Projects

Artificial Intelligence | Deep Learning | Machine Learning

  • Complete System Integration for Autonomous Vehicles and Programming on a Real Car - Developing an entire system of core functionality required for operating an autonomous vehicle. The project involves writing ROS nodes to implement core functionality like traffic light detection, control, and way-point following. The final system was deployed on an actual 2016 Lincoln MKZ vehicle which has installed sensors and parts like 2 Velodyne VLP-16 LiDARs, 1 Delphi radar, 3 Point Grey Blackfly cameras, an Xsens IMU, an ECU, a power distribution system, and more. The car successfully completed the required number of laps around the test track. My contributions were in the traffic light detection and classification module. The model was trained with Transfer Learning on top of the ResNet model using the TensorFlow Object Detection API and achieved an outstanding accuracy of more than 98%.
  • End-to-End automatic speech recognition (ASR) program - Developed an end-to-end system using the LibriSpeech dataset to create a Voice User Interface (VUI). The algorithm will first convert any raw audio to feature representations that are commonly used for ASR. Next step is building neural networks that can map these audio features to transcribed text.
  • Path Planning for Autonomous Vehicles - The goal in this project is to create a path planner that is able to create smooth, safe paths for the car to follow. The highway track has other vehicles, all going different speeds, but approximately obeying the 50 MPH speed limit. The car will be transmitting its localization information, along with its sensor fusion data, which estimates the location of all the vehicles on the same side of the road. Use the map, which consists of way-points along the middle of the highway, to figure out where the lanes are, and what the curvature of the road is. Used the concepts of Model-Predictive-Control (MPC), Sensor Fusion and Localization.
  • Air Cargo Transport Planning - Used logic and planning techniques to create an AI that finds the most efficient route to route cargo around the world to their respective destinations. This project used a combination of propositional logic and search along with A* heuristics to find optimal planning solutions.
  • Kalman Filter for Pedestrian Detection - Implemented a Extended as well as Unscented Kalman Filter algorithm in C++ using sample Lidar and Radar data for the purposes of obstacle detection and pedestrian tracking in two dimensions.
  • Image Generation using Generative Adversarial Networks - Trained a Deep Convolutioal Generative Neural Network on a set of images from the MNIST and CelebA datsets. The trained network is then able to generate a whole new set of images which are quite close to original images in an unsupervised way.
  • Model Predictive Control for Autonomous Driving - Implemented Model Predictive Control to drive a vehicle around a track even with additional latency between commands. The Model Predictive Controller (MPC) calculates the trajectory, actuations and sends back steering to the simulator for vehicle manoeuvring.
  • Sign Language Recognition System with Probabilistic Models - Built a system that can recognize words communicated using the American Sign Language (ASL). Trained a set of Hidden Markov Models (HMMs) using part of a pre-processed dataset of tracked hand and nose positions extracted from video to try and identify individual words from test sequences. Experimented with model selection techniques including BIC, DIC, and K-fold Cross Validation.
  • Deep Learning to Clone Human Driving Behaviour - Use Deep Neural Networks and Convolutional Neural Networks to predict the vehicle steering angles given a set of image inputs of the road ahead. Model trained using Convolutional Neural Networks on TensorFlow (AWS GPU instance) and tested using a track simulator which simulated a road complete with lane markings and turns like a the ones present in popular online games. Used optimization techniques such as regularization and dropout to generalize the network for driving on multiple tracks.
  • Language Translator using Recurrent Neural Networks - Use the knowledge of Transfer Learning and RNNs to write a program to translate from one language to another. Used a small subset of English and French conversations to teach a RNN network. Program written using Tensorflow and Python.
  • Traffic Sign Classification Using Deep Learning - Use Deep Neural Networks and Convolutional Neural Networks to classify traffic signs. Model trained so it can decode traffic signs from natural images by using the German Traffic Sign Dataset on Tensorflow (AWS GPU instance). Experimented with different network architectures. Performed image pre-processing and validation to guard against over-fitting. The model performed with an accuracy of 94.8% on new and previously unseen images of traffic signs.
  • Artificial Intelligence for Game Playing Agent - Created an AI that beats human opponents in the game of Isolation using Minimax, AlphaBeta Search, and Iterative Deepening.
  • Particle Filter for Autonomous Vehicle Localization - Implemented a two dimensional particle filter in C++ capable of localizing a vehicle within desired accuracy and time.
  • Road Detection using Fully Convolutional Networks (FCNs) - The aim of this project is to understand the concepts of Fully Convolutional Network (FCN) and write a program to label the pixels of a road in images. The training was done using the manually annotated KITTI dataset. The program correctly classified more than 95% of the image regions.
  • Vehicle Detection using HOG and SVM - Created a vehicle detection and tracking pipeline with OpenCV, histogram of oriented gradients (HOG), and support vector machines (SVM). Optimized and evaluated the model on video data from a automotive camera taken during highway driving.
  • PID Controller for Autonomous Driving - Using the knowledge of Control Systems to implement a Proportional–Integral–Derivative Controller (PID Controller or three-term controller) in C++ to manoeuvre the vehicle around the track. The program uses the cross track error (CTE) from pre-defined waypoints data and the velocity (mph) in order to compute the appropriate steering angle for vehicle maneuvering.

Big Data | Advanced Analytic | Data Engineering

  • Feature Selection on a data-set of Sparse Vectors - Working on developing and implementing feature selection techniques for customer conversion data in digital advertising domain. The aim is to identify a certain specific set of features for various campaigns that can be leveraged for better consumer targeting. This will also save a lot of resources on data-storage and make the prediction platforms run faster. Learning and implementing various algorithms based on Filter and Wrapper Methods like Anova, Fisher Score, Mutual Information, Bayes Error . Deploying the systems for test on Tera-bytes on data and optimizing the performance.
  • Customer Churn Prediction - Develop an end to end working system for the purposes of predicting Churn Prediction of a Telecommunications Company. Prototyping using Python sklearn and further system developed using Apache Spark and Spark MLlib. Utilized Decision Tree Classifier and improved accuracy with Random Forest and Grid Search. Gained understanding about how businesses can leverage the machine learning techniques to improve profits.
  • Customer Segmentation - Performed customer segmentation and clustering using the Call Data Records to better target customers for promotional contents and offers. Development using Apache Spark MLlib. Explored various attributes on which segmentation can be performed. Developed an end-to-end framework for fetching the user data records and classifying the users into distinct buckets on the basis of their usage behaviours.
  • Data exploration pipeline with Apache Spark - Developed an end to end system to extract data from HDFS and plot various parameters of the data using Oracle Visual Analyzer. Developed a program using Processing and Google Maps API to plot the data points on maps using Java applets.
  • Integrating Spark into OCAP - Successfully integrated and demonstrated the integration of Apache Spark into Oracle Communications Analytics Platform (OCAP) which earlier supported only Hadoop. Gained understanding of the ETL framework. Developed tools to generate relevant XML files from CSV files automatically.

Some Academic Projects

  • Optimal Power and Subcarrier allocation for Green Cognitive Radio by Combining Multi-objective Optimization with Differential Evolution - In this study, the problem of determining the power allocation that maximizes the energy efficiency of cognitive radio network was investigated as a constrained optimization problem and solved using Multi-objective Optimization with Differential Evolution (CMODE). The energy-efficient maximization problem was transformed into a parametric optimization problem and then solved to obtain an optimal solution.
  • Weather Monitoring System - Study the various wind resource assessment techniques and industry guidelines and to conduct a feasibility study of the wind turbine installations within the university campus. Identified the reasons for less than desired output from the current wind turbine installations and suggested new sites and turbine designs for future installations by using the results from the simulation of the wind flow around the entire campus using a 3D model constructed in Solidwork, Autodesk 3ds max and performing Computational Fluid Dynamics in Ansys Fluent. Developed a program in R for the purpose of cleaning and analysis of the data generated from the wind monitoring station on campus.