Isolater - Feed

Ax Kadir-Kaan \"Ozer, Ren\'e Ebeling, Markus Enzweiler 3/16/2026

Surprised by Attention: Predictable Query Dynamics for Time Series Anomaly Detection

AxonAD detects time series anomalies via attention query predictability, identifying coordination shifts missed by residual detectors.

Ax Md Anwar Hossen, Nathan R. Tallent, Luanzheng Guo, Ali Jannesary 3/16/2026

SCOPE: Semantic Coreset with Orthogonal Projection Embeddings for Federated learning

SCOPE semantic coreset selection for federated learning on imbalanced high-resolution instrument data using orthogonal projections.

Ax Yijun Quan, Wentai Wu, Giovanni Montana 3/16/2026

Exact Federated Continual Unlearning for Ridge Heads on Frozen Foundation Models

Exact federated unlearning method for removing sample/user influence from frozen foundation models with trainable heads.

Ax Bumjun Kim, Dongjae Jeon, Moongyu Jeon, Albert No 3/16/2026

Dependency-Aware Parallel Decoding via Attention for Diffusion LLMs

DAPD training-free decoding method for diffusion LLMs using attention to model inter-token dependencies during parallel decoding.

Ax Chen Feng, Zhuo Zhi, Zhao Huang, Jiawei Ge, Ling Xiao, Nicu Sebe, Georgios Tzimiropoulos, Ioannis Patras 3/16/2026

Deconstructing the Failure of Ideal Noise Correction: A Three-Pillar Diagnosis

Diagnoses why noise transition matrix methods underperform empirical sample selection in learning with noisy labels.

Ax Chenlong Yin, Runpeng Geng, Yanting Wang, Jinyuan Jia 3/16/2026

PISmith: Reinforcement Learning-based Red Teaming for Prompt Injection Defenses

PISmith RL-based red teaming framework for systematically evaluating robustness of prompt injection defenses in LLM applications.

Ax Michael Scholkemper, Sach Mukherjee 3/16/2026

Causal Cellular Context Transfer Learning (C3TL): An Efficient Architecture for Prediction of Unseen Perturbation Effects

C3TL enables efficient prediction of chemical and genetic perturbation effects on cell states using causal transfer learning.

Ax Ephrem Tibebe Mekonnen, Luca Longo, Lucas Rizzo, Pierpaolo Dondio 3/16/2026

L2GTX: From Local to Global Time Series Explanations

L2GTX explains time series classification decisions through local-to-global explanation synthesis respecting temporal dependencies.

Ax Yihao Ding, Yiran Zhang, Chris Gonzalez, Eun-Jung Holden, Wei Liu 3/16/2026

GeoChemAD: Benchmarking Unsupervised Geochemical Anomaly Detection for Mineral Exploration

GeoChemAD open-source benchmark for unsupervised geochemical anomaly detection in mineral exploration across multiple regions.

Ax Ann Dooms 3/16/2026

Fractals made Practical: Denoising Diffusion as Partitioned Iterated Function Systems

Interprets diffusion models as partitioned iterated function systems, deriving geometric quantities for unified design of denoising schedules and architectures.

Ax Jose Marie Antonio Mi\~noza, Paulo Mario P. Medina, Sebastian C. Iba\~nez 3/16/2026

Influence Malleability in Linearized Attention: Dual Implications of Non-Convergent NTK Dynamics

Analyzes learning dynamics of linearized attention mechanisms through neural tangent kernel framework, revealing trade-offs in attention-based models.

Ax Denis Huseljic, Paul Hahn, Marek Herde, Christoph Sandrock, Bernhard Sick 3/16/2026

BoSS: A Best-of-Strategies Selector as an Oracle for Deep Active Learning

BoSS develops an oracle-based selector for active learning strategies to improve robustness across different models and datasets.

Ax Jie Ji, Gen Li, Kaiyuan Deng, Fatemeh Afghah, Xiaolong Ma 3/16/2026

ZO-SAM: Zero-Order Sharpness-Aware Minimization for Efficient Sparse Training

ZO-SAM proposes zero-order sharpness-aware minimization for sparse neural network training, reducing computational costs in resource-constrained environments.

Ax Callum McLean, Luke Y. Prince, Alexandre Payot, Paul Balan\c{c}a, Carlo Luschi 3/16/2026

MXNorm: Reusing MXFP block scales for efficient tensor normalisation

MXNorm technique reuses MXFP block scales for efficient tensor normalization in low-precision deep learning accelerators.

Ax Xingli Fang, Jung-Eun Kim 3/16/2026

Learnability and Privacy Vulnerability are Entangled in a Few Critical Weights

Research showing privacy vulnerability concentrates in small fraction of neural network weights that also critically impact model utility, enabling targeted privacy interventions.

Ax Helen Qu, Rudy Morel, Michael McCabe, Alberto Bietti, Fran\c{c}ois Lanusse, Shirley Ho, Yann LeCun 3/16/2026

Representation Learning for Spatiotemporal Physical Systems

Proposes representation learning approach for spatiotemporal physical systems as alternative to next-frame prediction, focusing on downstream scientific tasks.

Ax Parth Patne, Mahdi Taheri, Christian Herglotz, Maksim Jenihhin, Milos Krstic, Michael H\"ubner 3/16/2026

DART: Input-Difficulty-AwaRe Adaptive Threshold for Early-Exit DNNs

Framework for adaptive early-exit in neural networks that adjusts confidence thresholds based on input difficulty to optimize inference cost on edge devices.

Ax Boyu Qiao, Sean Guo, Xian Yang, Kun Li, Wei Zhou, Songlin Hu, Yunya Song 3/16/2026

Diagnosing Retrieval Bias Under Multiple In-Context Knowledge Updates in Large Language Models

Analysis of retrieval bias in LLMs when multiple conflicting facts are provided in context, exploring how models resolve competition between historically valid knowledge versions.

Ax Wenwen Hou, Xinyuan Song, Shiwei Liu 3/16/2026

ActTail: Global Activation Sparsity in Large Language Models

ActTail proposes magnitude-based activation sparsity for LLM inference acceleration by adapting sparsity levels across heterogeneous Transformer projections rather than applying uniform sparsity.

Ax Thomas Kleine Buening, Jonas H\"ubotter, Barna P\'asztor, Idan Shenfeld, Giorgia Ramponi, Andreas Krause 3/16/2026

Aligning Language Models from User Interactions

Method for training LLMs from multi-turn user interactions by leveraging follow-up messages as implicit feedback signals about response quality and alignment.

Ax Md Tanvir Hasan Turja 3/16/2026

Predictive Analytics for Foot Ulcers Using Time-Series Temperature and Pressure Data

Predictive analytics using wearable sensor timeseries data for early detection of diabetic foot ulcers.

Ax Yiwen Song, Tomas Pfister, Yale Song 3/16/2026

VQQA: An Agentic Approach for Video Evaluation and Quality Improvement

VQQA multi-agent framework for evaluating and improving video generation quality across diverse tasks.

Ax Haining Pan, Nakul Aggarwal, J. H. Pixley 3/16/2026

Pruning-induced phases in fully-connected neural networks: the eumentia, the dementia, and the amentia

Studies phase transitions induced by pruning in fully-connected networks and their universality classes.

Ax Tianhao Qian, Zhuoxuan Li, Jinde Cao, Xinli Shi, Hanjie Liu, Leszek Rutkowski 3/16/2026

Alternating Gradient Flow Utility: A Unified Metric for Structural Pruning and Dynamic Routing in Deep Networks

Alternating Gradient Flow Utility metric for structural pruning and dynamic routing in neural networks.

Ax Yulin Li, Tengyao Tu, Li Ding, Junjie Wang, Huiling Zhen, Yixin Chen, Yong Li, Zhuotao Tian 3/16/2026

Efficient Reasoning with Balanced Thinking

Method to improve efficiency of Large Reasoning Models by balancing computation across reasoning problems.

Ax Zheda Mai, Ke Zhang, Fu-En Wang, Zixiao Ken Wang, Albert Y. C. Chen, Lu Xia, Min Sun, Wei-Lun Chao, Cheng-Hao Kuo 3/16/2026

Revisiting Model Stitching In the Foundation Model Era

Revisits model stitching technique for Vision Foundation Models to probe representational compatibility across different models.

Ax Nina Wiedemann, Quentin Leboutet, Michael Paulitsch, Diana Wofk, Benjamin Ummenhofer 3/16/2026

KernelFoundry: Hardware-aware evolutionary GPU kernel optimization

KernelFoundry uses LLMs with hardware awareness for evolutionary GPU kernel optimization and code generation.

Ax Prabhu Vellaisamy, Shreesh Tripathi, Vignesh Natarajan, Surya Santhan Thenarasu, Shawn Blanton, John P. Shen 3/16/2026

TaxBreak: Unmasking the Hidden Costs of LLM Inference Through Overhead Decomposition

TaxBreak methodology for decomposing and analyzing host-side overhead costs in LLM inference for latency-sensitive systems.

Ax Rujie Wu, Haozhe Zhao, Hai Ci, Yizhou Wang 3/16/2026

Less Data, Faster Convergence: Goal-Driven Data Optimization for Multimodal Instruction Tuning

Goal-Driven Data Optimization framework for efficient multimodal instruction tuning with reduced training data requirements.

Ax Aadyaa Maddi, Prakhar Naval, Deepti Mande, Shane Duan, Muckai Girish, Vyas Sekar 3/16/2026

Generating Expressive and Customizable Evals for Timeseries Data Analysis Agents with AgentFuel

AgentFuel framework for generating evaluations of timeseries data analysis agents across IoT, observability, and analytics domains.

Ax Joong Ho Kim, Nicholas Thai, Souhardya Saha Dip, Dong Lao, Keith G. Mills 3/16/2026

Na\"ive PAINE: Lightweight Text-to-Image Generation Improvement with Prompt Evaluation

Method to improve text-to-image generation quality through prompt evaluation without additional inference.

Ax Selim Furkan Tekin, Yichang Xu, Gaowen Liu, Ramana Rao Kompella, Margaret L. Loper, Ling Liu 3/16/2026

Vision Verification Enhanced Fusion of VLMs for Efficient Visual Reasoning

Multi-VLM ensemble method using vision and language modalities to select complementary models for efficient visual reasoning.

Ax Sihao Ding 3/16/2026

Colluding LoRA: A Composite Attack on LLM Safety Alignment

Composite attack on LLM safety alignment where multiple LoRA adapters appear benign individually but suppress safety when composed.

Ax Nandish Chattopadhyay, Anadi Goyal, Chandan Karfa, Anupam Chattopadhyay 3/16/2026

STRAP-ViT: Segregated Tokens with Randomized -- Transformations for Defense against Adversarial Patches in ViTs

Defense mechanism against adversarial patches in Vision Transformers using token segregation and randomized transformations.

Ax Chaojie Sun, Bin Cao, Tiantian Li, Chenyu Hou, Ruizhe Li, Qing Fan 3/16/2026

FGTR: Fine-Grained Multi-Table Retrieval via Hierarchical LLM Reasoning

Hierarchical LLM-based approach for fine-grained multi-table retrieval using compositional reasoning instead of coarse-grained similarity matching.

Ax Yali Du, San-Zhuo Xi, Hui Sun, Ming Li 3/16/2026

Design-Specification Tiling for ICL-based CAD Code Generation

In-context learning strategy for CAD code generation using design-specification tiling to improve LLM performance on domain-specific tasks.

Ax Jillur Rahman Saurav, Thuong Le Hoai Pham, Pritam Mukherjee, Paul Yi, Brent A. Orr, Jacob M. Luber 3/16/2026

UNIStainNet: Foundation-Model-Guided Virtual Staining of H&E to IHC

Foundation model-guided approach for virtual immunohistochemistry staining from H&E images to accelerate pathology diagnostics.

Ax Yonghun Jeong, David Yoon Suk Kang, Yeon-Chang Lee 3/16/2026

Anchored Alignment: Preventing Positional Collapse in Multimodal Recommender Systems

Multimodal recommendation framework using anchor-based alignment in projection space to prevent modality collapse and ID dominance.

Ax Yuchen Hua, Xingang Peng, Jianzhu Ma, Muhan Zhang 3/16/2026

VecMol: Vector-Field Representations for 3D Molecule Generation

Novel 3D molecule generation framework using vector-field representations to address modality entanglement and geometry-chemistry constraints.

Ax James Akl, Jose Nicolas Avendano Arbelaez, James Barabas, Jennifer L. Barry, Kalie Ching, Noam Eshed, Jiahui Fu, Michel Hidalgo, Andrew Hoelscher, Tushar Kusnur, Andrew Messing, Zachary Nagler, Brian Okorn, Mauro Passerino, Tim J. Perkins, Eric Rosen, Ankit Shah, Tanmay Shankar, Scott Shaw 3/16/2026

Show, Don't Tell: Detecting Novel Objects by Watching Human Videos

Self-supervised system for robots to detect and recognize novel objects from human video demonstrations without prompt engineering.

Ax Jiaming Zhang, Yuyuan Li, Xiaohua Feng, Li Zhang, Longfei Li, Jun Zhou, Chaochao Chen 3/16/2026

Taming the Long Tail: Efficient Item-wise Sharpness-Aware Minimization for LLM-based Recommender Systems

Efficient optimization technique addressing long-tail distribution problem in LLM-based sequential recommender systems.

Ax Nazar Puriy, Johannes Jakubik, Benedikt Blumenstiel, Konrad Schindler 3/16/2026

TerraFlow: Multimodal, Multitemporal Representation Learning for Earth Observation

Multimodal foundation model for Earth observation using temporal training objectives robust to variable-length satellite and sensor data.

Ax Zezhou Zhang, Songxin Zhang, Xiao Xiong, Junjie Zhang, Zejian Xie, Jingyi Xi, Zunyao Mao, Zan Mao, Zhixin Mai, Zhuoyang Song, Jiaxing Zhang 3/16/2026

PVI: Plug-in Visual Injection for Vision-Language-Action Models

Method for injecting auxiliary visual features into vision-language-action models to improve geometric understanding and temporal reasoning for robotic manipulation.

Ax Zhuchenyang Liu, Yao Zhang, Yu Xiao 3/16/2026

NanoVDR: Distilling a 2B Vision-Language Retriever into a 70M Text-Only Encoder for Visual Document Retrieval

Model distillation approach compressing 2B vision-language retriever into 70M text-only encoder for efficient document retrieval.

Ax You Wu, Zhenguo Wang, Naiyu Wang 3/16/2026

From AI Weather Prediction to Infrastructure Resilience: A Correction-Downscaling Framework for Tropical Cyclone Impacts

AI-based framework transforming global weather forecasts into fine-grained wind field predictions and infrastructure failure probabilities for tropical cyclones.

Ax Shaofeng Guo, Jiequan Cui, Richang Hong 3/16/2026

Rethinking VLMs for Image Forgery Detection and Localization

Research on using vision-language models for detecting and localizing forged images, studying how VLM priors affect forgery detection performance.

Ax Yamin Arefeen, Sidharth Kumar, Steven Warach, Hamidreza Saber, Jonathan Tamir 3/16/2026

Accelerating Stroke MRI with Diffusion Probabilistic Models through Large-Scale Pre-training and Target-Specific Fine-Tuning

Data-efficient MRI reconstruction strategy using diffusion probabilistic models with pre-training and fine-tuning.

Ax Matthew Butler, Yi Fan, Christos Faloutsos 3/16/2026