Sashank Gondala

sgondala@gatech.edu


I am a 2nd year Master's student in Computer Science (Specializing in Machine Learning) at Georgia Tech where I'm advised by Prof. Devi Parikh and also closely work with Prof. Dhruv Batra. For the summers of 2020, I worked as a Machine Learning Research Intern in the Language Modeling team at Apple.

In a past life, I heavily worked on Systems. I was a Senior Software Developer for 3 years at Oracle, where I worked on speeding up Oracle Analytics Platform, a C++ query engine. I also co-founded a startup to deliver a Decentralized CDN over blockchain. Though it didn't take off as expected, it was pretty exciting to work on!

I earned a Bachelor's degree in Computer Science in 2016 from IIT Bombay.

In my spare time I enjoy reading books, cooking, and travel.

E-mail  |  CV  |  LinkedIn  |  Github  |  Twitter  |  Scholar  |  Books



Research Interests

I’m interested in building intelligent agents that are capable of communicating with humans via natural language and are able to perform tasks. Towards this larger goal, I'm currently focusing on three main research directions

  • Teaching via language - Currently, most learning-based robots are taught via hundreds of demonstrations of a task. Humans, however, can be easily taught using natural language instructions. I am interested in using language as a medium of instructing and teaching robots.
  • Vision and Language grounding - Most existing image captioning models require paired training data, which is difficult to collect and annotate. If we were to deploy robots that communicate with humans in natural language, they would have to learn novel visual concepts without paired data.
  • Language generation - Most language models still aren't semantically accurate (GPT-2 writes about ‘fires happening under water’). I believe this is because current language models are learnt only from text, whereas human language understanding happens via visual and motor controls. Language models which incorporate visual and embodied signals will demonstrate better language understanding and are likely to produce semantically accurate sentences.


Publications
3DSP Error-driven Pruning of Language Models for Virtual Assistants
Sashank Gondala*, Lyan Verwimp*, Ernest Pusateri, Manos Tsagkias, Christophe Van Gysel.
Keywords: ASR, Language Model Pruning, discriminative, data selection, error prediction.
Work done as a Research Intern at Apple. Published at ICASSP 2021


Selected Projects
3DSP Test-time training for image captioning
Sashank Gondala, Rachel Zhong, Harsh Agarwal.

Worked on the idea of test-time training to improve performance of image captioning models on novel images. Incorporated several metrics that estimate the accuracy and fluency of captions without using paired data. Using these metrics as rewards, trained captioning models to optimize for them using PPO.


3DSP Vision & Language Navigation
Work under progress

Building models to better solve the task of Vision & Language Navigation in Continuous Environment.


Companies co-founded
3DSP Decentralized CDN

Co-founded a startup to provide decentralized CDN services by sharing the spare bandwidth and hard drive space of Internet users over blockchain. IP developed include a prototype to support HLS video streams and a whitepaper that describes the challenges and solutions including Service Certificates, Probabilistic Micropayments, etc.


Academic Achievements
  • Secured All India Rank 14 in IIT-JEE 2012 among 500,000 test-takers.
  • Secured All India Rank 8 in Nationwide Education and Scholarship Test 2014 and was awarded scholarship (Rs. 25,000/-) for the same.
  • Was placed National Top 1% in several Astronomy, Physics, and Junior Science Olympiads (2010-2012).
  • Attended (Top 35 students across India) Indian National Astronomy Olympiad(INAO) and Indian National Junior Science Olympiad(INJSO) Orientation-cum-Selection Camps held by HBCSE (2010)
  • Obtained 7th position in State Mathematics Olympiad (APAMT) (2009)

News
  • Aug 2020 → Started serving as the Graduate Teaching Assistant for the CS7643 Deep Learning.
  • May 2020 → Started Summer internship at Apple, working on Siri Language Understanding team.
  • Jan 2020 → Started serving as the Graduate Teaching Assistant for the CS8803 Special Topics in Systems for Machine Learning.
  • Aug 2019 → Started serving as a Graduate Teaching Assistant for the CS 7641 Machine learning
  • Aug 2019 → Started my MS degree in Computer Science at Georgia Tech.
  • Aug 2018 → Promoted to Senior Member of Technical Staff at Oracle.
  • Jul 2016 → Started working as a Member of Technical Staff in the Oracle Analytics team at Oracle.
  • May 2016 → Graduated from IIT Bombay with a Bachelor's degree in Computer Science.

Teaching
  • Teaching Assistant → Deep Learning, Georgia Tech, Spring 2021 (Scheduled).
  • Teaching Assistant → Deep Learning, Georgia Tech, Fall 2020.
  • Teaching Assistant → Systems for Machine Learning Research, Georgia Tech, Spring 2020.
  • Teaching Assistant → Machine Learning, Georgia Tech, Fall 2019.
  • Teaching Assistant → Intro to Computer Programming, IIT Bombay, Spring 2016.


"The best way to predict the future is to invent it." ~ Alan Kay
Design inspired from here