Advance Lane Lines

Completed as a part of the Self-Driving Car Nanodegree Program by Udacity.

Developed a sophisticated pipeline to detect lane lines in videos robust to shadows and curving roads, completed as part of the Self-driving Car Nanodegree Program by Udacity. The pipeline consists of several steps: camera calibration to obtain distortion coefficients for undoing the distortion in each frame, perspective transform to obtain a bird-eye’s view, applying color and gradient-based thresholds and using a sliding window based approach for fitting quadratic polynomials to pixels corresponding to the lane lines

Links: Code | Report
Finding Lane Lines on the Road

Completed as a part of the Self-Driving Car Nanodegree Program by Udacity.

When we drive, we use our eyes to decide where to go. The lines on the road that show us where the lanes are act as our constant reference for where to steer the vehicle. Naturally, one of the first things we would like to do in developing a self-driving car is to automatically detect lane lines using an algorithm.
​
In this project we detect lane lines in images using Python and OpenCV.

Links: Code | Report
Continuous Control: learning the optimal policy in a continuous action space

Completed as a part of the Deep Reinforcement Learning Nanodegree by Udacity.

Deep Deterministic Policy Gradient (DDPG) for learning the optimal policy in continuous action spaces. This is my second project as part of the Deep Reinforcement Learning Nanodegree by Udacity.

There are 20 agents in the environments and each agent is a double-jointed arm which can move to target locations.  A reward of +0.1 is provided for each step that the agent's hand is in the goal location. Thus, the goal for each agent is to maintain its position at the target location for as many time steps as possible. There are also single agent versions of the environment but distributed training shows performance improvements by multiple agents sharing their experience.

Links: Code | Report
Navigation: training an RL agent to navigate in a Unity environment

Completed as a part of the Deep Reinforcement Learning Nanodegree by Udacity.

The project demonstrates the ability of value-based methods, specifically, Deep Q-learning and its variants, to learn a suitable policy in a model-free Reinforcement Learning setting using a Unity environment, which consists of a continuous state space of 37 dimensions, with the goal to navigate around and collect yellow bananas (reward: +1) while avoiding blue bananas (reward: -1). There are 4 actions to choose from: move left, move right, move forward and move backward. The trained agent can be seen in action on the left.

Links: Code | Report
Splice site prediction in DNA sequences using Deep Learning

Prof. Ashish Anand, Assistant Professor, IIT Guwahati

Employing Natural Language Processing (NLP) ​ techniques for the task of understanding the genome, modeling each nucleotide (A, G, T, C) as a letter and each DNA sequence as a sentence. A DNA sequence consists of alternating intron-exon pairs. Here, exon are the ones that code the protein and introns can be considered the non-useful, which are removed in a process called splicing. When the splicing doesn't behave ideally (which happens ~95%) we get what is called alternative splicing, which is the primary cause of diseases like cancer. The interface between an intron and exon is called a splice site, and our task is splice-site prediction. Presently, I have been studying past work done in the field, reading and implementing architectures that have been successful in NLP tasks like Recurrent Neural Networks (RNN)(and its variants like Gated Recurrent Units [GRU], Long Short Term Memory [LSTM]), Recursive Neural Networks, Convolutional Neural Networks (CNN) and also intend to experiment with architectures that have not been tried in the NLP domain before. Paper submitted to Bioinformatics.

​Quora Question Pairs on Kaggle

Trained the Gated Recurrent Unit (GRU) variant of RNN in combination with Random Forest for predicting duplicate question pairs to land in the top 25% on Kaggle among 3000+ teams with a cross-entropy loss of 0.21.
Cancer recurrence prediction - Research project
I worked as a part of the research group under Dr. Amit Sethi, where we tried to predict the recurrence of cancer among patients who have been treated for the disease once, using actual patient data (histopathological images) of 1600 patients collected from hospitals in Guwahati.
XYZ Reader - Android Developer Nanodegree
Applying Material Design Principles to an existing app.
Build it Bigger - Android Developer Nanodegree
Android Application implementing a Java Library, Android Library, GCE, jUnit Tests and Free/Paid Flavours 
Social Network using Laravel - Personal project
A mock social networking website with features like adding/removing friend/status/like/dislike/replies.
Eventify - Winners, CodeIO, IIT Guwahati
Eventify sends event's details directly from a student's webmail to the app using a chrome extension.
Sunshine - Android Developer Nanodegree
A weather app showing the weather forecast of the next week, as a part of Udacity's Android Developer Nanodegree Program.
Popular Movies - Android Developer Nanodegree
This app displays the list of movies sorted according to the users choice, as a part of Udacity's Android Developer Nanodegree Program.
Stock Hawk - Android Developer Nanodegree
Improved on an existing project to make it production ready 
Badigaadi - Android Developer
Badigaadi is an 'Uber for Trucks' that enables goods to be transferred via booking trucks through the app.
SWC - Project
An app for the freshers of IIT Guwahati, that shows their entire schedule along with announcement notifications.
Spirit, IIT Guwahati- Core Team, WebOps
Spirit is the largest sports festival of the north-east organised annually in IIT Guwahati.
Yash Infotech - Android Developer
Yash Infotech is a Kolkata based company which deals in computer peripherals and accessories.