Isolater - Feed

Ax Henrik Voigt, Michael Habeck, Joachim Giesen 5/1/2026

Beyond the Training Distribution: Mapping Generalization Boundaries in Neural Program Synthesis

arXiv paper on evaluating generalization in neural program synthesis using controlled environments to detect memorization.

Ax Branislav Kveton, Michal Valko, Matthai Phillipose, Ling Huang 5/1/2026

Online semi-supervised perception: Real-time learning without explicit feedback

arXiv paper on online semi-supervised learning combining graph learning with real-time updates without explicit feedback.

Ax Branislav Kveton, Michal Valko 5/1/2026

Learning from a single labeled face and a stream of unlabeled data

arXiv paper on single-sample face recognition with semi-supervised learning from unlabeled data streams.

Ax Zhenzhou Jin, Li You, Xiang-Gen Xia, Xiqi Gao 5/1/2026

Statistical Channel Fingerprint Construction for Massive MIMO: A Unified Tensor Learning Framework

arXiv paper on tensor learning framework for MIMO channel fingerprinting. ML research but not AI/LLM focused.

Ax Al Zadid Sultan Bin Habib, Tanpia Tasnim, Md. Ekramul Islam, Muntasir Tabasum 5/1/2026

ZAYAN: Disentangled Contrastive Transformer for Tabular Remote Sensing Data

Presents ZAYAN, a self-supervised contrastive framework for learning representations from heterogeneous tabular remote sensing data.

Ax Chengcao Yang, Jun Chen 5/1/2026

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

Proposes ANCORA, a framework where language models learn to generate verifiable problems and improve via self-play without human supervision.

Ax Kathrin Korte, Joachim Winter Pedersen, Eleni Nisioti, Sebastian Risi 5/1/2026

When Does Structure Matter in Continual Learning? Dimensionality Controls When Modularity Shapes Representational Geometry

Investigates how dimensionality and modularity affect representational geometry in continual learning systems managing stability-plasticity tradeoffs.

Ax Corinna Cortes, Anqi Mao, Mehryar Mohri, Yutao Zhong 5/1/2026

Optimized Deferral for Imbalanced Settings

Improves learning to defer systems for imbalanced settings by routing uncertain inputs to specialized experts, reducing errors and computational cost.

Ax Mehryar Mohri, Yutao Zhong 5/1/2026

Mind the Gap: Structure-Aware Consistency in Preference Learning

Addresses theoretical inconsistency in Direct Preference Optimization and related surrogate losses for LLM alignment, proposing structure-aware consistency improvements.

Ax Mehryar Mohri, Yutao Zhong 5/1/2026

Linear-Core Surrogates: Smooth Loss Functions with Linear Rates for Classification and Structured Prediction

Proposes Linear-Core Surrogates, a family of convex loss functions balancing smooth optimization rates with fast consistency bounds for classification.

Ax Chendi Qian, Christopher Morris 5/1/2026

On the Expressive Power of GNNs to Solve Linear SDPs

Studies expressive power of graph neural networks as computational surrogates for solving large semidefinite programming problems in convex optimization.

Ax Bokai Pan, Mingyue Cheng, Zhiding Liu, Shuo Yu, Xiaoyu Tao, Yuchong Wu, Qi Liu, Defu Lian, Enhong Chen 5/1/2026

CastFlow: Learning Role-Specialized Agentic Workflows for Time Series Forecasting

CastFlow uses role-specialized LLM agents in iterative workflows for time series forecasting beyond static generative approaches.

Ax Logan G Wright, Tianyu Wang, Tatsuhiro Onodera, Peter L. McMahon 5/1/2026

Physical Foundation Models: Fixed hardware implementations of large-scale neural networks

Physical foundation models implementing large-scale neural networks (10^12+ parameters) in fixed hardware for diverse downstream tasks.

Ax Nghia Bui, Lijing Wang 5/1/2026

Dynamic Scaled Gradient Descent for Stable Fine-Tuning for Classifications

Dynamic scaled gradient descent algorithm for stable fine-tuning of pretrained models on sparse imbalanced datasets.

Ax Jingcheng Deng, Zihao Wei, Liang Pang, Junhong Wu, Shicheng Xu, Zenghao Duan, Huawei Shen 5/1/2026

Latent-GRPO: Group Relative Policy Optimization for Latent Reasoning

Latent-GRPO applies group relative policy optimization to latent reasoning in LLMs for stable RL training with compressed representations.

Ax Shijin Gong, Kai Ye, Jin Zhu, Xinyu Zhang, Hongyi Zhou, Chengchun Shi 5/1/2026

Kernelized Advantage Estimation: From Nonparametric Statistics to LLM Reasoning

Kernelized advantage estimation for LLM reasoning combining nonparametric statistics with RL policy gradient methods.

Ax Prabhjot Singh, Abhishek Gupta, Chris Betz, Abe Flansburg, Brett Ives, Sudeep Lama, Jung Hoon Son 5/1/2026

Learning from Disagreement: Clinician Overrides as Implicit Preference Signals for Clinical AI in Value-Based Care

Framework reframing clinician overrides of AI recommendations as preference signals for RLHF-style learning in clinical AI.

Ax Clara Mohri, Amir Globerson, Haim Kaplan, Tomer Koren, Yishay Mansour 5/1/2026

Cost-Aware Learning

Cost-aware stochastic gradient descent algorithm optimizing total cost to reach target error for finite-sum objectives.

Ax Zhiqiang Kou, Junxiang Wu, Wenke Huang, Wenwen He, Ming-Kun Xie, Changwei Wang, Yuheng Jia, Di Jiang, Yang Liu, Xin Geng, Qiang Yang 5/1/2026

FedHarmony: Harmonizing Heterogeneous Label Correlations in Federated Multi-Label Learning

Federated learning approach for multi-label classification with heterogeneous label distributions across clients.

Ax Jeanne Monnier, Thomas George, Fr\'ed\'eric Guyard, Christ\`ele Tarnec, Marios Kountouris 5/1/2026

MIFair: A Mutual-Information Framework for Intersectionality and Multiclass Fairness

MIFair framework for fairness assessment and bias mitigation in multiclass and intersectional ML settings using mutual information.

Ax Marc Dymetman 5/1/2026

Exponential families from a single KL identity

Theoretical analysis of exponential families via KL identity with applications to variational inference, RL, and RLHF.

Ax Arthur Corr\^ea, Paulo Nascimento, Samuel Moniz 5/1/2026

FiLMMeD: Feature-wise Linear Modulation for Cross-Problem Multi-Depot Vehicle Routing

Applies feature-wise linear modulation to neural combinatorial optimization for solving multi-depot vehicle routing problems.

Ax Junqi Gao, Dazhi Zhang, Zhichang Guo, Biqing Qi, Yi Ran, Wangmeng Zuo 5/1/2026

Auto-FlexSwitch: Efficient Dynamic Model Merging via Learnable Task Vector Compression

Proposes Auto-FlexSwitch for efficient dynamic model merging in multi-task adaptation through learnable task vector compression.

Ax Usha Bhalla, Thomas Fel, Can Rager, Sheridan Feucht, Tal Haklay, Daniel Wurgaft, Siddharth Boppana, Matthew Kowal, Vasudev Shyam, Jack Merullo, Atticus Geiger, Ekdeep Singh Lubana 5/1/2026

Do Sparse Autoencoders Capture Concept Manifolds?

Investigates whether sparse autoencoders capture concept manifolds rather than independent linear directions, questioning core assumptions in interpretability research.

Ax Florian Wolf, Ilyas Fatkhullin, Niao He 5/1/2026

Global Optimality for Constrained Exploration via Penalty Regularization

Penalty regularization method for constrained maximum-entropy exploration in reinforcement learning.

Ax Haidong Zhao, Nikolaos Georgantas 5/1/2026

Strait: Perceiving Priority and Interference in ML Inference Serving

Strait system for prioritization and latency estimation in ML inference serving under concurrent GPU execution.

Ax Himanshu Pandey, Ratikanta Behera 5/1/2026

An adaptive wavelet-based PINN for problems with localized high-magnitude source

Adaptive wavelet-based physics-informed neural networks addressing spectral bias and loss imbalance.

Ax Eyon Jang, Damon Falck, Joschka Braun, Nathalie Kirch, Achu Menon, Perusha Moodley, Scott Emmons, Roland S. Zimmermann, David Lindner 5/1/2026

Exploration Hacking: Can LLMs Learn to Resist RL Training?

Study on LLM behavior during RL training: models may strategically reduce exploration to manipulate post-training outcomes.

Ax Daghash K. Alqahtani, Aamir Cheema, Adel N. Toosi 5/1/2026

Benchmarking Deep Learning Models for Object Detection on Edge Computing Devices

Benchmarking YOLOv8 and EfficientDet models for real-time object detection on resource-constrained edge devices.

Ax Yifei Wang, Hancheng Ye, Yechen Xu, Cong Guo, Chiyue Wei, Qinsi Wang, Dongting Li, Tingjun Chen, Hai "Helen" Li, Danyang Zhuo, Yiran Chen 5/1/2026

MARS: Efficient, Adaptive Co-Scheduling for Heterogeneous Agentic Systems

MARS system for efficient co-scheduling of heterogeneous GPU-CPU resources in multi-turn agentic LLM workloads.

Ax Mary Nusrat, Sarfuddin Bhuiyan, Gahangir Hossain 5/1/2026

Learning-to-Explain through 20Q Gaming: An Explainable Recommender for Cybersecurity Education

Explainable AI educational framework using 20 Questions game for adaptive cybersecurity training.

Ax Mandar Karhade 5/1/2026

Not All Memories Age the Same: Autodiscovery of Adaptive Decay in Knowledge Graphs

Framework for adaptive decay in knowledge graphs recognizing different temporal dynamics per knowledge type.

Ax Shawqi Al-Maliki, Ammar Gharaibeh, Mohamed Rahouti, Mohammad Ruhul Amin, Mohamed Abdallah, Junaid Qadir, Ala Al-Fuqaha 5/1/2026

Budget-Constrained Online Retrieval-Augmented Generation: The Chunk-as-a-Service Model

Chunk-as-a-Service model for budget-constrained retrieval-augmented generation addressing RAG limitations.

Ax Mar\'ia Florencia Acosta, Rodrigo Garc\'ia Arancibia, Pamela Llop, Mariel Lovatto, Lucas Mansilla 5/1/2026

Value-Aware Product Recommendation by Customer Segmentation using a suitable High-Dimensional Similarity Measure

arXiv paper on value-aware product recommendation using customer segmentation with similarity measures.

Ax Gustav Olaf Yunus Laitinen-Fredriksson Lundstrom-Imanov, Taner Yilmaz 5/1/2026

EdgeSpike: Spiking Neural Networks for Low-Power Autonomous Sensing in Edge IoT Architectures

EdgeSpike framework for spiking neural networks on IoT edge devices with hardware-aware NAS and energy budgets.

Ax Minhee Park, Seongyeon Son, Yonghyun Lee, Eunchan Kim 5/1/2026

SCOPE-FE: Structured Control of Operator and Pairwise Exploration for Feature Engineering

SCOPE-FE structured search space for automatic feature engineering in tabular learning, reducing combinatorial explosion of operator-feature combinations.

Ax Katelyn Crumpacker, Dimitrios Nikolopoulos 5/1/2026

LLM-Guided Runtime Parameter Optimization for Energy-Efficient Model Inference

Uses LLMs to optimize runtime parameters (batch size, quantization) for energy-efficient inference in production LLM services.

Ax Emma Casey, David Roberts, David Sim, Ian Beaver 5/1/2026

When Your LLM Reaches End-of-Life: A Framework for Confident Model Migration in Production Systems

Framework for migrating production LLM systems using Bayesian calibration of automated metrics against human judgments for confident model replacement.

Ax Yibin Luo, Shiwei Gao, Huichuan Zheng, Youyou Lu, Jiwu Shu 5/1/2026

Efficient Training on Multiple Consumer GPUs with RoundPipe

RoundPipe framework for efficient LLM fine-tuning on consumer GPUs using pipeline parallelism and CPU offloading, addressing weight binding limitations.

Ax Paraskevas V. Lekeas, Giorgos Stamatopoulos 5/1/2026

What Suppresses Nash Equilibrium Play in Large Language Models? Mechanistic Evidence and Causal Control

Mechanistic study of why LLM agents deviate from Nash equilibria in game theory, with causal interventions on Llama-3 and Qwen2.5 models to reverse deviations.

Ax Austin C. Kozlowski, Andrei Boutyline 5/1/2026

Semantic Structure of Feature Space in Large Language Models

Analysis showing LLM hidden state feature geometry mirrors human semantic associations, validated through projection onto 32 semantic axes with human rating correlations.

Ax Shouren Wang, Wang Yang, Chuang Ma, Debargha Ganguly, Vikash Singh, Chaoda Song, Xinpeng Li, Xianxuan Long, Vipin Chaudhary, Xiaotian Han 5/1/2026

Path-Lock Expert: Separating Reasoning Mode in Hybrid Thinking via Architecture-Level Separation

Path-Lock Expert (PLE) architecture separates explicit thinking and non-thinking modes in hybrid LLMs through dedicated feed-forward parameters to reduce reasoning leakage.

Ax Anh Ta, Junjie Zhu, Shahin Shayandeh 5/1/2026

Reinforced Agent: Inference-Time Feedback for Tool-Calling Agents

Framework for tool-calling agents that moves evaluation into inference-time execution loop with real-time feedback to correct agent behavior during active tool use.

Ax Chung-Hsiang Lo, Lu Li, Diji Yang, Tianyu Zhang, Yunkai Zhang, Yoshua Bengio, Yi Zhang 5/1/2026

When 2D Tasks Meet 1D Serialization: On Serialization Friction in Structured Tasks

Studies how LLMs handle 2D structured data through 1D serialization, identifying representational burden and 'serialization friction' in tasks requiring explicit spatial relationships.

Ax Mehmet Iscan 5/1/2026

Learning When to Remember: Risk-Sensitive Contextual Bandits for Abstention-Aware Memory Retrieval in LLM-Based Coding Agents

Risk-sensitive contextual bandits framework for LLM-based coding agents to safely retrieve and reuse external memory while avoiding unsafe injections from superficially similar failures.

Ax Ming-Hong Yao, Di Wang, Jian Cui, Jin-Yan Chen, Zi-Hao Cui, Fa Wang, Chen Wei, Qiu-Ye Yu 5/1/2026