection.

Ax Olivier Jiyoun Jung, Jonghyeon Park, Myungwoo Oh 7/1/2026

Listening Between the Lines: Joint Learning of ASR Embeddings and LLM-Augmented Linguistics for Dementia Detection

Multimodal framework combining speech embeddings and LLM-augmented linguistics for dementia detection.

Ax Muhammad Hamza 7/1/2026

Criticality-Constrained Iterative Pruning for Energy-Efficient Spiking Neural Networks via Combined Importance Scoring

CQP: pruning method for energy-efficient spiking neural networks on neuromorphic hardware.

Ax Kaier Liang, Hengde Dai, Cristian-Ioan Vasile 7/1/2026

ViTL: Temporal Logic-Guided Zero-Shot Natural Language Navigation via Vision-Language Models

ViTL: zero-shot robotic navigation using vision-language models and temporal logic constraints.

Ax Ludovic K. Tuncay (IRIT-SAMoVA), Etienne Labb\'e (IRIT-SAMoVA), Thomas Pellegrini (IRIT-SAMoVA) 7/1/2026

BEST-RQ-2: Contextualize-Then-Predict, a Two-Step Approach for Self-Supervised Audio Representations

BEST-RQ-2: self-supervised audio representation learning using two-step contextualize-then-predict pretraining scheme.

Ax Alessandro Morosini, Sarah H. Cen, Andrew Ilyas, Hedi Driss, Aleksander M\k{a}dry, Chara Podimata 7/1/2026

Using AI Agents to Automate Black-Box Audits of Personalization Algorithms at Scale

AI agents for automated black-box auditing of personalization algorithms at scale, combining simulated and real-user approaches.

Ax Zhe Dong (University of Maine at Presque Isle), Fang Qin (Stanford University), Manish Shah (Independent Researcher) 7/1/2026

When Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning Models

LearnStop: learned stopping rule for reasoning LLMs that predicts prefix correctness to optimize computation budget across instances.

Ax Brent A. Griffin, Jason J. Corso 7/1/2026

The Label Imitation Game: Turing Test Network for Zero-Shot Pseudo-Label Pruning

Label Imitation Game: adversarial framework for zero-shot pseudo-label pruning from foundation models using Turing test networks.

Ax Idelfonso B. R. Nogueira, Sigurd Skogestad 7/1/2026

A Systematic Approach to Multi-Agent AI from Advanced Regulatory Control Theory: Safe and Auditable LLM Operator Agents for Process Control

Multi-agent LLM system using process control theory for safe, auditable operator agents in domain-specific process control tasks.

Ax Yongbin Kim, Yashar Talebirad, Osmar R. Zaiane 7/1/2026

Why Solve It Twice? Hierarchical Accumulation of Skills for Transfer-Efficient ML Engineering

HASTE: hierarchical multi-agent system that accumulates ML techniques across competitions, enabling transfer-efficient engineering via LLM-driven abstraction.