About me

I’m a third-year undergraduate student from School of Computer Science, Harbin Institute of Technology， ShenZhen. I am very fortunate to be advised by Prof. Xuebo Liu of HITSZ-ICI-HappyTrans team from School of Computer Science, Harbin Institute of Technology，ShenZhen. My CV is here: Xinyu Ma’s Curriculum Vitae.

Research Interest

My research interest includes natural language processing and machine translation ：

【2023.10.8】EMNLP2023 accept main “Clustering Pseudo Language Family in Multilingual Translation Models with Fisher Information Matrix”
【2023.6.24】EMNLP2023 Conference submission, my very first paper “Clustering Pseudo Language Family in Multilingual Translation Models with Fisher Information Matrix”.

【EMNLP2023 main】”Clustering Pseudo Language Family in Multilingual Translation Models with Fisher Information Matrix”
- We introduce a novel approach to pseudo language family clustering, which is inherently dependent on the MNMT model itself. Our proposed methodology hypothesizes that language pairs exhibiting similarities in their impact on model parameters are likely to possess a high level of congruence and should, consequently, be grouped together. We employ the fisher information matrix (FIM) to quantify such similarities between language pairs. We demonstrate the efficacy of our approach by enhancing low-resource NMT with pseudo language family clustering, yielding superior results when compared to the conventional use of language family categorization.

The 4th IKCEST The Belt and Road International Big Data Competition and the 8th Baidu & Xian Jiaotong University Big Data Competition
- We complete eight directions from Chinese \(\leftrightarrow\) French, Chinese \(\leftrightarrow\) Russian, Chinese \(\leftrightarrow\) Thai,and Chinese \(\leftrightarrow\) Arabic.
- Our team enployed many approachs, including backtranslation, rdrop, using r2l model for reranking, using pre fine-tuned model for data selection. I have conducted most of the training approach.
- The average increase of approximately 2.0 BLEU score, we attained the third prize in the final competition as an undergraduate team, achieving an average BLEU score of 32.965 for Chinese \(\leftrightarrow\) Arabic translation.
ASC22 Student Supercomputer Challenge
- I was responsible for optimizing the training process of the deepmp-kit machine learning molecular dynamics tool. Utilized knowledge of OpenMP, assembly language, and SIMD (Single Instruction, Multiple Data) to perform loop unrolling, memory optimization, and parallel processing on the algorithm.
- Our team was awarded the second prize.

Prof. Xuebo Liu is an exceptionally outstanding mentor who has provided me with invaluable assistance in my research endeavors. He has guided me in developing systematic research habits and has been a guiding light throughout my research journey.
I am fond of playing football and badminton.