Most of my recent work has been in the field of deep learing. Computer Vision, Natural Langaue Processing, Reinforcement Learning... I want to get my hands dirty in all of them and so I have. I have tried to get myself involved in whole range of different projects, some of them academic and some of them out of sheer curiosity.
(Active Project, Deep learning)
Roma, is a neural network that will learn music and then make it - something that's considered as deeply human. It uses Pytorch and will be primarily tested with Orchestral and Dubstep Music. More details will come as progress is made.
(2020, Computer Vision, Jupyter Notebook)
SinGAN generates unconditional realistic images from a single natural input image with akin quality to state-of-the art Generative Adversarial Networks (GANs) trained on complete classes of datasets.
(2019, Reinforcement Learning, Github)
Using curiosity as an internal reward, we make an agent explore virtual environments even in sparse reward scenarios. By modyfying the agent to make predictions for N steps in the future, effects on training and performance is investigated.
Over time, I got a chance to work with interesting and very capable people on multiple interdisciplinary projects.
In this project I implemented XNORNet architecture for CIFAR-10 dataset in Keras. My team designed ensembling of XNORNet and Residual Network to get better classification performance. We also evaluated and compared performance of our ensembled system with the state-of-the-art.
PythonKerasAs a part of my senior design project, my team designed a walking stick for blind people to help them easilly navigate through Purdue campus. I developed the PCB design to efficiently pack the hardware. This was my first time designing a PCB so the project was very interesting and it was really fun to do. We made a project video which can be found here.
EAGLE-CADCVerilogDuring my time at NUS, I worked to design an addressbook app catered to businessmen. I reworked the entire UI for an easier navigation of the app. I also added features for listing people by tags, converting phone numbers in the national convention and adding password to the address book. I also performed automated testing of the implemented features using Junit and Gradle
JavaJunitGradleI collaborated with 3 other students to design an ASIC chip that performed RC4 decryption and Sobel-Edge detection a given input image. I worked on smooth data transactions through AHB-Lite and correct sobel-edge detection in Verilog. I also modified the input method to edge-detection unit which improved the efficiency of the process by 450%.
SystemVerilog