Projects

Current Projects

• The details of current projects cannot be made public as these projects are under NDA.

Previous Projects

• Incremental Learning

  • Funding Agency: Industrial Technology Research Institute, Taiwan
  • As part of my PhD research, I have worked on the development of an algorithm for class incremental learning in Deep Neural Networks (DNNs). The main aims of the algorithm development were to develop a hardware friendly and high performance incremental learning technique.
• Learning On-Chip
  • As part of my PhD research, I have worked on the development of an online learning chip that was able to perform inference as well as training on the chip. The main goal of this research was to develop a digital hardware design in such a way that it brings the possibility of optimizing and training a Deep Neural Network (DNN) model on the edge.

• Designing of ECG compression Chip (Master Thesis)

  ​ In this project, I completed the digital design and layout of ECG compression algorithm using TSMC90nm technology. The algorithm consisted of adaptive linear prediction as the prediction part and Golomb Rice Code as entropy coding engine. The main aims of the hardware design were to process real-time data without memory usage and low-area utilization in ASIC. ‘Synthesis’ of the implemented design was performed using Design Compiler. Floor planning, placement, clock tree synthesis and routing was completed in IC Compiler. DRC and LVS were performed in Calibre. The project was supervised by Professor Tsung-Han Tsai.

• Designing of Digital Hardware Design of FBCOT Encoding for HTJ2K

  ​ Fast Block Coding with Optimized Truncation (FBCOT) is new proposed engine for data compression in JPEG2000 Part 15. The standard for High-Throughput JPEG200 (HTJ2K) can be found here. An open source implementation of HTJ2K is available here. HTJ2K is intended to replace EBCOT due to its low latency and high throughput features. I have worked on the implementation of encoder in digital hardware design. For encoder module, variable length coding, MEL coding, significance pattern coding, and bit stuffing and packing were designed in digital hardware. For this project, the collaboration work was performed with Professor David Taubman from University of New South Wales, Australia and a team from University of Stuttgart, Germany.

• Automatic Iris Segmentation and Recognition

  ​ Automatic iris segmentation and recognition were performed using total variation model. This project was implemented using Matlab. The main purpose of this project was to increase the speed of iris segmentation while maintaining the highest accuracy.