Graph-based Knowledge Distillation by Multi-head Attention Network
This addresses the need for more effective knowledge distillation in machine learning, particularly for improving small student networks, and is incremental as it builds on existing methods by focusing on dataset units rather than data units.
The paper tackles the problem of knowledge distillation by proposing a method to distill dataset-based knowledge using a multi-head attention network, achieving a 7.05% performance improvement over the student network alone on CIFAR100, which is 2.46% higher than the state-of-the-art.
Knowledge distillation (KD) is a technique to derive optimal performance from a small student network (SN) by distilling knowledge of a large teacher network (TN) and transferring the distilled knowledge to the small SN. Since a role of convolutional neural network (CNN) in KD is to embed a dataset so as to perform a given task well, it is very important to acquire knowledge that considers intra-data relations. Conventional KD methods have concentrated on distilling knowledge in data units. To our knowledge, any KD methods for distilling information in dataset units have not yet been proposed. Therefore, this paper proposes a novel method that enables distillation of dataset-based knowledge from the TN using an attention network. The knowledge of the embedding procedure of the TN is distilled to graph by multi-head attention (MHA), and multi-task learning is performed to give relational inductive bias to the SN. The MHA can provide clear information about the source dataset, which can greatly improves the performance of the SN. Experimental results show that the proposed method is 7.05% higher than the SN alone for CIFAR100, which is 2.46% higher than the state-of-the-art.