Sanshuiii
0%

About me

Yifan Shen

I am pursuing a master’s degree in Machine Learning at Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi. I am supervised by Dr. Kun Zhang. My research interest mainly focuses on Causality and Causal Representation Learning. Before that, I achieved my bachelor’s degree in Computer Science at Nanjing University of Informational Science and Technology.

Education

  • 2022-2024 (expected)
    MSc, Machine Learning, MBZUAI

  • 2018-2022
    BSc, Computer Science, NUIST

Professional Experience

  • SDE Intern, Tencent, Tencent Cloud (2022)
  • SDE Intern, Microsoft STCA, Bing Ads (2021)
  • Research Intern, Microsoft Research Asia, Machine Learning Group (2021)

Projects

  • 3D Rougelike Game AI Based on Vision (2021)
    For some games with a small amount of code, it is undoubtedly inefficient to design a set of APIs for testing. The amount of code of such APIs may be equivalent to the game itself. Given this pain point, we designed a set of distributed hierarchical reinforcement learning models with only a vision signal as input, which can automatically complete all levels of the game without APIs. The model can effectively help testers conduct stress testing. AI completion time cost can be used as a measure of difficulty. What is more, the model can efficiently find bugs in the game during exploration. As a result, this project won the first prize in ByteDance summer camp among 50+ teams.

  • Database index compression with machine learning (2020)
    In the database, the index is often implemented by traditional data structures such as B + tree. These data structures can accurately locate the value corresponding to the key. However, in some specific situations, such indexing is not efficient. For example, in a log database, the number of bytes each tuple consumes is small, which makes B+ tree indexing less spatially efficient. However, if we can encode features of the keys and decode them during a query, storage overhead can be reduced. Therefore, we study this kind of special database, which satisfies 1) the key is regular and increases monotonically with the physical address, 2) read-only, and 3) each tuple is small. We design and implement a prototype of a database index based on a machine learning algorithm, which reduces the space consumption to 1% (compared to the B+ tree) without losing query performance. In addition, it supports efficient parallel queries as a bonus.

Honors & Awards

  • 2021
    • 1st prize in ByteDance summer camp
  • 2020
    • Gold Award in Jiangsu Collegiate Programming Contest
    • Bronze Award in International Collegiate Programming Contest (East Continent Final)
  • 2019
    • Silver Award in International Collegiate Programming Contest (Nanjing Site)
    • Silver Award in Chinese Collegiate Programming Contest (Qinghuangdao Site)