When Sensing Varies with Contexts: Context-as-Transform for Tactile Few-Shot Class-Incremental Learning

arXiv cs.AI / 3/27/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper addresses how Few-Shot Class-Incremental Learning (FSCIL) performance degrades in tactile sensing because acquisition contexts (devices, contact states, interaction settings) are not standardized.
It proposes Context-as-Transform FSCIL (CaT-FSCIL), decomposing context into a structured low-dimensional part and a high-dimensional residual, then handling each with different mechanisms.
For the low-dimensional component, it models tactile interaction effects using an approximately invertible Context-as-Transform family and performs inverse-transform canonicalization with a pseudo-context consistency loss.
For the high-dimensional residual (platform/device differences), it introduces Uncertainty-Conditioned Prototype Calibration (UCPC) to adjust prototypes and decision boundaries based on context uncertainty.
Experiments on HapTex and LMT108 show CaT-FSCIL outperforms existing approaches on the tactile FSCIL benchmarks, indicating the proposed context modeling and calibration improve robustness.

Abstract

Few-Shot Class-Incremental Learning (FSCIL) can be particularly susceptible to acquisition contexts with only a few labeled samples. A typical scenario is tactile sensing, where the acquisition context ({\it e.g.}, diverse devices, contact state, and interaction settings) degrades performance due to a lack of standardization. In this paper, we propose Context-as-Transform FSCIL (CaT-FSCIL) to tackle the above problem. We decompose the acquisition context into a structured low-dimensional component and a high-dimensional residual component. The former can be easily affected by tactile interaction features, which are modeled as an approximately invertible Context-as-Transform family and handled via inverse-transform canonicalization optimized with a pseudo-context consistency loss. The latter mainly arises from platform and device differences, which can be mitigated with an Uncertainty-Conditioned Prototype Calibration (UCPC) that calibrates biased prototypes and decision boundaries based on context uncertainty. Comprehensive experiments on the standard benchmarks HapTex and LMT108 have demonstrated the superiority of the proposed CaT-FSCIL.