Isolater - Feed

Ax Marzi Heidari, Hanping Zhang, Hao Yan, Yuhong Guo 20d ago

Bi-Level Optimization for Single Domain Generalization

Bi-level optimization framework (BiSDG) for single domain generalization without target domain access during training.

Ax Gautham Vinod, Siddeshwar Raghavan, Bruce Coburn, Fengqing Zhu 20d ago

DietDelta: A Vision-Language Approach for Dietary Assessment via Before-and-After Images

Vision-language framework for dietary assessment and nutritional analysis from before-and-after food images.

Ax Charlotte Pouw, Hosein Mohebbi, Afra Alishahi, Willem Zuidema 20d ago

In-Context Learning in Speech Language Models: Analyzing the Role of Acoustic Features, Linguistic Structure, and Induction Heads

Analyzes in-context learning in speech language models, studying acoustic features, linguistic structure, and induction heads for text-to-speech tasks.

Ax Ziwei Li, Rumali Perera, Angus Forbes, Ken Moreland, Dave Pugmire, Scott Klasky, Wei-Lun Chao, Han-Wei Shen 20d ago

GS-Surrogate: Deformable Gaussian Splatting for Parameter Space Exploration of Ensemble Simulations

Deformable Gaussian Splatting surrogate model for interactive exploration of ensemble simulations.

Ax Ahmed Alansary, Molham Mohamed, Ali Hamdi 20d ago

A Severity-Based Curriculum Learning Strategy for Arabic Medical Text Generation

Severity-based curriculum learning strategy for Arabic medical text generation using LLMs.

Ax Guruprasad Viswanathan Ramesh, Asmit Nayak, Basieem Siddique, Kassem Fawaz 20d ago

WebSP-Eval: Evaluating Web Agents on Website Security and Privacy Tasks

WebSP-Eval benchmark for evaluating web agents on security and privacy tasks like cookie management and account settings configuration.

Ax Syed Mohammad Kashif, Ruiyin Li, Peng Liang, Amjed Tahir, Qiong Feng, Zengyang Li, Mojtaba Shahin 20d ago

Beyond Functional Correctness: Design Issues in AI IDE-Generated Large-Scale Projects

Empirical study of design issues in large-scale code generation by AI IDEs with agentic capabilities, beyond functional correctness.

Ax Rishab Balasubramanian, Pin-Jie Lin, Rituraj Sharma, Anjie Fang, Fardin Abdi, Viktor Rozgic, Zheng Du, Mohit Bansal, Tu Vu 20d ago

The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment

Proposes Master Key Hypothesis and UNLOCK method for training-free capability transfer across different LLM scales via linear subspace alignment.

Ax Samuel Yanes Luis, Alejandro Casado P\'erez, Alejandro Mendoza Barrionuevo, Dame Seck Diop, Sergio Toral Mar\'in, Daniel Guti\'errez Reina 20d ago

Uncertainty Estimation for Deep Reconstruction in Actuatic Disaster Scenarios with Autonomous Vehicles

Compares uncertainty quantification methods (Gaussian Processes, MC Dropout, Deep Ensembles, Evidential Deep Learning) for environmental field reconstruction from autonomous vehicle sensors in aquatic monitoring.

Ax Hikmat Khan, Usama Sajjad, Metin N. Gurcan, Anil Parwani, Wendy L. Frankel, Wei Chen, Muhammad Khalid Khan Niazi 20d ago

MorphDistill: Distilling Unified Morphological Knowledge from Pathology Foundation Models for Colorectal Cancer Survival Prediction

Framework distilling pathology foundation models for colorectal cancer survival prediction using knowledge distillation techniques.

Ax Sakib Mostafa, Lei Xing, Md. Tauhidul Islam 20d ago

Toward a universal foundation model for graph-structured data

Research on foundation models for graph-structured data in biomedical applications, extending language/vision foundation model approaches to graph analysis.

Ax Sylvia Vassileva, Ivan Koychev, Svetla Boytcheva 20d ago

FMI@SU ToxHabits: Evaluating LLMs Performance on Toxic Habit Extraction in Spanish Clinical Texts

arXiv paper on using LLMs for toxic habit extraction in Spanish clinical texts. Zero-shot and few-shot approaches for substance use classification.

Ax Yunze Xiao, Wenkai Li, Xiaoyuan Wu, Ningshan Ma, Yueqi Song, Weihao Xuan 20d ago

Say Something Else: Rethinking Contextual Privacy as Information Sufficiency

arXiv paper on privacy-preserving strategies for LLM-drafted messages. Evaluates suppression and generalization approaches for sensitive information.

Ax Yasamin Fayyaz, Li Yang, Khalil El-Khatib 20d ago

Towards Resilient Intrusion Detection in CubeSats: Challenges, TinyML Solutions, and Future Directions

Addresses cybersecurity vulnerabilities in CubeSats using TinyML-based intrusion detection systems for resource-constrained satellite environments.

Ax Rebecca M. M. Hicke, Sil Hamilton, David Mimno, Ross Deans Kristensen-McLachlan 20d ago

Attention Flows: Tracing LLM Conceptual Engagement via Story Summaries

Evaluates LLM performance on long-form text understanding by comparing human and model-generated novel summaries, assessing conceptual engagement patterns.

Ax Jonathan Nemitz, Carsten Eickhoff, Junyi Jessy Li, Kyle Mahowald, Michal Golovanevsky, William Rudman 20d ago

When to Call an Apple Red: Humans Follow Introspective Rules, VLMs Don't

Introduces Graded Color Attribution dataset to evaluate whether Vision-Language Models follow their own introspective reasoning rules compared to human behavior.

Ax Georgi Grazhdanski, Sylvia Vassileva, Ivan Koychev, Svetla Boytcheva 20d ago

Team Fusion@ SU@ BC8 SympTEMIST track: transformer-based approach for symptom recognition and linking

Transformer-based NER and entity linking approach for medical symptom recognition in SympTEMIST shared task using RoBERTa and SapBERT.

Ax Mingchen Zhuge, Changsheng Zhao, Haozhe Liu, Zijian Zhou, Shuming Liu, Wenyi Wang, Ernie Chang, Gael Le Lan, Junjie Fei, Wenxuan Zhang, Yasheng Sun, Zhipeng Cai, Zechun Liu, Yunyang Xiong, Yining Yang, Yuandong Tian, Yangyang Shi, Vikas Chandra, J\"urgen Schmidhuber 20d ago

Neural Computers

Proposes Neural Computers (NCs), a model architecture unifying computation, memory, and I/O as learned runtime state, aiming toward completely neural computing systems.

Ax Yi Xu, Philipp Jettkant, Laura Ruis 20d ago

The Depth Ceiling: On the Limits of Large Language Models in Discovering Latent Planning

Research on limits of latent reasoning in LLMs, testing whether models can discover multi-step planning strategies without supervision using graph path-finding tasks.

Ax Manuel Barusco, Francesco Borsatti, David Petrovic, Davide Dalle Pezze, Gian Antonio Susto 20d ago

Continual Visual Anomaly Detection on the Edge: Benchmark and Efficient Solutions

Benchmark and solutions for visual anomaly detection on edge devices with continual learning to adapt to evolving data distributions.

Ax Manish Bhatt, Sarthak Munshi, Vineeth Sai Narajala, Idan Habler, Ammar Al-Kahfah, Ken Huang, Blake Gatto 20d ago

The Defense Trilemma: Why Prompt Injection Defense Wrappers Fail?

Theoretical proof that no continuous wrapper defense can prevent all prompt injections in LLMs with connected prompt space, characterizing defense failure modes.

Ax Srinidhi Madabhushi, Pranesh Vyas, Swathi Vaidyanathan, Mayur Kurup, Elliott Nash, Yegor Silyutin 20d ago

From Load Tests to Live Streams: Graph Embedding-Based Anomaly Detection in Microservice Architectures

Graph embedding-based anomaly detection system identifies under-represented services in microservice architectures using unsupervised learning.

Ax Mario Iacobelli, Adrian Robert Minut, Tommaso Mencattini, Donato Crisostomi, Andrea Santilli, Iacopo Masi, Emanuele Rodol\`a 20d ago

Multi-objective Evolutionary Merging Enables Efficient Reasoning Models

Multi-objective evolutionary merging approach to reduce computational overhead of reasoning models while maintaining accuracy with fewer tokens.

Ax Afrah Gueriani, Hamza Kheddar, Ahmed Cherif Mazari 20d ago

Hybrid ResNet-1D-BiGRU with Multi-Head Attention for Cyberattack Detection in Industrial IoT Environments

Hybrid ResNet-1D-BiGRU-MHA model for intrusion detection in Industrial IoT systems achieving 98.71% accuracy on EdgeHoTset dataset.

Ax Dev Arpan Desai, Shaoyi Huang, Zining Zhu 20d ago

Distributed Interpretability and Control for Large Language Models

Practical implementation of activation-level interpretability and steering techniques for large language models distributed across multiple GPUs.

Ax David Cho, Yifan Wang, Fanping Sui, Ananth Grama 20d ago

Inference-Time Code Selection via Symbolic Equivalence Partitioning

Symbolic Equivalence Partitioning uses symbolic execution for inference-time code selection in LLM-based code generation without expensive verifiers.

Ax Maojiang Su, Po-Chung Hsieh, Weimin Wu, Mingcheng Lu, Jiunhau Chen, Jerry Yao-Chieh Hu, Han Liu 20d ago

Discrete Flow Matching Policy Optimization

DoMinO framework unifies reinforcement learning fine-tuning of Discrete Flow Matching models by reformulating sampling as a multi-step MDP.

Ax Weiyue Li, Ruizhi Qian, Yi Li, Yongce Li, Yunfan Long, Jiahui Cai, Yan Luo, Mengyu Wang 20d ago

MedConclusion: A Benchmark for Biomedical Conclusion Generation from Structured Abstracts

MedConclusion benchmark dataset of 5.7M PubMed abstracts for evaluating LLMs on biomedical conclusion generation from structured evidence.

Ax Mohammed Nowaz Rabbani Chowdhury, Kaoutar El Maghraoui, Hsinyu Tsai, Naigang Wang, Geoffrey W. Burr, Liu Liu, Meng Wang 20d ago

Efficient Quantization of Mixture-of-Experts with Theoretical Generalization Guarantees

Efficient quantization method for Mixture-of-Experts models with theoretical generalization guarantees to reduce inference memory overhead.

Ax Puja Saha, Eranga Ukwatta 20d ago

Adaptive Differential Privacy for Federated Medical Image Segmentation Across Diverse Modalities

Adaptive differential privacy approach for federated medical image segmentation across diverse imaging modalities and clinical sites.

Ax Basil Kyriacou, Mo Kordzanganeh, Maniraman Periyasamy, Alexey Melnikov 20d ago

Soft-Quantum Algorithms

Soft-Quantum Algorithms explores classical simulation of variational quantum circuits for few-qubit problems with large datasets.

Ax Yinghan Hou, Zongyou Yang 20d ago

SkillSieve: A Hierarchical Triage Framework for Detecting Malicious AI Agent Skills

SkillSieve is a three-layer detection framework for identifying security vulnerabilities in AI agent skills, addressing both code and natural language prompt injection attacks.

Ax Audrey Cheng, Harald Ng, Aaron Kabcenell, Peter Bailis, Matei Zaharia, Lin Ma, Xiao Shi, Ion Stoica 20d ago

AI-Driven Research for Databases

AI-Driven Research for Systems uses LLMs to automate database performance optimization through automated code generation instead of manual design.

Ax Joshua Castillo, Ravi Mukkamala 20d ago

LLM-based Schema-Guided Extraction and Validation of Missing-Person Intelligence from Heterogeneous Data Sources

Guardian Parser Pack uses LLMs to parse and normalize heterogeneous investigative documents for missing-person cases with varying layouts and data quality.

Ax Maotian Ma, Zheni Zeng, Zhenghao Liu, Yukun Yan 20d ago

Scientific Knowledge-driven Decoding Constraints Improving the Reliability of LLMs

SciDC method reduces LLM hallucination by incorporating scientific knowledge and rules as decoding constraints to improve reliability.

Ax Nan Zhang, Zishuo Wang, Shuyu Huang, Georgios Diamantopoulos, Nikos Tziritas, Panagiotis Oikonomou, Georgios Theodoropoulos 20d ago

TwinLoop: Simulation-in-the-Loop Digital Twins for Online Multi-Agent Reinforcement Learning

TwinLoop framework uses simulation-in-the-loop digital twins for online multi-agent reinforcement learning to adapt policies when operating conditions change.

Ax Hanyang Wang, Mingxuan Zhu 20d ago

The Detection--Extraction Gap: Models Know the Answer Before They Can Say It

Research finding that 52-88% of chain-of-thought tokens in reasoning models are generated after the answer is already recoverable, revealing a detection-extraction gap in model behavior.

Ax Mingyu Yang, Wentao Li, Wei Wang 20d ago

CubeGraph: Efficient Retrieval-Augmented Generation for Spatial and Temporal Data

CubeGraph: efficient retrieval-augmented generation system for hybrid queries combining vector similarity search with spatio-temporal filters for RAG workloads.

Ax Evgeny Skvortsov, Yilin Xia, Ojaswa Garg, Shawn Bowers, Bertram Lud\"ascher 20d ago

Logical Robots: Declarative Multi-Agent Programming in Logica

Logical Robots: declarative multi-agent programming platform using logic programming language Logica for robot behavior specification combining reactive control and planning.

Ax Zheng Jiang, Nan He, Yiming Chen, Lifeng Sun 20d ago

SubFLOT: Submodel Extraction for Efficient and Personalized Federated Learning via Optimal Transport

SubFLOT: Federated learning method using optimal transport for efficient submodel extraction, addressing heterogeneity and enabling client-side personalization.

Ax Zhengyang Ai, Zikang Shan, Xiaodong Ai, Jingxian Tang, Hangkai Hu, Pinyan Lu 20d ago

SHAPE: Stage-aware Hierarchical Advantage via Potential Estimation for LLM Reasoning

SHAPE: Framework for improving LLM reasoning through process supervision, formalizing reasoning as state-space trajectory with stage-aware advantage estimation.

Ax Cheng Peng, Mengxian Lyu, Ziyi Chen, Yonghui Wu 20d ago

A Parameter-Efficient Transfer Learning Approach through Multitask Prompt Distillation and Decomposition for Clinical NLP

Parameter-efficient multitask prompt distillation framework for clinical NLP adapting shared metaprompts across diverse medical tasks.

Ax Xiaoyou Qin, Zhihong Li, Xiaoxiao Cheng 20d ago