Hi there! I was a research intern at BIGAI . Before joining BIGAI, I was a research assistant in TSAIL in Department of Computer Science and Technology at Tsinghua University  in Beijing, China. I got my Master degree from Department of Electrical and Computer Engineering, University of California, Los Angeles (UCLA ). Before that, I spent my 4 years of undergraduate study at Tsinghua University , majoring in Electrical Engineering and Automation.

My research interests lie primarily in AI-related areas, including but are not limited to reinforcement learning, natural language processing, computer vision, robotics, etc. I am especially into fusing these theories and methods, and applying them to real-world appliactions.

   


Education


Publications

Object Gathering with a Tethered Robot Duo
Yao Su, Yuhong Jiang, Yixin Zhu, Hangxin Liu
IEEE Robotics and Automation Letters, 2022 
IEEE International Conference on Robotics and Automation (ICRA), 2022 
[Paper]  [PDF]  [Code]  [Video]  [Web]

Honors and Awards


Skills


Research Experience

Robotics, 2021.07 - 2021.10
Intern at Beijing Institute for General Artificial Intelligence 
  • Devised a cooperative planning framework and an iterative optimization scheme to generate smooth and collision-free trajectories for a tethered Turtlebot3 Waffle Pi duo, which is tasked to gather scattered objects spread in a large area using a flexible net while properly avoiding obstacles.
  • Successfully implemented both simulation in Gazebo and the real-world experiment using MRAC to cope with increased payload with unknown dynamics.
  • Came up with the idea of leveraging reinforcement learning to handle uncertainty in VKC-based motion planning in a mobile manipulation task of door opening; set up simulation environment in MuJoCo with OpenAI Gym-like APIs.
    • Model-based Reinforcement Learning, 2021.03 - 2021.06
    Supervised by Prof. Jun Zhu  at Tsinghua University
  • Integrated physical knowledge into state-of-the-art model-based RL algorithms by introducing Hamiltonian canonical ODE into the learning process.
  • Resorted to the neural ODE to model the underlying transition dynamics, and achieved both higher sample efficiency and better performance than state-of-the-art methods.
    • Phase Identification of Power Distribution Network Based on Big Data, 2018.12 - 2019.06
    Supervised by Prof. Jin Lin  at Tsinghua University
  • Utilized Random Forest Classifier, t-SNE and Topological Data Analysis(TDA) in the phase identification of smart meters, based on collected historical big data of voltage magnitude, and fine-tuned the model.
  • Applied Monte Carlo method to the model, which helped to select the training sets effectively and to improve accuracy of phase identification to 97% with a small workload of 10%.
    • Implementation and Analysis of Bitcoin system and Ethereum system, 2018.07 - 2018.09
    Supervised by Prof. Xiaodong Wang  at Columbia University
  • Simulated two blockchain systems: Bitcoin system and Ethereum system in Matlab.
  • Analyzed the difference of decentralization level of the two systems, as well as their different behavior and security degree when under attack.

    • Selected Projects

    Full-stack Development of a Basketball Player Data Management System, 2021.04 - 2021.06
  • Developed a comprehensive data management system for student basketball players using Java and Spring.
  • Realized data storage and manipulation with MySQL, including insertion, deletion, update and selection.
  • Facilitated smooth operations of the system with a front-end interface developed by HTML and JavaScript; allowed easy management of both players and their game data, as well as customizable visualization of history stats and rankings.
    • Geotag Prediction of Covid-19 Related Tweets Based on a Multi-view Network, 2020.11 - 2020.12
  • Evaluated the end-to-end network proposed by this paper , on the dataset consisting of covid-19 related tweets that were extracted through Twitter API.
  • Replaced the text network in the model with several pretrained Word2Vec models, and slightly improved the model performance.
    • Deep Reinforcement Learning with Double Q-learning, 2020.05 - 2020.06
  • Implemented both DQN and Double DQN on a PLE game 'Pixelcopter' .
  • Optimized and compared the two networks, as well as evaluated their issues of value overestimation.
  • Achieved humal-level performance with the trained RL agent.
    • Combined Network Based Deep Learning Methods for EEG Dataset Classification, 2020.02 - 2020.03
  • Implemented and combined different architectures of CNN and RNN, and optimized them over hyperparameters.
  • Leveraged data augmentation, voting methods, etc. to improve the model, which boosted the classification accuracy to over 70%.
    • Classification & Clustering Analysis on Textual Data, 2020.01 - 2020.02
  • Extracted proper features from raw textual data of textual dataset "20Newsgroups" by preprocessing texts, generating TF-IDF matrix, and reducing the dimensionality of TF-IDF matrix by PCA/LSI/NMF.
  • Performed different classification methods on the processed representations, and analyzed their difference.
  • Performed K-Means clustering on the processed representations, and analyzed the effect of different preprocessing techniques: dimensionality reduction methods, scaling, logarithmic non-linear transformation to the data vectors, etc.
    • Grocery Shopping Helper Based on Interactively Customized Image Classifier, 2019.11 - 2019.12
  • Implemented a customizable web application, which classified goods based on MobileNet  by Google.
  • Applied transfer learning to the application by removing the last layer of MobileNet and appending KNN to it, enabling the model to be trained and customized simultaneously and smoothly.
    • The Development of a Chinese Input Method, 2019.03
  • Developed a Chinese Input Model by Python, using Hidden Markov Model, Viterbi Algorithm, etc.
  • Realized the output of corresponding Chinese characters of highest probability by identifying the input phonetic transcription of sentences.

    • Extracurricular Activities


    Contact

    jiangyuhong1997 [at] g [dot] ucla [dot] edu
    jiangyuhong1997 [at] 163 [dot] com

    © Yuhong Jiang 2022