Class Incremental Learning for Algorithm Selection

2025-06-02

AI Summary

This work addresses the challenge of dynamic solver category growth and continual model updating in streaming algorithm selection. It introduces class-incremental learning (CIL) to the algorithm selection domain for the first time and systematically investigates its applicability. To mitigate catastrophic forgetting, eight continual learning methods are comparatively evaluated on a bin-packing dataset; rehearsal-based strategies consistently outperform alternatives, exhibiting only ~7% average performance degradation, low forgetting rates, and high stability. The study demonstrates that rehearsal mechanisms—when integrated with an appropriate classification training paradigm—effectively support dynamic expansion of solver sets, preserving prior knowledge while accurately recognizing newly introduced solver categories. This work establishes the first systematic CIL-based solution and empirical benchmark for adaptive algorithm selection in streaming optimization settings.

Technology Category

• Machine Learning: Online Learning & Bandits
• Search and Optimization: Mixed Discrete/Continuous Search
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Algorithm selection is commonly used to predict the best solver from a portfolio per per-instance. In many real scenarios, instances arrive in a stream: new instances become available over time, while the number of class labels can also grow as new data distributions arrive downstream. As a result, the classification model needs to be periodically updated to reflect additional solvers without catastrophic forgetting of past data. In machine-learning (ML), this is referred to as Class Incremental Learning (CIL). While commonly addressed in ML settings, its relevance to algorithm-selection in optimisation has not been previously studied. Using a bin-packing dataset, we benchmark 8 continual learning methods with respect to their ability to withstand catastrophic forgetting. We find that rehearsal-based methods significantly outperform other CIL methods. While there is evidence of forgetting, the loss is small at around 7%. Hence, these methods appear to be a viable approach to continual learning in streaming optimisation scenarios.

Problem

• Addressing catastrophic forgetting in algorithm selection models
• Evaluating continual learning methods for streaming optimization scenarios
• Benchmarking rehearsal-based methods against other CIL approaches

Innovation

• Class Incremental Learning for algorithm selection
• Benchmarks 8 continual learning methods
• Rehearsal-based methods outperform others