Mitigating Negative Flips via Margin Preserving Training

2025-11-11

AI Summary

In continual learning for AI systems, introducing new classes often compresses decision boundaries of original classes, causing negative forgetting—where previously correctly classified samples become misclassified. To address this, we propose a boundary-aware co-optimization method. Our approach features: (1) an explicit inter-class boundary calibration term in the logits space to preserve the discriminative structure of original classes; and (2) a dual-source focal distillation loss that jointly leverages predictions from the old model and ground-truth labels, dynamically balancing learning priorities between old and new classes. The method operates without storing old-class data and supports incremental class expansion. Evaluated on multiple image classification benchmarks, it reduces negative forgetting by 38.2% on average while maintaining overall accuracy comparable to independent training—demonstrating effectiveness, efficiency, and robust generalization.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Minimizing inconsistencies across successive versions of an AI system is as crucial as reducing the overall error. In image classification, such inconsistencies manifest as negative flips, where an updated model misclassifies test samples that were previously classified correctly. This issue becomes increasingly pronounced as the number of training classes grows over time, since adding new categories reduces the margin of each class and may introduce conflicting patterns that undermine their learning process, thereby degrading performance on the original subset. To mitigate negative flips, we propose a novel approach that preserves the margins of the original model while learning an improved one. Our method encourages a larger relative margin between the previously learned and newly introduced classes by introducing an explicit margin-calibration term on the logits. However, overly constraining the logit margin for the new classes can significantly degrade their accuracy compared to a new independently trained model. To address this, we integrate a double-source focal distillation loss with the previous model and a new independently trained model, learning an appropriate decision margin from both old and new data, even under a logit margin calibration. Extensive experiments on image classification benchmarks demonstrate that our approach consistently reduces the negative flip rate with high overall accuracy.

Problem

• Mitigating negative flips in updated AI image classification models
• Preserving original class margins when adding new training categories
• Balancing margin calibration with accuracy for new classes

Innovation

• Margin preserving training for model updates
• Double-source focal distillation loss integration
• Logit margin calibration to mitigate negative flips