Knowledge distillation is a method to help learning a new network (student) by using
information of already trained network (teacher). As a new learning method, knowledge
distillation is actively studied for various applications such as network compre...
Knowledge distillation is a method to help learning a new network (student) by using
information of already trained network (teacher). As a new learning method, knowledge
distillation is actively studied for various applications such as network compression.
However, the analysis of knowledge distillation is not sufficient. In this dissertation, we
pursue a comprehensive understanding on the knowledge distillation mechanism and
develop advanced distillation methods for knowledge transfer. This dissertation includes
three distillation technique advances: distillation of decision boundary, distillation of
activation boundary, and comprehensive design of distillation loss.
First, we provide a new perspective based on a decision boundary, which is one
of the most important components of a classifier. The generalization performance of a
classifier is closely related to the adequacy of its decision boundary, so a good classifier
bears a good decision boundary. Therefore, transferring information closely related
to the decision boundary can be a good attempt for knowledge distillation. To realize
this goal, we utilize an adversarial attack to discover samples supporting a decision
boundary. Based on this idea, to transfer more accurate information about the decision
boundary, the proposed algorithm trains a student classifier based on the adversarial
samples supporting the decision boundary. Experiments show that the proposed method
indeed improves knowledge distillation and achieves the state-of-the-arts performance.
Second, based on the idea of decision boundary, we moved from distillation based on
network output to distillation based on hidden layer response. An activation boundary
for a neuron refers to a separating hyperplane that determines whether the neuron
is activated or deactivated. It has been long considered in neural networks that the
activations of neurons, rather than their exact output values, play the most important
role in forming classification-friendly partitions of the hidden feature space. However, as
far as we know, this aspect of neural networks has not been considered in the literature
of knowledge transfer. We propose a knowledge transfer method via distillation of
activation boundaries formed by hidden neurons. For the distillation, we propose an
activation transfer loss that has the minimum value when the boundaries generated by
the student coincide with those by the teacher. Since the activation transfer loss is not
differentiable, we design a piecewise differentiable loss approximating the activation
transfer loss. By the proposed method, the student learns a separating boundary between
activation region and deactivation region formed by each neuron in the teacher. Through
the experiments in various aspects of knowledge transfer, it is verified that the proposed
method outperforms the current state-of-the-art.
Lastly, we investigate the design aspects of feature distillation methods achieving
network compression and propose a novel feature distillation method in which the
distillation loss is designed to make a synergy among various aspects: teacher transform,
student transform, distillation feature position and distance function. Our proposed
distillation loss includes a feature transform with a newly designed margin ReLU, a
new distillation feature position, and a partial L2 distance function to skip redundant
information giving adverse effects to the compression of student. In ImageNet, our
proposed method achieves 21.65% of top-1 error with ResNet50, which outperforms the
performance of the teacher network, ResNet152. Our proposed method is evaluated on
various tasks such as image classification, object detection and semantic segmentation
and achieves a significant performance improvement in all tasks