Research Experience
Laboratory: Software/Hardware Co-design Lab (Sharc)
Advisor: Prof. Cong (Callie) Hao Georgia Institute of Technology ECE
Role: Undergraduate Research Assistant
Jan.2021-Present Title: Low-latency and energy-efficient object detection and tracking on FPGA [More details]
Objective:
Proposed a hardware-friendly grid-based mask generation method, which requires much less information and can skip redundant computations at an early stage.
Contributions:
Applied the method on SkyNet as a case study, deployed the new architecture on ZCU106 FPGA and evaluated it on DAC-SDC dataset.
Increased the throughput by 35.0% and reduced the energy by 31.3% with only a slight IOU drop in the case study.
Extended our technique on other backbones, e.g., ResNet and UltraNet and evaluated it on other single object detection and tracking datasets, e.g., GOT-10k.
Jan.2021-Mar.2021 Title: Research on improving the performance of multivariate gas sensor by using neural networks
Objective:
Applied Temporal Convolutional Network (TCN) to realize instantaneous gas type and concentration prediction.
Contributions:
Implemented Pytorch to construct TCN to process data from gas sensor and improved the gas types and concentration instantaneous prediction accuracy from 79% to 95%.
Laboratory: Novel Electronic Systems Group at MIT
Advisor: Dr. Mindy D. Bishop Massachusetts Institute of Technology EECS
Role: Undergraduate Research Assistant
Mar.2021-May.2021 Title: Designing machine learning hardware for image denoising
Objective:
Mapped Denoising Convolutional Neural Network (DnCNN) on Google Pixel architecture to realize image denoising on smartphones.
Responsibilities:
Simulated image data flow on Google Pixel architecture in MATLAB and compared the execution time and energy consumption with some classic hardware (e.g., CPUs, GPUs and NPUs).
Laboratory: Lab of Mesoscopic Physics and Quantum Devices
Advisor: Prof. Feng Miao School of Physics, Nanjing University
Role: Undergraduate Research Assistant
Sept. 2020-Present Title: Research on Network Behavior of Memristor Circuit for Large Scale Brain-like Operation
Objective:
Mapped some unsupervised learning algorithms, e.g., STDP algorithm onto the memristor array to realize digital classification and language recognition.
Responsibilities:
Used Pytorch to construct Spiking Neuron Network (SNN) and realized STDP algorithm.
Jun.2020-Sept.2020 Title: Simulation of the function of Gaussian synapses in PNN
Objective:
Mapped PNN onto a memristor array consisting of memristive gaussian synapses.
Contributions:
Built a Probabilistic Neural Network (PNN) with MATLAB and achieved 96% accuracy using the Iris dataset.
Sept.2019-Jun. 2020 Title: Networking retinomorphic vision sensor with memristive crossbar for brain-inspired visual perception [More details]
Objective:
Constructed a neuromorphic vision system prototype to realize moving object tracking.
Achieved a reconfigurable vision sensor to simulate the function of Convolutional Neural Network (CNN) in image sensing, processing and classification.
Contributions:
Programmed to develop Recurrent Neural Network (RNN) to successfully predict the location and trajectory of a moving cross and the mean squared error was less than 10^-2.
Constructed a Convolutional Neural Network (CNN) to help verify the efficiency and superiority of the vision sensor prototype based on a memristor array.