Tarun Bisht

RESEARCH INTERESTS

Research interests include Machine Learning, Computer Vision, Signal Processing and Generative AI.

EDUCATION

Ph.D in Computer Science and Engineering, IIT Bombay (2023 - Present)

Research - Trustworthy AI systems
CGPA - 8.7/10.0

M.S. Research in Industrial Engineering and Operations Research, IIT Bombay (2022 - 2023 [dropped])

Research - Deep Learning for Combinatorial Optimization Problems
CGPA - 8.08/10.0

M. Sc. in Computer Science, Kumaun University (2018 – 2020)

Research - Application of object detection in home surveillance system
University Rank - 1 &#124 Vice Chancellor Gold Medal

B.Sc. in Computer Science, Physics and Mathematics, Kumaun University (2015 – 2018)

WORK EXPERIENCE

Research Co-Coordinator | The Sound of AI OSR, Remote | 2020 - 2022

Introduced transformer-based models into the automatic speech-to-text component of NeuralNet music synthesizer. Achieved a 20% reduction in transcription error rates through model fine-tuning, data augmentation techniques, and integration of decoding algorithms with language models.

Co-Founder and Developer | The Gamer Studio | (2017 - 2019)

Co-founded and led a game and app development venture, successfully delivering 8+ games, 3+ mobile applications, and 2+ software solutions, with 9 products published on Google Play Store, achieving significant market traction (two applications exceeding 40K+ and 13K+ downloads respectively)

SELECTED RESEARCH PROJECTS

NLP for Optimization Problems, Techfest 2023

Modelling optimization problems from natural language text descriptions into solver-compatible modelling language. A key emphasis is leveraging language models and extracting valuable representations from scarce data resources due to the labour-intensive nature of annotating data for optimization problems.

Direct speech-to-speech translation with Voice Interpolation, WiSSAP 2023

Direct speech-to-speech translation between languages, bypassing text as an intermediate step. Designed a latent space to represent and interpolate speaker voice characteristics, enabling dynamic voice modifications. Generated speech variations for both the same speaker and different speakers. Focused on enhancing naturalness and flexibility in multilingual voice synthesis.

NeurIPS Large Language Model Efficiency Challenge, NeurIPS Challenge 2023

Adapted a foundation model to specialized tasks by fine-tuning on a single GPU (4090 or A100 40GB) within a 24-hour timeframe. Secured a top 10 ranking in the 4090 GPU track.

AI-based surveillance system, Kumaun University 2020

Developed an affordable alert system for detecting monkey theft with high accuracy by training advanced single-shot object detection algorithms. Optimized the system’s performance by converting the trained model to TensorFlow Lite, increasing detection speed from 1 FPS to 24 FPS, enabling deployment on Raspberry Pi. Additionally, provided a software solution compatible with existing IP cameras.

TEACHING EXPERIENCE

Teaching Assistant (TA) - EE739 Processor Design | Spring 2024-2025
Teaching Assistant (TA) - IE643 Deep Learning Theory and Practice | Autumn 2023-2024
Freelance Educator - Computer Science, Physics and Mathematics | 2017-2020
Create and publish educational tutorial blogs on my personal website to share knowledge and insights

PUBLICATIONS

Tarun Bisht, Sarath Babu, and Virendra Singh | Critical Behavior Sequence Monitoring for Early Malware Detection | SINCONF 2024 | Read
The Sound of AI Community | From Words to Sound: Neural Audio Synthesis of Guitar Sounds with Timbral Descriptors | AIMC 2022 | Read

TALKS

Generative AI and Large Language Model (LLM) | Samsung Open Source Conference (SOSCON 2023) | Samsung R&D Institute Bangalore, India | Invited Speaker

ACHIEVEMENTS AND AWARDS

Recipient of the Vice Chancellor’s Gold Medal for outstanding academic achievement at the 17th Convocation of Kumaun University Nainital
Awarded UGC NET Junior Research Fellowship (2021) for Phd in Computer Science
Qualified UGC NET (2020) for Assistant Professor in Computer Science
Qualified GATE (2021) in Computer Science and Engineering organised by IIT Bombay
Rank 2 out of 41 in object detection competition organized by dockship.io