Knowledge Distillation (KD) is a prominent neural model compression technique that heavily relies on teacher network predictions to guide the training of a student model. Considering the ever-growing size of pre-trained language models (PLMs), KD is often adopted in many NLP tasks involving PLMs. However, it is evident that in KD, deploying the teacher network during training adds to the memory and computational requirements of training. In the computer vision literature, the necessity of the teacher network is put under scrutiny by showing that KD is a label regularization technique that can be replaced with lighter teacher-free variants such as the label-smoothing technique. However, to the best of our knowledge, this issue is not investigated in NLP. Therefore, this work concerns studying different label regularization techniques and whether we actually need them to improve the fine-tuning of smaller PLM networks on downstream tasks. In this regard, we did a comprehensive set of experiments on different PLMs such as BERT, RoBERTa, and GPT with more than 600 distinct trials and ran each configuration five times. This investigation led to a surprising observation that KD and other label regularization techniques do not play any meaningful role over regular fine-tuning when the student model is pre-trained. We further explore this phenomenon in different settings of NLP and computer vision tasks and demonstrate that pre-training itself acts as a kind of regularization, and additional label regularization is unnecessary.

When multiple conversations occur simultaneously, a listener must decide which conversation each utterance is part of in order to interpret and respond to it appropriately. This task is referred as dialogue disentanglement. A significant drawback of previous studies on disentanglement lies in that they only focus on pair-wise relationships between utterances while neglecting the conversation structure which is important for conversation structure modeling. In this paper, we propose a hierarchical model, named Dialogue BERT (DIALBERT), which integrates the local and global semantics in the context range by using BERT to encode each message-pair and using BiLSTM to aggregate the chronological context information into the output of BERT. In order to integrate the conversation structure information into the model, two types of loss of conversation-structure loss and tree-structure loss are designed. In this way, our model can implicitly learn and leverage the conversation structures without being restricted to the lack of explicit access to such structures during the inference stage. Experimental results on two large datasets show that our method outperforms previous methods by substantial margins, achieving great performance on dialogue disentanglement.

Knowledge Distillation (KD) is a model compression algorithm that helps transfer the knowledge in a large neural network into a smaller one. Even though KD has shown promise on a wide range of Natural Language Processing (NLP) applications, little is understood about how one KD algorithm compares to another and whether these approaches can be complimentary to each other. In this work, we evaluate various KD algorithms on in-domain, out-of-domain and adversarial testing. We propose a framework to assess adversarial robustness of multiple KD algorithms. Moreover, we introduce a new KD algorithm, Combined-KD, which takes advantage of two promising approaches (better training scheme and more efficient data augmentation). Our extensive experimental results show that Combined-KD achieves state-of-the-art results on the GLUE benchmark, out-of-domain generalization, and adversarial robustness compared to competitive methods.