New method improves remembering many tasks in ai systems
Distribution-Conditioned Task Routing for Class-Incremental Learning
Machine Learning
Summary
Class-incremental learning means teaching AI systems to recognize many classes over time without forgetting old ones and without knowing which task a new input belongs to. The authors focus on a setting where AI uses small, task-specific model updates but needs to decide which task module to use during inference without retraining or extra training. They identify three sources of error in guessing the right task and propose Feature Distribution Calibration (FDC) to fix them by filtering task features, checking residuals, and comparing to class prototypes. Their approach improves accuracy consistently across many benchmarks and methods without changing the original models.
What this means in practice
- •For machine learning engineers: Improve AI systems that need to recognize classes added over time without retraining or knowing task identity in advance.
- •For edge device developers: Enhance AI models on limited hardware by improving task inference accuracy with small task-specific updates and no extra training.
Authors
Longhuan Xu, Zhipeng Zhou, Wei Ji, Chunyan Miao, Peilin Zhao, Lijun Zhang
Abstract
Parameter-efficient adaptation enables continual learners to acquire task-specific knowledge through compact model updates while maintaining strong within-task performance. However, class-incremental inference requires each input to be classified among all classes seen so far without access to its task identity. For learners equipped with task-specific parameter-efficient modules, this introduces a critical task-routing challenge beyond catastrophic forgetting. We study post-hoc task routing without retraining the learner or introducing a separately trained router. Such training-free inference-time calibration remains comparatively underexplored in parameter-efficient class-incremental learning. We identify three sources of routing error (feature-level, task-level, and class-level misalignment) and propose Feature Distribution Calibration (FDC). Its three components address these misalignments: Task Subspace Filtering (TSF) suppresses feature components outside each task's principal subspace, Residual Likelihood Calibration (RLC) evaluates the typicality of its subspace residual, and Prototype Affinity Calibration (PAC) measures compatibility with the task's class prototypes. Experiments demonstrate plug-and-play applicability to eight parameter-efficient class-incremental methods using a shared encoder. With one component configuration selected per method across all five benchmarks, FDC improves final accuracy in all 40 method-dataset pairs by 4.39 percentage points on average. Enabling all components improves 35 of the 40 pairs, with an average gain of 4.45 points. When applied to a simple baseline, FDC achieves strong overall performance.