arXiv Machine Learning

TASTE: Task-Aware Out-of-Distribution Detection via Stein Operators

TASTE (Task-Aware STEin operators) is a framework that links distribution shift to a model’s input sensitivity using Stein operators. It provides a geometric interpretation of shift as a projection onto the model’s sensitivity field, offers coordinate‑wise localisation, and delivers interpretable per‑pixel diagnostics for image data. Experiments on Gaussian shifts, MNIST, and CIFAR‑10/ImageNet‑1K benchmarks show that TASTE closely tracks task degradation and performs competitively with existing baselines, with gains varying by shift regime.

arXiv Machine Learning

Oct 19

TASER: Task-Aware Stein Regularisation for Geometry-Driven Smoothness

The paper introduces TASER, a training-time regularisation method that couples neural network derivatives to density scores via the Langevin Stein operator, providing a weighted Sobolev interpretation and a hierarchy of stability bounds. A lighter variant, TASERLight, uses a first-order Stein operator and avoids explicit tangent-space estimation or inner perturbation searches. Experiments on spectroscopy, skeleton motion, ECG, MNIST, and CIFAR-10 show improved robustness to distribution shifts and complementary benefits when combined with isotropic regularisation.

By Micha{\l} Kozyra, Gesine Reinert

arXiv Machine Learning

Jun 9

Need We Teach Foundation Models What is a Generative Image? Gradient-Free Generative Artifact Detection via Analytic Spectral Adaptation

arXiv:2606. 07660v1 Announce Type: cross Abstract: Adapting foundation models to detect generative artifacts via gradient-based updates compromises their intrinsic representations.

By Qiaoyu Chen, Bing Zhang

diffusion fine-tuning safety

arXiv Machine Learning

Aug 4

Generated Images Are Easier to Forget: A Machine Unlearning Perspective for Synthetic Image Detection

arXiv:2608. 00716v1 Announce Type: cross Abstract: Robust detection of generated images is critical to counter the misuse of generative models.

By Jun Nie, Yonggang Zhang, Tongliang Liu, Yiu-ming Cheung, Bo Han, Xinmei Tian

benchmarks

arXiv AI

Aug 20

Prior-Conditioned Gaussian Discriminants for Generalizable AI-generated Image Detection

The paper introduces prior-conditioned Gaussian discriminants as a diagnostic tool for AI-generated image detection. By fitting closed‑form classifier heads from first‑ and second‑order feature statistics, the authors evaluate performance across 39 public datasets (7.1 million images) using the Percept‑Lens protocol. Results show that these Gaussian heads can match or surpass existing detector heads when matched on prior and encoder, highlighting sensitivity to training prior, data efficiency, and representation dependence.

By Shashank Kotyan, Makoto Shing, Yuki Imajuku, Rujikorn Charakorn, Tarin Clanuwat

diffusion

arXiv AI

Jun 24

Zero-Shot Test-Time Canonicalization using Out-of-Distribution Scoring

arXiv:2606. 24178v1 Announce Type: cross Abstract: Pretrained vision models often misclassify inputs that are rotated, scaled, or sheared, even though these affine transformations leave the object class unchanged.

By Dominik Lindner, Johann Schmidt, Tom Siegl, Martin Becker, Sebastian Stober

benchmarks

arXiv Computer Vision

Oct 22

Stochastic Nonlinearities Improve Uncertainty Estimation

The paper introduces stochastic activation switching, a lightweight technique that randomly selects between multiple nonlinear activation functions during inference. This approach generates predictive diversity without disabling intermediate features, enabling uncertainty estimation from a single trained model. Experiments on classification, semantic segmentation, and out‑of‑distribution detection with both convolutional and transformer architectures show improved uncertainty quality while maintaining competitive primary‑task performance.

By Mohammad Sarhangzadeh, Amirreza M. Shebly, Doruk Oner

llms computer-vision

arXiv Machine Learning

Oct 15

Discriminative Score Function: Turning Pretrained Models into Functional Generative Priors

arXiv:2610.19743v1 Announce Type: cross Abstract: Discriminative models are trained with task-specific objectives such as classification, detection, or representation learning, yet their practical ut...

By Junhoo Lee, Hyeonjin Kim, Sangbum Han, Nojun Kwak

efficiency

arXiv AI

Jun 9

LargeMonitor: Monitoring Online Task-Free Continual Learning via Large Pretrained Models

arXiv:2606. 09430v1 Announce Type: cross Abstract: Online task-free continual learning (TFCL) requires intelligent agents to sequentially accumulate knowledge from an unbounded, non-stationary data stream under strict single-pass constraints and without any explicit task identifiers.

By Mingqi Yuan, Xiaoquan Sun, Shihao Luo, Jiayu Chen

agents multimodal benchmarks

arXiv AI

Oct 27

Simple yet Effective Semi-supervised Knowledge Distillation from Vision-Language Models via Dual-Head Optimization

arXiv:2505.07675v3 Announce Type: replace-cross Abstract: Semi-supervised learning (SSL) has emerged as a practical solution for addressing data scarcity challenges by leveraging unlabeled data. Rece...

By Seongjae Kang, Dong Bok Lee, Hyungjoon Jang, Sung Ju Hwang

llms efficiency multimodal benchmarks

arXiv AI

Jun 8

GP-Adapter: Gaussian Process CLIP-Adapter for Few-Shot Out-of-Distribution Detection

arXiv:2606. 07102v1 Announce Type: cross Abstract: We propose GP-Adapter, a training-free framework that augments CLIP (Contrastive Language-Image Pre-training) with Gaussian Process (GP) uncertainty modeling for few-shot classification and out-of-distribution (OOD) detection.

By Taisei Saito, Koretaka Ogata, Takafumi Hiroi

llms rag fine-tuning multimodal benchmarks

arXiv Computer Vision

Sep 3

Uniformity First: Uniformity-aware Test-time Adaptation of Vision-language Models against Image Corruption

The paper introduces UnInfo, a test‑time adaptation method for vision‑language models like CLIP that addresses image corruption—a realistic distribution shift caused by sensor conditions. UnInfo leverages uniformity‑aware confidence maximization, information‑aware loss balancing, and knowledge distillation from an EMA teacher to preserve embedding uniformity and improve zero‑shot classification accuracy. Experiments show that UnInfo outperforms existing TTA methods on corrupted image datasets.

By Kazuki Adachi, Shin'ya Yamaguchi, Tomoki Hamagami

llms rag fine-tuning efficiency multimodal

arXiv AI

Sep 16

Same Answer, Different Representations: Hidden instability in VLMs

arXiv:2602.06652v2 Announce Type: replace Abstract: The robustness of Vision Language Models (VLMs) is commonly assessed through output-level invariance, implicitly assuming that stable predictions r...

By Farooq Ahmad Wani, Alessandro Suglia, Rohit Saxena, Aryo Pradipta Gema, Wai-Chung Kwan, Fazl Barez, Maria Sofia Bucarelli, Fabrizio Silvestri, Pasquale Minervini

llms rag multimodal benchmarks safety

Related stories