Divyam Goel

Divyam Goel

Deep Learning Junior Researcher

Rephrase AI

Biography

Hi! I am a Deep Learning Junior Researcher at Rephrase AI and am also a research assistant at the GIST Computer Vision Lab. I recently graduated from the Indian Institute of Technology (IIT) Roorkee with a bachelor’s degree in Electronics and Communications Engineering. My research interests revolve around deep learning, and its applications in natural language and computer vision for embodied agents.

Previously, I spent time as a Mitacs Globalink Research Intern advised by Prof. Fatemeh H. Fard at the University of British Columbia (Summer 2021) where I led a collaborative effort to develop efficient but accurate language models for software engineering tasks. I have also worked as a Research and Development Intern at Pando Labs (Fall 2020). We developed novel deep reinforcement learning solutions for operations research problems in the supply chain industry.

At IIT Roorkee, I was a core member of a Deep Learning research centric reading group named the Vision and Language Group, and the IEEE Student Chapter, IIT Roorkee.

Interests

  • Computer Vision
  • Deep Learning
  • Embodied AI
  • Natural Language Processing

Education

  • B.Tech. in Electronics and Communications Engineering, 2022

    Indian Institute of Technology Roorkee

News

  • June 2022 Joined Rephrase AI as a Deep Learning Junior Researcher!
  • June 2022 Paper titled “Language Guided Meta-Control for Embodied Instruction Following” accepted at the Embodied AI Workshop @ CVPR 2022! Preprint coming soon.
  • May 2022 Paper titled “Leveraging Dependency Grammar for Fine-Grained Offensive Language Detection using Graph Convolutional Networks” accepted at SocialNLP@NAACL 2022! Preprint coming soon.
  • March 2022: Paper “On the Cross-Modal Transfer from Natural Language to Code through Adapter Modules” accepted at ICPC 2022! Preprint coming soon.
  • Feb. 2022: Recipient of the IIT Roorkee Heritage Foundation Excellence Award for 2021!
  • July 2021: Joined Prof. Jonghyun Choi’s GIST Computer Vision Lab, South Korea (virtually) as a Research Assistant.
  • June 2021: Headed to Prof. Fatemeh H. Fard’s lab at UBC, Canada (virtually) to work as a Mitacs Globalink Research Intern.
  • Dec. 2020: Selected for the prestigious MITACS Globalink Research Internship 2021!!
  • July 2020: Will be working at PandoCorp as a research intern this fall in their machine learning research lab - Pando Labs.
  • March 2020: Recipient of the IIT Roorkee Heritage Foundation Excellence Award for 2019!
  • April 2019: Selected for the Quantum Computing summer training organized by QuLabs.

Experience

 
 
 
 
 

Deep Learning Junior Researcher

Rephrase AI

Jun 2022 – Present Bangalore, India
  • Part of the research arm of the company aimed at developing tech to generate highly personalised video from input text scripts.
  • My research mainly revolves around deep learning for computer vision and speech audio with forays into natural language processing.
 
 
 
 
 

Research Assistant

GIST Computer Vision Lab

Jul 2021 – Present Gwangju, South Korea (virtually)
  • Working under the aegis of Prof. Jonghyun Choi to improve the current state of vision and language navigation for embodied agents.
  • Collaborated with Prof. Dongyeop Kang on a research project which aims to build multi-modal transformers for story visualization.
 
 
 
 
 

MITACS Globalink Research Intern

University of British Columbia

Jun 2021 – Sep 2021 Kelowna, Canada (virtually)
  • Led a collaborative effort to develop efficient but accurate language models for software engineering tasks.
  • Exploited cross-modal transfer of adapters to achieve 140×better parameter budget and ∼95% efficient storage over the state-of-the-art.
  • Built the Super Code Clone Detection-88 (SCD-88) dataset to evaluate the proposed approach on python-specific code clone detection.
 
 
 
 
 

Research and Development Intern

Pando Labs - PandoCorp PLC

Aug 2020 – Nov 2020 Chennai, India
  • I worked on developing a practical Deep Reinforcement Learning solution to optimize shipping costs in the Supply Chain industry.
  • Achieved ~85% packing efficiency in 3D bin packing and ~50% cost efficiency in capacitative vehicle routing over existing solutions using off-policy agents in simulated environments.
  • Re-architected a heuristic algorithm to improve efficiency and allow parallelised operations on GPUs.
 
 
 
 
 

Undergraduate Student Researcher

Machine Vision Lab

Dec 2019 – Mar 2020 IIT Roorkee
  • Worked under the supervision of Prof. Balasubramanian Raman on a project titled - Medical Image Segmentation and Classification Problems.

Achievements

IIT Roorkee Heritage Foundation Excellence Award 2021

  • Awarded to 45 students from a pool of over 5000 undergraduates.
  • The Excellence Awards recognize students for outstanding academic, co-curricular and extra-curricular achievements.

MITACS Globalink Research Internship

  • Selected for the prestigious MITACS Globalink Research Internship 2021 program.
  • The Mitacs-SICI partnership awards this research grant to top international undergraduates for a research internship under the supervision of Canadian university faculty members.
See certificate

IIT Roorkee Heritage Foundation Excellence Award 2019

  • Awarded to 45 students from a pool of over 5000 undergraduates.
  • The Excellence Awards recognize students for outstanding academic, co-curricular and extra-curricular achievements.
See certificate

Quantum Computing Summer Training

  • Among the 15 students selected for the Quantum Computing Summer Training - 2019.
  • The course comprised of intensive lectures on linear algebra, quantum mechanics and topics in quantum computation including entanglement, multi-qubit computation, quantum key distribution and quantum algorithms.

Featured Projects

*

Super Code Clone Detection - 88

A dataset consisting of 11,400 python codes for a retrieval based code clone detection.
LINK

HyperspecAE

PyTorch reproduction of the paper titled ‘Hyperspectral Unmixing Using a Neural Network Autoencoder’ (Palsson et al. 2018).
CODE

Customer Churn Management

Led a collaborative effort to develop an agent that predicts consumer churn in the Indian Telecom Industry.
REPORT DEMO

Policy Gradient Methods

Implemented some Policy Gradient RL algorithms (using PyTorch) on the CartPole environment by OpenAI.
CODE

Indic-Language Translation

Trying to utilize a large pre-trained language model to build a English-Hindi machine translation system.
CODE

Sentiment Analysis with BERT

Fine-Tuning a BERT model for Sentiment Classification on the SMILE: Twitter Emotion Dataset.
CODE

NNVisualizer

Developed a Web Application using Flask and StreamLit for the visualization of the weights on neural network.
CODE

Self Driving Car

Designed an autonomous vehicle using a combination of image processing and IoT methods.
CODE DEMO