When Sensing Varies with Contexts Context Probing for Tactile Few-Shot Class-Incremental Learning

Jul 16, 2026

Few-shot class-incremental learning (FSCIL) must recognize novel tactile classes from a few labeled samples while retaining previously learned knowledge. In tactile sensing, the same material can yield different observations across acquisition contexts such as sensing devices, contact states, scanning trajectories, and interaction conditions. These contexts bias the few-shot support prototypes, shifting classifier decision boundaries in later sessions. CoP-FSCIL addresses this with context probing, quotient adaptation, and probe-stability prototype calibration.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages14

Reading Time35 min

Abstract

Introduction

When Sensing Varies with Contexts: Context Probing for Tactile Few-Shot Class-Incremental Learning Yifeng Lin, Aiping Huang, Wenxi Liu, Senior Member, IEEE, Si Wu, Member, IEEE, Tiesong Zhao, Senior Member, IEEE, Zechao Li, Senior Member, IEEE, and Zheng-jun Zha, Member, IEEE arXiv :2603 .25 1 15v2 [ cs .AI] 13 Jul 2026 Abstract—Few-shot class-incremental learning (FSCIL) aims to recognize novel classes from only a few labeled samples while retaining previously learned knowledge. Although recent FSCIL methods have achieved substantial progress on visual benchmarks, they remain limited in tactile sensing, where the same material may produce markedly different observations under different acquisition contexts, such as sensing devices, contact states, scanning trajectories, and interaction conditions. In tactile FSCIL, the challenges of few-shot learning and classincremental learning are further amplified by acquisition context: the limited support samples may not only be scarce, but also carry context-induced biases. Once the resulting biased prototypes are inserted into the classifier, they may affect the decision boundaries in subsequent sessions. To address this problem, we propose Context-Probing Few-Shot Class-Incremental Learning (CoP-FSCIL), a context-aware framework for tactile FSCIL. CoP-FSCIL first employs Context-Probing Intervention (CPI) to diagnose local context-sensitive variations in tactile representations. It then introduces a Probe-Conditioned Quotient Adapter (PCQA) to suppress context-sensitive components identified by the probes. Finally, Probe-Stability Prototype Calibration (PSPC) estimates support sample reliability from probe-induced embedding fluctuations and calibrates stochastic prototypes accordingly. Experiments on HapTex and LMT108 show that CoP-FSCIL consistently outperforms representative FSCIL baselines, and extended experiments on audio FSCIL further demonstrate the generality of the proposed context probing mechanism. The source code is currently being prepared and will be released soon. Index Terms—Few-shot class-incremental learning, Tactile sensing, Acquisition context I. INTRODUCTION REAL-WORLD intelligent systems rarely operate under a fixed and fully annotated class taxonomy. New categories may appear over time, while collecting sufficient labeled data This work is supported by National Natural Science Foundation of China (Grant No. 62571131) . Corresponding author: Tiesong Zhao. Yifeng Lin, Aiping Huang and Tiesong Zhao are with the Fujian Key Laboratory for Intelligent Processing and Wireless Transmission of Media Information, College of Physics and Information Engineering, Fuzhou University, Fuzhou 350108, China. (e-mails: [241120007@fzu.edu.cn](241120007@fzu.edu.cn), [sxxhap@163.com](sxxhap@163.com), [t.zhao@fzu.edu.cn](t.zhao@fzu.edu.cn)) . Wenxi Liu is with the College of Computer and Data Science, Fuzhou University, Fuzhou 350108, China. ([e-mail: wenxi.liu@hotmail.com](e-mail: wenxi.liu@hotmail.com)). Si Wu is with the School of Computer Science and Engineering, South China University of Technology, Guangdong, China (e-mail: [cswusi@scut.edu.cn](cswusi@scut.edu.cn)) . Zechao Li is with School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China. (e-mail: [zechao.li@njust.edu.cn](zechao.li@njust.edu.cn)). Zheng-jun Zha is with the School of Information Science and Technology, University of Science and Technology of China, Hefei 230027, China. (e-mail: [zhazj@ustc.edu.cn](zhazj@ustc.edu.cn)). Fig. 1. (a) Acquisition context entangles with target features and degrades FSCIL through two coupled failure modes. Context Overfitting occurs when different classes are covered by similar contexts, causing prototypes to encode context cues and even yielding unreliable features that are difficult to recover. Context Variation Failure occurs when an unseen context is matched to the closest observed one, resultin

FAQs

What makes tactile few-shot class-incremental learning harder than visual FSCIL?

Tactile observations are tightly coupled to the acquisition process, so the same material can produce very different responses under different sensing devices, contact states, and trajectories, creating context-induced biases.

How does CoP-FSCIL diagnose context-sensitive variations in tactile representations?

CoP-FSCIL uses Context-Probing Intervention (CPI) to identify local context-sensitive components in the tactile embedding space.

What are the key components of the CoP-FSCIL framework and what do they do?

CoP-FSCIL applies CPI for diagnosis, Probe-Conditioned Quotient Adapter (PCQA) to suppress probe-identified context-sensitive components, and Probe-Stability Prototype Calibration (PSPC) to estimate support reliability via embedding fluctuations and calibrate stochastic prototypes.

Jul 16, 2026

Upload by Olivia Brown