C/C++, CUDA, video，trace flow，driver

Support with robotic platform for radar-based Neuromorphic projectProgramming GeNN (GPU enhanced Neuronal Network) models (Spiking neural network) in C++ for deploying them into the robotic platform (NVIDIA Jetson TX2).

Extension of the Machine Learning Framework in C++(1): Literature research on RNN, LSTM, and GRU, especially for their training algorithm(2): Familarization with the machine learning framework based on C++ developed by the chair(3): Implement the forward propagation of multilayer GRU (multithreading)(4): Implement the training algorithm BPTT (back propagation through time) of multilayer GRU based on the chain rule and the jacobi matrix (multithreading)(5): Make a comparision of the GRU weights and the loss function value generated by our ML framework with the corresponding value produced by Pytorch(6): Take the ML framework into a real application (such as sin/cos function prediction, trajectory prediction)

Embedded Software Programming and Digitial Integrated Circuit Design(1): Implementation of a JTAG-adapter using a micro controller system (Arduino board)(2): Application of the JTAG adapter to operate and control a test chip(3): Generalization of the software(4): Investigation on resolution, accuracy and iteration depth for a Cordic algorithm (arctan calculation)(5): Implementation of that Cordic algorithm using Verilog HDL(6):Simulation of the Cordic implementation and estimation of numerical deviations of the hardware implementation from correct precise values

Characterization of Machine Learning Hardware(1): Classify the MNIST dataset using deep neural network and convolutional neural network in Tensorflow(2): Using Qm,n method to quantize the floating-point weights of DNN and CNN into fixed-point weights for the preparation of the inference phase in ARM(3): Using the library (CMSIS-NN) provided by ARM to transfer the two network architectures to ARM Cortex M4.(4): Compare the inference accuracy between ARM Cortex M4 and Tensorflow(5): Using machine learning accelerator provided by university to replace CMSIS-NN and write neural network into FPGA.(6): Evaluate the inference accuracy of neural network model in FPGA