Education
Graduate Tutor and Grader for the class of Introduction to Deep Learning
Coursework
Large Language Models(LLM), Introduction to Deep Learning, Introduction to Machine Learning, Big Data Processing, Practical AI, Data Management, Data Analysis, Ethical & Legal Issues in Data Science.
In Practical AI course, I also worked with Nvidia’s Jetson Orin Nano and JetBot AI Robot Kit to implement latest ML models through Dockers.
Publication
Efficient Deep Learning Models for Facial Expression Recognition (FER) using Transformers
- Developed efficient Transformer-based computer vision algorithm for Facial Expression Recognition (FER).
- Achieved competitive accuracy while maintaining computational efficiency, distinguishing from top-tier large models.
- Presented paper at IEEE BSN hosted at the MIT Media Lab, Cambridge! Link
- Extended the work across diverse massive datasets achieving state of the art results.
Transformer-Based Emotion Recognition with EEG
- In this paper, we introduce an innovative approach for understanding emotions through brain signals, leveraging electroencephalography (EEG) and Transformer-based models.
- Designed and implemented a novel Transformer architecture tailored for processing multiple EEG signals simultaneously, overcoming the limitation of traditional transformer models that handle only one sequence at a time.
- Demonstrated the significance of the breakthrough by enhancing the decoding of emotional states from EEG data, with potential applications in improving brain-computer interfaces and human-robot interaction.
- Achieved higher accuracy in predicting emotions, with mean accuracies exceeding 91% for both valence and arousal levels, establishing our method as a reliable tool for advancing EEG-based emotion recognition.
Work Experience
ML | AI | Computer Vision | Natural Language Processing
Machine Learning Researcher
Sensorimotor Control Laboratory, UMBC – A Lab specialized in Brain-Machine interface
• Designed and developed machine learning models; efficiently managed model versions using MLflow for testing and validation
• Co-authored paper on efficient transformers algorithm for facial expression recognition.
• Leveraged Transformers-based algorithm for EEG, boosting accuracy by 32%, with tailored multi-sequence input for transformer encoder
• Engineered emotion-based speed control for a robotic arm, showcasing seamless human-robot collaboration.
Machine Learning Engineer
WebOccult Technology, India – A Software development company
Text-to-Image search
• Developed a pipeline for images text-to-image search, enabling efficient querying of a million-image dataset via text input
• Integrated Object Detection, Color Analysis, OCR (Optical Character Recognition), Vector Similarity, improving search and retrieval accuracy
• Utilized NLP (Natural Language Processing) techniques including keyword extraction and text cleaning.
Real-Time Hand and Body Pose Detection for Interactive Gaming
• Integrated and implemented real-time detection of hand and body pose to Unity game engine with Python API endpoint.
• Enabled depth estimation using OpenCV AI Kit (OAK-D camera), resulting in performance improvement.
Adjunct Lecturer – Machine Learning
Bath Spa University, U.A.E.
• Taught two full-credit courses, covering fundamentals of machine learning, computer vision and Python programming.
• Graduate students applied their knowledge to real-world problems and develop their own projects.
Artificial Intelligence Engineer
Doctor on Click, Singapore – Provides remote healthcare
• Implemented image segmentation with deep learning for Traditional Chinese Medicine with an accuracy of 0.95 IoU.
• Resulted in a web app for the doctors to get the segmented tongue image and the diagnosis of the patient.
Instructor, Python
Akshar CompuSoft Education Center, India
• Taught fundamentals of Python, data science, machine learning, and provided individualized mentorship to students.
Projects
ML | AI | Computer Vision | Natural Language Processing
TL-GAN (Transparent Latent Space Generative Adversarial Network)
- Programmed TL GAN to better understand the latent space and improve control over GAN-generated features.
- Controlled human face features with the linear model and visualized the latent space with the multi-label classifier.
Song Lyrics Generator with Large Language Model(LLM)
- Created a song lyrics generator capable of producing pop and rap lyrics from a limited dataset.
- Leveraged transfer learning on OpenAI’s GPT-2 (small) with 124M parameters to develop the model.
Real-time writing on live web-camera screen
- Developed a real-time interactive system enabling touchless handwriting on live web-camera screens through advanced hand tracking technology.
- The project gained viral success on LinkedIn with 2.5 million views. Post link.