Projects

DAC: A Double Accelerating Contrastive Learning Framework

Combining Masked Image Modeling with Contrastive Learning by proposing a new data augmentation method and an assymetric architecture for Contrastive Learning. Both modifications help achieve better performance within faster training time compared to previous methods. Achieved SOTA level results on image classification.

Jan 2022 ~ Apr 2022

[Report] [Github] [Slides]

FAD: Feature Alignment Discriminator for Abstractive Text Summarization

Tackling the discrete nature of GAN in NLP with feature matching discriminator. The discriminator is applied at finetuning stage for text summarization, where real/fake samples are from generated/reference summaries. Achieved better ROGUE score than BART only.

Feb 2022 ~ Apr 2022

[Report] [Github] [Slides]

Vocoder review

Performance comparison on several popular vocoders on different aspects. A brief report and synthesized samples are included.

[Github]

Mar 2021 ~ May 2021

Long paper review on Solving Min-Max Optimization with Hidden Structure via Gradient Descent Ascent

A comprehensive review of method on the paper, with reimplementation of part of its applications.

Mar 2022 ~ Apr 2022

[Report] [Github]

Reimplementation of WGAN-GP

Reimplementation and performance comparison of WGAN and WGAN-GP on LSUN bedroom, CelebA, and Animefaces datasets.

Nov 2021 ~ Dec 2021

[Report] [Github]