Responsibilities include:

• Investigating how real-world semantic representations can be combined with generative latent diffusion models
• Visualizing where LDM-generated images land in the latent space and developing evaluation methods for generative quality with and without representation autoencoders

Responsibilities include:

• Refined the term tables for Chinese-to-English, English-to-French, and English-to-German translations to support the overseas release of Tencent’s Honor of Kings (HOK) and Delta Force (DF)
• Successfully trained translation models for the two games using Llama-3-70B, enhancing automation and efficiency in language localization for their international launches
• Introduced the retrieval-augmented generation (RAG) technique to the models to address term translation challenges and improve model performance, achieving translation edit rates (TER) below 15 for HOK and below 10 for DF, in line with company targets

Responsibilities include:

• Designed 5 homework projects for the course, covering multiple aspects including word embedding and pre-trained language models, and scored the completion of all 92 students in the course
• Provided students with 10 tutorials about the implementation details of AI models covered in lectures