Vyshnavi Gutta

I am a second year graduate student in the department of Computer Science at Georgia Institute of Technology. I am presently working on improving continual learning methods in budget/compute-restricted environments under the guidance of Professor Zsolt Kira. Previously, I have worked on developing universal peturbation attacks for image & video-level privacy protection against deep object detection networks under the guidance of Professor Ling Liu. My interests span a broad range of sub-fields in Computer Science, including Deep Learning, Machine Learning, Computer Vision and Natural Language Processing. I have worked on various projects which involved extracting meaning from text and images.

Prior to pursuing my Master's I was working as a Data Scientist at Reliance Jio, working on Intent modelling, Machine translation and diving into Natural Language Processing for chat-bot automation. I started working at Jio after graduating from IIIT Hyderabad where I did my major in Computer Science.

In my free time I like to read books and appreciate the unexplored beauty of nature.

Email: vgutta7@gatech.edu

LinkedIn  /  Resume

CODA-Prompt: COntinual Decomposed Attention-based Prompting for Rehearsal-Free Continual Learning

Contributed in developing CODA-Prompt for continual learning (CL) which uses a novel attention-based end-to-end key-query scheme to learn a set of weighted expanding prompt components conditioned on inputs.

Replaces the prior state-of-the-art method DualPrompt as the new SOTA, beating it by 5.4% and 6.6% in Average accuracy on CL benchmarks with both class and domain-incremental task shifts.

Accepted at IEEE/CVF Conference on Computer Vision and Pattern Recognition 2023 (CVPR, 2023).

Syllabified sequence-to-sequence attention for transliteration

Developed Syllabified Seq2Seq net for transliteration. Scaled it to 8 language scripts using Sonority sequencing.

E2E tested using Azure CI/CD pipeline before containerizing with Docker. Deployed and currently hosted on Kubernetes server using Amazon AWS. Serving ∼ 2000 clients daily.

Accepted at PACLIC, 2022.

An improved human-in-the-loop model for fine-grained object recognition with batch-based question answering

Successfully demonstrated cost-sensitivity in the existing fine-grained recognition approaches and proposed Recognition via Image and batch-based local question answering to this end.

Proposed a novel dynamic cluster-centric local question mining method based on the concept of locality degree of an attribute in a cluster.

Adversarial gradient attacks on real-time object detection

Developing a universal adversarial perturbation technique for video-level privacy preservation against state-of-the-art deep object detection networks.

Deep reinforcement learning for autonomous driving

Working on leveraging deepRL techniques for improving average return performance in AD simulation environments.

Representation learning via multi-view contrastive selection

Optimal view-selection for obtaining better image representations with info-min and multi-view contrastive learning.

4.5% improvement overall with respect to State-of-the-art methods.

Geo-location prediction on twitter data
Developed GeoAttn, a scalable and memory-efficient deep CNN model for location prediction from tweets.

Natural language understanding for medical sense disambiguation in user dialogue

Unseen symptom, attribute, value extraction & mapping to structured data given an utterance.

Implemented Induction network for few shot symptom recognition via text classification and Capsule neural net as further re-enforcement for joint intent detection and slot filling.

Crop darpan: engineering crop diagnosis framework

Led and designed Hierarchy-based knowledge acquisition and Coverage-set based question mining models on visual symptoms of an affected crop for disease prediction with question answering.

Graduate Teaching Assistant, Georgia Tech | Aug 2021 - Present

Working as a teaching assistant for CS4400, Intro to databases.

Data scientist, Reliance Jio | Nov 2019 - Jul 2021

Core member of the R&D team. Worked as a researcher in the natural language processing domain.

Designed and successfully completed projects spanning context-aware Q/A, machine translation, intent modelling and entity detection.

Undergraduate researcher, IIIT Hyderabad | Aug 2018-Oct 2019

Successfully demonstrated cost-sensitivity in the existing fine-grained recognition approaches and proposed Recognition via Image and batch-based local question answering to this end.

Proposed a novel dynamic cluster-centric local question mining method based on the concept of locality degree of an attribute in a cluster. The work was published at a top Data science conference (CODS-COMAD'20)

Research assistant, IIIT Hyderabad | Aug 2017 - Aug 2018

UI designer for the project Climate change adaption and prediction.

Engineering a crop diagnosis framework, CropDarpan, for early pest prediction affecting crops. As part of this, led and designed Hierarchy-based knowledge acquisition and Coverage-set based question mining models on visual symptoms of an affected crop for disease prediction with question answering.

Teaching assistant, IIIT Hyderabad | Jan 2016-August 2017

Served as teaching assistant for Database systems, Science II and Mathematics II.

Responsibilities include holding tutorial sessions, quizzes, as- sistant hours and labs for a class of 150+ undergraduate students.

Master of Science in Computer Science

Georgia Institute of Technology, Atlanta | Specialization in Machine Learning Expected Graduation: May 2023.

B.Tech and M.S by Research in Computer Science Engineering.

IIIT Hyderabad | Aug 2014 - Oct 2019.

inspired from this website