Shivanshu Tyagi

June 24, 2022

I wake up at dawn; I see no light. I wake up at dawn; I feel tired and lite. I give my wings a flick and my face a wipe.....

Brain Imaging and Analysis

Age and Gender Prediction in Low resolution CCTV footage - InterIIT 2022

A drastic difference between math and story processing

Worked with 3 other people on this research project to differentiate between the areas of brain responsible for math and language processing using fMRI data from Human Connectome Project. we used GLM to predict the task using the activity from selected areas of brain. Model was able to predict correctly for hypothesized parcels (areas) and not for random ones.

Why would you do that

Mentored this project aimed at using Reinforcement Learning techniques with Evolutonary game theory theory to find a best strategy in an environment. The project started by asking "Why would an agent become altruistic in a survival setting?". We create a simulation based on stocastic processes and used Q-Learning to simulate multi-agent game. Currently we're working on this project to use Deep RL based techniques for this game.

Computer Vision A Classic & Modern approach

Mentored a project under Stamatics IITK where we taught the basic concepts of Computer Vision and some classical algorithms used in CV. With a gentle introduction of CNNs. It was a fun experience to first revisit (learn) and solve doubts.

Doori

We developed a system to allocate access rights for shopping places to decrease the congestion amidst the havoc created by COVID-19. This system can be used to allocate any scarce resource. The allocation algorithm is based on Game Theory and mechanism design with monetary transactions. It computes in polynomial time, which is wrapped by a server (NodeJS based) deployed on IITK's servers and an android app (Flutter based) on the play store. The concerned paper is under the proceedings of the AAAI conference.

Model-Zoo

We read and implemented a collection of Machine Learning Algorithms from various domains such as Natural Language Processing, Computer Vision, Object detection, GANs, etc.. using PyTorch and Tensorflow. The code is Open Source and can be used by anyone for developing their systems.

Facial Expression Recognition

We developed a deep learning system to recognise human facial expressions. We created it in 3 steps; first, we trained a network to recognise human emotions from an image, then we did the same for recorded videos, and in the final step, we implemented it for a live video feed. The code, along with examples, is available on GitHub.

Autonomous Underwater Vehicle(Triton)

We are developing a highly optimised software stack for Triton. Triton is designed for autonomous navigation in an underwater environment and performs a predefined set of tasks to compete in SAVe using depth estimation, SLAM, object recognition, etc.

PoleBalance

A naive attempt in making AI learn to balance a lope on a cart. I developed this while procrastinating on my assignments XD. This is a pole balance game that is run along with a Neural Network training loop that learns using SGD.

CP-List

When I wanted to start Competitive Programming, initially, it was quite hard for me. I was more interested in developing things than competing in some online competitions, but many people liked it, so I gave it a shot. So, I was having trouble finding the competitions which were happening, so I developed an app to show all upcoming CP events in a sorted list, with an option to add events to your calendar to make it easier for me. Later on, I realised that many other services exist that provide a google calendar for that :p.

IMDB-Player

I liked watching movies, but I did not have Netflix and finding links for movies can be challenging. So, I created a chrome Extention that edits the IMDB movie/series page and embeds the movie in the page itself, saving me from the trouble of finding a link every time I come across an excellent movie.

GnS Website

I developed a website for Games and Sports Council IIT Kanpur.