ABHISHEK GOYAL

As a Data Scientist at my current organization, I have been working on product development problems using deep learning algorithms for Automotive Driving Scenarios. I spend most of my time reading and implementing state-of-the-art research papers in tensorflow and pytorch. Unsupervised Domain Adaptation: Deep learning algorithm development for unsupervised domain adaptation for generating semantic segmented images for synthetic dataset based on entropy minimization. Goal of the project was to use realistic images along with semantic segmented counterpart and develop a model that generated semantic images for SSG dataset. Water Droplet Dataset Generation: Alignment of images from fish eye stereo cameras using rigid (homography) and non-rigid transforms. Training the processed dataset on pix2pixHD for waterdroplet synthesis on fisheye camera images. Semantic to photorealistic image synthesis: Developing and implementing GAN based algorithms for image domain translation - transforming output of gaming engine to photo-realistic images. Implementing SPADE paper in pytorch. Developing and improving results for semantic segmentation using Deep Learning architectures like UNET on limited dataset. Using self-supervised techniques to learn better visual representation of the images and transfering the learned weights for improving the performance of the UNET model. Evaluating performance of GAN generated images using image evaluation metrics such as NIQE score.

Stationary Vehicle classification using Radar Dataset. Developing Algorithm to obtain data for vehicle from the radar data consisting of data related to moving objects, stationary objects such as (vehicles, pedestrian, gantry sign, bridges, guard rail, etc.). This includes trying various clustering algorithms (such as DBSCAN, K-Means, etc.) and various distance metric such as Euclidean distance and Mahalanobis distance. Performing Exploratory Data Analysis to identify features of interest and identifying patterns in the data enabling us to decide which model would perform better with the data in hand. Data Preprocessing to balance the imbalance dataset. Applying machine learning classification algorithms (Logistic Regression, SVM, KNN, Neural Network and LSTM) and comparing there performance for classifying truck and car objects.

Automating classification of patents/ invention ideas into different IPC classes by developing a deep learning based text classifier. Developing python scripts for automating various tasks related to web scraping, relevant document retrieval, portfolio analysis of patents, etc. Infringement Analysis, Licensing, EoU generation, Patentability search.