Efficient Backdoor Removal Through Natural Gradient Fine-tuning
This addresses the security vulnerability of backdoor attacks in deep learning models, offering an efficient defense with broad applicability, though it is incremental as it builds on existing fine-tuning and optimization techniques.
The paper tackles the problem of backdoor attacks in deep neural networks by proposing Natural Gradient Fine-tuning (NGF), a method that purifies models by fine-tuning only one layer with a geometry-aware optimizer and regularizer, achieving state-of-the-art performance across four datasets and 13 backdoor attacks.
The success of a deep neural network (DNN) heavily relies on the details of the training scheme; e.g., training data, architectures, hyper-parameters, etc. Recent backdoor attacks suggest that an adversary can take advantage of such training details and compromise the integrity of a DNN. Our studies show that a backdoor model is usually optimized to a bad local minima, i.e. sharper minima as compared to a benign model. Intuitively, a backdoor model can be purified by reoptimizing the model to a smoother minima through fine-tuning with a few clean validation data. However, fine-tuning all DNN parameters often requires huge computational costs and often results in sub-par clean test performance. To address this concern, we propose a novel backdoor purification technique, Natural Gradient Fine-tuning (NGF), which focuses on removing the backdoor by fine-tuning only one layer. Specifically, NGF utilizes a loss surface geometry-aware optimizer that can successfully overcome the challenge of reaching a smooth minima under a one-layer optimization scenario. To enhance the generalization performance of our proposed method, we introduce a clean data distribution-aware regularizer based on the knowledge of loss surface curvature matrix, i.e., Fisher Information Matrix. Extensive experiments show that the proposed method achieves state-of-the-art performance on a wide range of backdoor defense benchmarks: four different datasets- CIFAR10, GTSRB, Tiny-ImageNet, and ImageNet; 13 recent backdoor attacks, e.g. Blend, Dynamic, WaNet, ISSBA, etc.