Pre-trained language models (PLMs) are known to be overly parameterized and have significant redundancy, indicating a small degree of freedom of the PLMs. Motivated by the observation, in this paper, we study the problem of re-parameterizing and fine-tuning PLMs from a new perspective: Discovery of intrinsic task-specific subspace. Specifically, by exploiting the dynamics of the fine-tuning process for a given task, the parameter optimization trajectory is learned to uncover its intrinsic task-specific subspace. A key finding is that PLMs can be effectively fine-tuned in the subspace with a small number of free parameters. Beyond, we observe some outlier dimensions emerging during fine-tuning in the subspace. Disabling these dimensions degrades the model performance significantly. This suggests that these dimensions are crucial to induce task-specific knowledge to downstream tasks.

Clustering short text streams is a challenging task due to its unique properties: infinite length, sparse data representation and cluster evolution. Existing approaches often exploit short text streams in a batch way. However, determine the optimal batch size is usually a difficult task since we have no priori knowledge when the topics evolve. In addition, traditional independent word representation in graphical model tends to cause “term ambiguity” problem in short text clustering. Therefore, in this paper, we propose an Online Semantic-enhanced Dirichlet Model for short sext stream clustering, called OSDM, which integrates the word-occurance semantic information (i.e., context) into a new graphical model and clusters each arriving short text automatically in an online way. Extensive results have demonstrated that OSDM has better performance compared to many state-of-the-art algorithms on both synthetic and real-world data sets.

Weight tying is now a common setting in many language generation tasks such as language modeling and machine translation. However, a recent study reveals that there is a potential flaw in weight tying. They find that the learned word embeddings are likely to degenerate and lie in a narrow cone when training a language model. They call it the representation degeneration problem and propose a cosine regularization to solve it. Nevertheless, we prove that the cosine regularization is insufficient to solve the problem, as the degeneration is still likely to happen under certain conditions. In this paper, we revisit the representation degeneration problem and theoretically analyze the limitations of the previously proposed solution. Afterward, we propose an alternative regularization method called Laplacian regularization to tackle the problem. Experiments on language modeling demonstrate the effectiveness of the proposed Laplacian regularization.