A Retention-Centric Framework for Continual Learning with Guaranteed Model Developmental Safety

2024-10-04

AI Summary

In cost-sensitive scenarios involving continual model iteration, catastrophic forgetting undermines the reliability of previously acquired capabilities. Method: This paper proposes a novel paradigm—“model development safety”—requiring new models to not only improve performance on target tasks but also strictly preserve critical capabilities of prior models. To this end, we formulate a capability-preserving continual learning framework, introducing for the first time a data-dependent capability constraint mechanism and an efficient constrained optimization algorithm with theoretical guarantees—overcoming the limitations of conventional average-performance trade-offs. Leveraging the CLIP pre-trained model, we adopt task-specific head fine-tuning and jointly optimize the target-task loss and capability preservation constraints. Results: Experiments on autonomous driving and scene recognition benchmarks demonstrate zero degradation in accuracy on key legacy tasks, significant gains on new tasks, and complete prevention of safety-critical capability deterioration.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Natural Language Processing: Safety and Robustness
• Computer Vision: Large Vision Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

In real-world applications, learning-enabled systems often undergo iterative model development to address challenging or emerging tasks, which involve collecting new data, training a new model and validating the model. This continual model development process raises a significant issue that acquiring new or improving existing capabilities may inadvertently lose good capabilities of the old model, also known as catastrophic forgetting. While existing continual learning aims to mitigate catastrophic forgetting by trading off performance on previous tasks and new tasks to ensure good average performance, it often falls short in cost-sensitive applications, where failing to preserve essential established capabilities introduces unforeseen costs and risks and substantial expenses for re-improving these capabilities. To address this issue, we impose a requirement on learning systems to ensure that a new model strictly retains important capabilities of the old model while improving target-task performance, which we term model developmental safety. To ensure model developmental safety, we propose a retention-centric framework with data-dependent constraints, and study how to continually develop a pretrained CLIP model for acquiring new or improving existing capabilities of image classification. We propose an efficient constrained optimization algorithm with theoretical guarantees and use its insights to finetune the CLIP model with task-dependent heads for promoting the model developmental safety. Experiments on autonomous driving and scene recognition datasets validate the efficacy of our method.

Problem

• Ensures new models retain old model's essential capabilities
• Addresses catastrophic forgetting in continual learning systems
• Improves target-task performance with developmental safety

Innovation

• Retention-centric framework for continual learning
• Data-dependent constraints ensure model safety
• Efficient constrained optimization algorithm