{"ID":2854617,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2510.14634","arxiv_id":"2510.14634","title":"SteeringTTA: Guiding Diffusion Trajectories for Robust Test-Time-Adaptation","abstract":"Test-time adaptation (TTA) aims to correct performance degradation of deep models under distribution shifts by updating models or inputs using unlabeled test data. Input-only diffusion-based TTA methods improve robustness for classification to corruptions but rely on gradient guidance, limiting exploration and generalization across distortion types. We propose SteeringTTA, an inference-only framework that adapts Feynman-Kac steering to guide diffusion-based input adaptation for classification with rewards driven by pseudo-label. SteeringTTA maintains multiple particle trajectories, steered by a combination of cumulative top-K probabilities and an entropy schedule, to balance exploration and confidence. On ImageNet-C, SteeringTTA consistently outperforms the baseline without any model updates or source data.","short_abstract":"Test-time adaptation (TTA) aims to correct performance degradation of deep models under distribution shifts by updating models or inputs using unlabeled test data. Input-only diffusion-based TTA methods improve robustness for classification to corruptions but rely on gradient guidance, limiting exploration and generali...","url_abs":"https://arxiv.org/abs/2510.14634","url_pdf":"https://arxiv.org/pdf/2510.14634v1","authors":"[\"Jihyun Yu\",\"Yoojin Oh\",\"Wonho Bae\",\"Mingyu Kim\",\"Junhyug Noh\"]","published":"2025-10-16T12:46:53Z","proceeding":"cs.CV","tasks":"[\"cs.CV\"]","methods":"[\"Diffusion Model\",\"LoRA\"]","has_code":false}