Experiences
Researches:
- Optical computing
Research Assistant; Supervised by Prof. Mercedeh Khajavikhan; Jun 2021 – present- Researched various architectures of Optical Neural Network(ONN), Quantum Optical Neural Network(QONN), Continuous/Discrete Variable Quantum Computation
- Developed PT-ONN architecture with cascading PT-symmetric couplers, based on modulations of gain/loss contrasts; implemented simulations of on-chip training of PT-ONN using finite difference method with Python; achieved 68% training accuracy on MNIST dataset; simulated directional couplers in COMSOL; verified system transfer matrix of PT couplers with Mathematica
- Reproduced on-chip training of MZI-based ONN with Python; achieved 71% on-chip training accuracy on MNIST dataset
- Investigated the detrimental effect of V-shaped, intensity-dependent optical nonlinearity on VGG network architecture
- Spiking Neural Networks: Long Short-Term Memory (LSTM)
Research Assistant, supervised by Prof. Peter Bereel; Aug 2021 - present- Developed custom Spiking LSTM that allows pipelined inferencing
- Implemented custom Analog LSTM and Spiking LSTM module with PyTorch;
- Implemented RNN-to-SNN conversion through weight transfer and spike-timing-dependent plasticity (STDP) training; achieved 85%, 96% testing accuracy on IMDB, temporal-MNIST datasets
- CMOS neuromorpic circuits
Research Assistant; Supervised by Prof. Alice Parker; Jun 2021 – August 2021- Developed VLSI circuits of excitatory/inhibitory synapse, Axon Hillock, STDP-dopamine-noise synapse, dendritic spiking, edge detector, voltage adder, using Cadence; designed transistors to operate at subthreshold regime; obtained ideal biological waveforms for each circuit component
- Designed a multi-layer neural network with edge-detection, STDP-synapse, and dopamine-modulation layers; used PyTorch to simulate the network topology, spike propagation, and synaptic weight update, with hardware-realistic parameters; achieved unity accuracy on self-designed four-class, nine-pixel dataset
- Report
- Quantum computing
Research Assistant; Supervised by Prof. Xuehai Qian; May 2019 – Sept 2020- Developed Accelerating Quantum Optimal Control (accQOC), a comprehensive compilation methodology that accelerates pulse generation by 9x:
- partitioned the DAG of quantum circuit into sub-components under size constraints, using Qiskit
- concatenated QOC pulses of each components using dynamic programming
- balance partitionioned the MST of computing nodes for efficient parallel computation, using METIS
- Worked on the optimization of VQE(Variational Quantum Eigen-solver) algorithm:
- worked on circuit-level implementation of VQE with pyQpanda
- optimized VQE measurement overhead through joint measurement of commuting Hamiltonians; projecting linearly-independent basis terms onto qubit computational basis using stabilizer formalism
- worked on pulse-level optimization of VQE circuits using Qiskit open-pulse and QOC
- Developed Accelerating Quantum Optimal Control (accQOC), a comprehensive compilation methodology that accelerates pulse generation by 9x:
- In-memory computing with InP memristor
Research Assistant; Supervised by Prof. Rehan Kapadia; Jun 2020 – Jan 2021- Researched various architectures of mapping ANN/SNN onto hardware using semiconductor devices that mimic synapses, neurons
- Simulated a temporal-encoded convolutional Spiking Neural Network(SNN) and a rate-encoded SNN with PyTorch, incorporating InP-synapse STDP parameters; achieved 89 & 91% training accuracies respectively, on MNIST dataset
- Collaborated on designing Arduino/breadboard implementation of neural networks, utilizing a crossbar array of InP memristors
- Grew InP using the Low-Temperature Templated Liquid-Phase(LT-TLP) technique
Internships
- Summer Internship, Shanghai AIKE Measurement Co. (July 2018 – August 2018)
- Learned to use Labview to sample data and control hardware
- Used Labview to automate a printer to print labels on a streamline
- Assisted in designing the structure of a dispenser
Activities
- Mobile App: Go Eat
- Developed a cross-platform app that recommends restaurants according to users’ preferences
- Developed KNN and genetic models to recommend restaurants based on users’ and restaurants’ features
- Implemented data fetching and storage using Firebase
- Desktop Game: Mind Palace
- Developed a game that trains players to master ”Mind Palace” mnemonic technique
- Designed game stages and implemented them with SDL library
- Extracted key information from sentences with Google Cloud API
- Neural Signal Sampling
- Collaborated on designing a neural-signal sampling device
- Realized data transfer between FPGA and PC using Opal-Kelly’s API; implemented SPI protocol to transfer data between FPGAs; visualized data on PC end with matplotlib library
Class Projects
- Reinforcement Learning in Games
- Developed customized gym-compatible, overhead-shooting game environment with pygame
- Trained AI agents to achieve close to full score with Deep Q learning (DQN), written with PyTorch
- Currently extending the current framework to allow multi-player collaboration
- CMOS VLSI
- Implemented a 32 bit MAC unit with Brent-kung Adder and Array Multiplier;
- Created schematics and drew layouts using Cadence
- Parallel Programming
- Developed grayscale-to-color image conversion using CNN
- Converted python model into C++ model using keras2cpp library
- Parallelized convolution-layer computation using OpenMP and Pthread library
- Video Game Programming
- Implemented 2D classical Arcade games such as Zelda, Super Mario, PAC-MAN
- Implemented 3D games such as Mario Cars, FPS, Parkor game
- Programmed game engines using C++ and SDL libraries
- FPGA Arcade Game “SPLATOON”
- Designed a 2D-board Splatoon game that runs on a Xilinx FPGA Spartan 6, implemented with Verilog
- Integrated joystick module for user control and VGA module for image display
- Web Game “Mission Universe”
- Created a web-based, multiplayer jet-fighting game, using HTML, JavaScript, Java
- Utilized Phaser API to construct game elements and WebSocket to transmit data in multiplayer mode
- Home Light System IoT Project
- Developed a remote light control system running on Raspberry Pi
- Fetched data from light sensor and transmit it through OpenMote
- Utilized MQTT library to transmit data and control illumination
- Code Compiler
- Implemented a compiler that parses a realistic program into tree structure, with Bison and C++
- Generated assembly language and computed values of variables and memories
- Accelerometer
- Developed an accelerometer running on Arduino
- Integrated a rotary encoder to set threshold and a buzzer to present speed, controlled by interrupt
Skills:
Programing/Software:
- Python
- C/C++
- Java
- PyTorch
- Qiskit
- pyqpanda
- Verilog
- Cadence
Related Course
- EE514: Quantum Error Correction
- EE301: Linear Systems
- EE370: Electromagnetism
- EE354: Introduction to Digital Circuits
- EE348: Electronic Circuits
- EE451: Parallel and Distributed Computation
- EE457: Computer System Organization
- EE477: MOS VLSI Circuit Design
- EE459: Embedded System Design Laboratory
- EE520: Introduction to Quantum Computing
- EE582: Neuromorphic Computing
- EE540: Quantum Electronics
- CSCI527: Applied Machine Learning in Games
- CSCI201: Pinciple of Software Engineering
- CSCI270: Introduction to Algorithms and Theory of Computing
- CSCI104: Objected Oriented Programming
- CSCI353: Internetworking
- Math407: Propability Theory
- Math445: Mathematics of Physics and Engineering
- Phys438: Introduction to Quantum Mechanics
- Phys304: Mechanics
- ITP445: Professional C++
- ITP380: Video Game Programming