Hi! I’m Jerome Francis, currently doing my Masters from Mila and University of Montreal. Previously worked as Data Scientist @ TCS. I’m interested in Deep Learning with a specific focus on Reinforcement Learning and Decision Making.
I did my undergraduate from TKM College of Engineering, Kollam in Computer Science and Engineering with Honors.
Recent news
What I’m doing now :
- Studying RL papers and implementing projects based on different concepts of RL.
- Reading “Godel,Escher,Bach : An Eternal Golden Braid by Douglas Hofstadter”
A bit more about me
I love reading fiction novels about science, philosophy and a bit of history, for example, the Dan Brown series, cycling in the evening – which mostly depends on the mood or work, and listening to film scores like “Aggressive Expansion”, “Iron Man 3 main theme”, “Love & Work”, and so on …
Publications
Abin Shoby, Jerome Francis, Jini Raju, Ushadevi Amma C., Ansamma John. A Novel approach for b value optimization in IVIM imaging using an LSTM based deep learning network on simulated data, International Journal of Medical Engineering and Informatics, 2023
Projects
RL + IL in Minecraft
Used imitation learning along with RL to solve the task of obtaining a Diamond in the MineRLObtainDiamond-v0 environment. My agent was able to obtain all the items till the stone-pickaxe object, with a few Iron Pickaxes. Careful distribution of data samples from datasets like TreeChop-v0, Navigate-v0 and ObtainIronPickaxe-v0 was maintained in the training set, along with removal of unnecessary task examples showed good improvements in sequential task execution.
Improving IVIM Imaging technique
During my senior year, I worked on improving an MRI Imaging technique known as "Intra-Voxel Incoherent Motion" or IVIM, with the application of Deep Learning, under the supervision of Dr. Ansamma John. We focussed on data augmentation of MRI scans, MRI scan reconstruction using learned visual representations and b-value optimization.
Video based Rail gap measurement
Developed a feature to measure the width of rail tracks present in video frames using Semantic Segmentation. It aided the detection of track width irregularities present in the tracks, which proved to be a crucial addition to the Rail Defect Detection product. This project was implemented during my internship @ Ignitarium Technologies, Kochi in July 2019. [DEMO]
RailSathi
Worked in a 3-person team to design a railway booking app as the project for a Design course. This application aims to provide booking ticket service for connecting trains if direct trains are not possible. It also included a sanitation rating feature, alongwith an interface for seat-exchanges and chat-room talking feature specific to each local coach . Presented the product at Ideafest, a techfest organised by Kerala Startup Mission.
Smart Waste Management
This is a part of an IoT based smart waste management system using smart bins to provide waste-management services. Its objective is to calculate an estimate amount of waste generated by a locality on a periodical basis (days, weeks). The data was experimented with different time-series models like SARIMA and ARIMA to study about the waste generation. This study was used to ensure the frequency of removing waste from smart bins. The smart waste management system was selected for final round of TEKCON'19 contest.
Q-Learning in Solving Taxi environment
A tabular version of Q-Learning along with Temporal Difference concept was used to make the taxi agent pickup a passenger from their source and drop them at their given destinations. Gym's taxi-task environment was used here.
Creation of Jazz music using LSTM
It uses LSTM network model to learn a jazz composition and then randomly sample musical tones according to the jazz style learned. This was an assignment of week 1 for a course "Sequence and Models" in Coursera.