Simran Deepak Makariye
Interests : Data Science, Deep Learning and Computer Vision (Generative AI)
About Me
Hello! I'm Simran Makariye , pursuing my M.S. degree in computer science at NYU-Courant. Currently interning at Audible, I'm working on some interesting recommendation systems, employing reinforcement learning to deliver personalized content to customers. Previously, I spent over 2 years at Qualcomm as a software engineer, focusing on machine learning and location technology. Additionally, I've collaborated on numerous deep learning and computer vision projects with professors/researchers from NYU and Georgia Tech.
My primary areas of interest are deep learning, computer vision, and Generative AI. Beyond building these models to address real-world challenges, I'm interested in developing more efficient and scalable AI solutions, making them practical and accessible. Therefore, I'm actively learning about accelerating deep learning workloads and optimizing compute utilization.
I am always open to collaborate with like-minded individuals who share my passion towards deep learning. If you have any intriguing research projects or work opportunities that align with my interests, please do reach out over email at sdm8499 [at] nyu [dot] edu. I am eager to continue learning and making meaningful contributions to these exciting fields.
Projects
Hate-LLaMA: An Instruction-tuned Audio-Visual Language Model for Hate Content Detection
Fine-tuned Video-LLaMA, which effectively processes video and audio information multimodally, on a hate video dataset using 4 RTX8000 GPUs, achieving an F1-score of 73% on the unseen test dataset. Additionally, released a benchmark dataset comprising 300 labeled hate/non-hate videos to address the scarcity of labeled hate video datasets. Also, deployed a demo for Hate-LLaMA (Multimodal LLM) on HuggingFace Spaces using Gradio.
Video Frame Prediction and Semantic Segmentation with Self-Supervised Learning
Implemented a fully CNN architecture - SimVP - for video frame prediction and trained a U-Net architecture for Semantic Segmentation on a diverse synthetic dataset consisting of video clips of 3D moving objects achieving a Jaccard Index of 0.251 with minimal training and fine-tuning.
Multi-Image Fusion and Semantic Segmentation for Power Substation Delineation
Collaborating with Transition Zero, a leading clean energy technology company, to develop deep neural models that fuse multiple revisits of low-resolution Sentinel-2 L1C imagery and generate a super-resolved image and then applying semantic segmentation task to accurately delineate power substations from the super-resolved image.
Integrated Content Management Platform - New York Public Library
Developing a generalized content management pipeline for the NYPL website, incorporating deep learning techniques to enhance collection item accessibility and enable automated meta-data tagging. Also, integrating crowdsourcing technologies to further refine and optimize these solutions, resulting in improved user experience and streamlined content management processes.
Semantic Role Labeling on NomBank Dataset
Implemented a fully CNN architecture - SimVP - for video frame prediction and trained a U-Net architecture for Semantic Segmentation on a diverse synthetic dataset consisting of video clips of 3D moving objects achieving a Jaccard Index of 0.251 with minimal training and fine-tuning.
Smart Mental Health Monitoring using Wearable Sensors
Designed and implemented an LSTM-based stress detection model using time-series data consisting of both wrist features and chest features emulated as wrist features. Achieved a remarkable accuracy of 90%.
Human Neuron Inspired CNN architecture for Music Emotion Recognition
Designed a CNN-based Music Emotion Recognition model that mimics the way a human brain responds to an emotion i.e. by firing a specific set of neurons for each emotion. It yielded an accuracy of 82% with the additional benefit of fewer computational parameters.
Work Experience
September 2023 - Present
Audible, Amazon
Data Science Intern - Product & Insights Team
July 2023 - October 2023
Georgia Institute of Technology
Research Assistant, Advisor: Prof. Eva Dyer
Research focus: DL in Neuroscience
May 2023 - September 2023
New York Univesity
Research Assistant, Advisor: Prof. Vasant Dhar.
Research focus: Deep Learning in Finance
June 2020 - July 2022
Qualcomm
Software Engineer - Location Technology Team
May 2019 - July 2019
Qualcomm
Software Engineering Intern - Location Technology Team
May 2018 - July 2018
Indian Institute of Technology, Hyderabad
Research Intern - Advisor: Prof. Manohar Kaul
Education
September 2022 - Present
New York University (Courant Institute of Mathematical Sciences)
Master of Science in Computer Science
Coursework:
Large Language and Vision Models, Deep Learning, Computer Vision, Big Data and ML Systems, Cloud and Machine Learning, Natural Language Processing, Algorithms, Operating Systems.
Teaching Assistant:
1. Intro to CV (Taught by Prof. Jean Ponce) - Spring 2024
2. Deep Learning (Taught by Prof. Yann Lecun and Prof. Alfredo Canziani) - Spring 2024, Fall 2023
3. Programming Bootcamp - Math in Finance - Summer 2023
July 2016 - June 2020
National Institute of Technology, Tiruchirappalli
B.Tech in Computer Science and Engineering and Minor Degree in Management
Coursework:
Natural Language Processing, Machine Learning, Probability Theory, Linear Algebra, Optimization Techniques.
