Despite the fact that deep neural networks are powerful models and achieve appealing results on many tasks, they are too gigantic to be deployed on edge devices like smart-phones or embedded sensor nodes. There has been efforts to compress these networks, and a popular method is knowledge distillation, where a large (a.k.a. teacher) pre-trained network is used to train a smaller (a.k.a. student) network. However, in this paper, we show that the student network performance degrades when the gap between student and teacher is large. Given a fixed student network, one cannot employ an arbitrarily large teacher, or in other words, a teacher can effectively transfer its knowledge to students up to a certain size, not smaller. To alleviate this shortcoming, we introduce multi-step knowledge distillation which employs an intermediate-sized network (a.k.a. teacher assistant) to bridge the gap between the student and the teacher. We study the effect of teacher assistant size and extend the framework to multi-step distillation. Moreover, empirical and theoretical analysis are conducted to analyze the teacher assistant knowledge distillation framework. Extensive experiments on CIFAR-10 and CIFAR-100 datasets and plain CNN and ResNet architectures substantiate the effectiveness of our proposed approach.
Since visual perception can give rich information beyond text descriptions for world understanding, there has been increasing interest in leveraging visual grounding for language learning. Recently, vokenization has attracted attention by using the predictions of a text-to-image retrieval model as labels for language model supervision. Despite its success, the method suffers from approximation error of using finite image labels and the lack of vocabulary diversity of a small image-text dataset. To overcome these limitations, we present VidLanKD, a video-language knowledge distillation method for improving language understanding. We train a multi-modal teacher model on a video-text dataset, and then transfer its knowledge to a student language model with a text dataset. To avoid approximation error, we propose to use different knowledge distillation objectives. In addition, the use of a large-scale video-text dataset helps learn diverse and richer vocabularies. In our experiments, VidLanKD achieves consistent improvements over text-only language models and vokenization models, on several downstream language understanding tasks including GLUE, SQuAD, and SWAG. We also demonstrate the improved world knowledge, physical reasoning, and temporal reasoning capabilities of our model by evaluating on the GLUE-diagnostics, PIQA, and TRACIE datasets. Lastly, we present comprehensive ablation studies as well as visualizations of the learned text-to-video grounding results of our teacher and student language models. Our code and models are available at: https://github.com/zinengtang/VidLanKD
Deep learning based models are relatively large, and it is hard to deploy such models on resource-limited devices such as mobile phones and embedded devices. One possible solution is knowledge distillation whereby a smaller model (student model) is trained by utilizing the information from a larger model (teacher model). In this paper, we present a survey of knowledge distillation techniques applied to deep learning models. To compare the performances of different techniques, we propose a new metric called distillation metric. Distillation metric compares different knowledge distillation algorithms based on sizes and accuracy scores. Based on the survey, some interesting conclusions are drawn and presented in this paper.
To put a state-of-the-art neural network to practical use, it is necessary to design a model that has a good trade-off between the resource consumption and performance on the test set. Many researchers and engineers are developing methods that enable training or designing a model more efficiently. Developing an efficient model includes several strategies such as network architecture search, pruning, quantization, knowledge distillation, utilizing cheap convolution, regularization, and also includes any craft that leads to a better performance-resource trade-off. When combining these technologies together, it would be ideal if one source of performance improvement does not conflict with others. We call this property as the orthogonality in model efficiency. In this paper, we focus on knowledge distillation and demonstrate that knowledge distillation methods are orthogonal to other efficiency-enhancing methods both analytically and empirically. Analytically, we claim that knowledge distillation functions analogous to a ensemble method, bootstrap aggregating. This analytical explanation is provided from the perspective of implicit data augmentation property of knowledge distillation. Empirically, we verify knowledge distillation as a powerful apparatus for practical deployment of efficient neural network, and also introduce ways to integrate it with other methods effectively.
It has been recently demonstrated that multi-generational self-distillation can improve generalization . Despite this intriguing observation, reasons for the enhancement remain poorly understood. In this paper, we first demonstrate experimentally that the improved performance of multi-generational self-distillation is in part associated with the increasing diversity in teacher predictions. With this in mind, we offer a new interpretation for teacher-student training as amortized MAP estimation, such that teacher predictions enable instance-specific regularization. Our framework allows us to theoretically relate self-distillation to label smoothing, a commonly used technique that regularizes predictive uncertainty, and suggests the importance of predictive diversity in addition to predictive uncertainty. We present experimental results using multiple datasets and neural network architectures that, overall, demonstrate the utility of predictive diversity. Finally, we propose a novel instance-specific label smoothing technique that promotes predictive diversity without the need for a separately trained teacher model. We provide an empirical evaluation of the proposed method, which, we find, often outperforms classical label smoothing.