CV
Click the pdf button to download my Resume.
Basics
Name | Prajjwal Gupta |
Label | Graduate Student |
gupta.praj@northeastern.edu | |
Url | https://ma1VAR3.github.io/ |
Summary | A graduate student at Northeastern University with research interest in machine learning privacy and security, explainable AI, and adversarial machine learning. |
Work
- 2023.08 - Present
Teaching Assistant
Northeastern University
Reviewed material, graded assignments, and conducted office hours for the courses: Introduction to Data Mining and Machine Learning in Fall 2023 and Foundations of AI in Spring 2024 semesters.
- Data Mining
- Reinforcement Learning
- Deep Learning
- Search and Optimization
- AI Ethics
- 2022.12 - 2023.07
Research intern
Indian Institute of Science (India Urban Data Exchange)
-Developed a cutting-edge privacy-preserving mean estimation algorithm, ensuring user-level privacy in a non-IID setting. Achieved exceptional results, surpassing existing approaches with a significant 20% increase in utility and an outstanding 36% enhancement in privacy.
-Designed a privacy-preserving approach for generative models based on quantization of latent space and exponential mechanism with differential privacy guarantees.- Data Mining
- Reinforcement Learning
- Deep Learning
- Search and Optimization
- AI Ethics
- 2022.02 - 2022.07
Federated Learning Consultant
DynamoFL
-Implemented a pipeline for federated neural collaborative filtering from scratch for recommendation systems.
-Integrated federated learning optimizers and aggregators for models like Xgboost and K-Means clustering.
Undertook the development of end-to-end solutions with the dynamofl framework for diverse client use cases.- Federated Learning
- Recommendation Systems
- XGBoost
- Clustering
- 2021.08 - 2022.01
Research Intern
National Institute of Technology, Kurukshetra (Security and AI Laboratory)
-Created a data poisoning attack for collaborative learning settings, which resulted in an adversarial success rate that was over 3 times improvement over the existing error-generic attacks.
- Investigated the use of encoder based models in distributed learning to achieve cross-domain generalization of network intrusion in a multi-client setting.- Federated Learning
- Adversarial Machine Learning
- Instrusion Detection
Education
-
2023.09 - 2025.04 Boston, MA, USA
MS
Northeastern University
Artificial Intelligence
- Unsupervised Machine Learning
- Foundations of AI
- Algorithms
- Program Design Paradigms
-
2019.08 - 2023.04 Vellore, India
BTech
Vellore Institute of Technology
Computer Science and Engineering
- Machine Learning
- Natural Langugage Processing
- Parallel and Distributed Computing
- Data Privacy
Awards
- 2023.05.15
Raman Research Award
Vellore Institute of Technology
Award for the publication 'A Novel Data Poisoning Attack in Federated Learning based on Inverted Loss Function'
Publications
-
2024.02.28 -
2023.07.01 A Novel Data Poisoning Attack in Federated Learning based on Inverted Loss Function
Elsevier Computers & Security
-
2022.09.27 Applications of IoT and Cloud Computing: A COVID-19 Disaster Perspective
Springer New Frontiers in Cloud Computing and Internet of Things
-
2022.06.24 A study of gene characteristics and their applications using deep learning
Springer Handbook of Machine Learning Applications for Genomics
-
2022.06.23 Early diagnosis of retinal blood vessel damage via deep learning-powered collective intelligence models
Hindawi Computational and Mathematical Methods in Medicine
Skills
Programming Languages | |
Python | |
R | |
Java | |
SQL |
Frameworks | |
Tensorflow | |
PyTorch | |
Keras | |
Scikit-learn | |
Huggingface Transformers | |
Flask |
Libraries and Databases | |
VectorDB | |
Numpy | |
Pandas | |
NLTK | |
OpenCV | |
Plotly | |
Matplotlib | |
Seaborn |
Tools | |
Git | |
Docker | |
Kubernetes | |
Kubeflow | |
BigQuery | |
VertexAI |
Skills | |
Deep Learning | |
Computer Vision | |
Reinforcement Learning | |
Natural Language Processing | |
Data Analysis |
Languages
English | |
Fluent |
Hindi | |
Native |
Projects
- 2021.10 - 2022.02
DAIDNet IDS
Conceptualized and implemented a novel deep learning model for the two-stage classification of network intrusions. DAIDNet leverages multiple autoencoders to learn patterns from distinct data subsets and has demonstrated state-of-the-art (SOTA) or near-SOTA performance on benchmark datasets.
- Autoencoders
- Tensorflow
- 2023.01 - 2023.05
FedPartial
A framework to enable collaborative learning across multiple hospitals, each using a different model architecture. The results demonstrated a remarkable 66% gain in communication efficiency, ensuring comparable utility to baseline models for a chest x-ray image classification task using the EfficientNet family of models.
- Federated Learning
- Tensorflow