Avatar

Raunak Sinha

Staff Research Engineer, Artificial Intelligence

IBM Research

Hey!

I am an ardent researcher interested in exploring and developing machine learning for computer vision, computational sustainability, and natural language understanding.

I am currently working on utilizing machine learning to jointly understand the semantics of programming languages and natural languages, making it easy to comprehend code. I am also working on recommending microservice partitions for a given piece of code for application modernization.

Previously I have worked on using machine learning to capture facial features to generate images of family members. I have also developed an open-source toolkit for freely experimenting and interacting with Generative Adversarial Networks in a code-less fashion.

Interests

  • Machine Learning
  • Computer Vision
  • Artificial Intelligence
  • Natural Langauge Processing
  • Computational Sustainibility

Education

  • BTech Computer Science and Engineering, 2019

    Indraprastha Institute of Information Technology, Delhi

Projects

Natural Language Code Search

Structured understanding of programming languages, like C and Java, by linking them to the semantics of human-understandable (natural) …

Generating kin images using Generative Adversarial Networks

Establising the problem statement on generating images of kin (family members). FamilyGAN achieves ~2% variation in kin-verification …

Recommending Microservices

Improving micro-service recommendation by detecting communities through code API calls for application modernisation

AuthorGAN: Improving GAN Reproducibility using a Modular GAN Framework

Toolkit for code-less authoring of GANs. Introducing a new abstractive view of the GAN framework which unifies various GANs. …

Tranfering Adversarial Perturbation

Deep learning models have been shown to perform extremely well for many image based tasks. Such tasks include image classification, …

Multi-label Triplet Embeddings for Image Annotation from User-Generated Tags

To predict (extract) semantic labels for a given image. This is achieved by using triplet loss on latent embedding (features) for …

noWhinge

Got a complaint, don't whinge!

Facial retouching detection using Subclass Restricted Boltzmann Machines

Utilizing subclass information to detect facial retouching in images, by incorporating ‘L2,1’ loss in the formulation of …

Monocular SLAM

Simultaneous location and mapping(SLAM) is the problem of creating a map of an unknown environment and simultaneously tracking agent’s …

Github Stack-Overflow User Recommendation System

Recommending users to Github repositories

Predicting Trajectory of Basketball Shots

Detecting moving basketball using advanced image analysis techniques for segmentation and extraction. Mapping ball trajectory for …

Recommendation System on MovieLens dataset

Almost all online services use a recommendation system in one way or other. Even though the field is widely explored, it still remains …

Experience

 
 
 
 
 

Staff Research Engineer, Artificial Intelligence

IBM Research

Jul 2019 – Present New Delhi, India
  • Using machine learning to understand the intricate semantics of programming language. Constructing a programming language agnostic framework to understand varied languages. Part of IBM Research's agenda on AI for code.
  • Accomplishing Natural Language Code Seach, by learning joint distribution for code and natural language (English). Improving retrieval accuracy by 16% by the novel use of siamese networks.
  • Exploring community detection and clustering algorithms on code artifacts (such as code classes) for recommending microservices from monolithic applications and programs.
 
 
 
 
 

Research Intern

IBM Research

May 2018 – Dec 2018 Bengaluru, India
  • Developed IBM GAN-Toolkit for code-less authoring of Generative Adversarial Networks. Formalized the ‘abstractive’ view of GAN framework, which unifies various GANs. Unification lets users to commix different GAN components making learning and experimenting with GANs easier.
  • The toolkit supports PyTorch, Keras and Tensorflow.
 
 
 
 
 

Undergraduate Researcher

Image Analysis and Biometric Lab - IIITD

Jan 2018 – May 2019 New Delhi, India
  • Established the research statement for generating the facial images of family members (kin) from a given image and the relationship type.
  • Developed a Generative Adversarial Network with updated training protocol and loss function for learning the features heredity between family members
  • The proposed method achieves ~2% variation in kin-verification scores for original and generated face-image pairs.