Deep repurposing adapts neural networks without retraining after task changes
Repurposing Obsolete Representations for Post-Deployment Adaptation
Machine Learning
Summary
Sometimes AI models need to stop using outdated parts of what they learned without having to retrain everything, which can be slow and costly. The authors propose a way called Deep Repurposing that cleverly adjusts a model’s internal knowledge to ignore obsolete outputs while keeping useful skills intact. This approach works without retraining the model and runs much faster than other methods. It helps AI systems stay effective even when their tasks change after they have already been deployed.
What this means in practice
- •For machine learning engineers: Adapt deployed AI models to new tasks quickly without expensive retraining by removing outdated outputs and reallocating useful internal features.
- •For software maintenance teams: Maintain long-running AI applications by updating models to reflect changed requirements without downtime due to full retraining cycles.
Authors
Daniel Bethell, Charmaine Barker, Simos Gerasimou
Abstract
Deep neural networks are increasingly deployed in long-lived systems, where task requirements may change after training. In such settings, part of the original output space may become obsolete: a class, prediction region, or learned behaviour may no longer be valid. Existing approaches either leave the obsolete behaviour intact or require fine-tuning, which can be expensive. We propose Deep Repurposing (DR), a post-hoc framework for adapting models under task obsolescence. DR estimates the latent geometry of obsolete and retained regions, removes obsolete-supporting components, and reallocates retained-compatible evidence through an analytic repair map without gradient updates. This yields repaired predictions and representations in which obsolete regions no longer act as valid outputs, while useful obsolete structure can support the retained task. Across multiple task settings, DR removes obsolete behaviour while preserving retained utility. More importantly, across classification benchmarks, DR matches or exceeds competing unlearning and editing baselines in retained accuracy, eliminates obsolete predictions, and adapts up to $60\times$ faster than competing unlearning methods.