project

Applied a deep learning ResNet model with transfer learning techniques to classify spectrograms of human and AI-generated voices, achieving high test accuracy in the ASVspoof2019 competition.

Leveraging fine-tuned LLMs and the GPT-4 model to identify and highlight unfair or significant clauses within the Terms of Service agreements of major service providers.

Leveraging Pyspark to perform tranditional text data analysis and NLP learning tasks when the scale of samples is large. And using the trained model to construct an automatic hash-tagging system for incoming tweets.

Using German General Social Survey (ALLBUS), this project tend to classify the survey’s responses into one of the five categories from 1 to 5 which represent the “self-asset financial health” of samples.