Ashish Pandey

Logo

View My GitHub Profile

Ashish Pandey

Welcome to my portfolio website! This site showcases my skills, projects, and professional experience in data science and machine learning. You can find detailed information about my background, education, and projects below.

About Me

I am a passionate Data Science Master’s student at the New Jersey Institute of Technology. My academic and research journey has been focused on exploring innovative solutions in machine learning, deep learning, and natural language processing. Specifically, I am interested in developing faster alternatives to transformers by leveraging state-space models. As a Research Assistant, I have worked on projects ranging from dynamic graph embedding to Alzheimer’s disease prediction, where I have developed models like DG-Mamba that push the boundaries of existing technologies.

Beyond research, I have practical experience in cloud infrastructure optimization, data pipeline implementation, and software development from my time at Accenture. I am deeply committed to advancing research that not only challenges the status quo but also has a tangible, positive impact on human well-being. Through my work, I strive to contribute to fields that can make a difference in the world.

Skills

Education

Master’s in Data Science | GPA: 3.75 / 4.0
New Jersey Institute Of Technology, Newark, NJ
September 2023 - May 2025

Projects

Prudential Dataset

Prudential Dataset Led a risk assessment optimization project for Prudential Life Insurance, developing a machine learning algorithm to assign risk factors using minimal customer data. Utilized Principal Component Analysis (PCA) and boosting algorithms on a 2015 dataset, achieving top Kaggle performance with 23% fewer features, while maintaining significant variance retention.

Malaria Topic Modelling Project in R

Malaria Topic Modelling Developed and implemented a data processing pipeline to systematically extract, clean, and organize data from a vast collection of malaria-related articles. Utilized clustering and natural language processing (NLP) techniques to identify main themes and significant keywords, revealing prevalent research areas and enhancing insights into the malaria landscape.

Multimodal Forecasting Model for Stock Price Prediction

Stock Price Prediction Developed an advanced oil price prediction system combining ARIMA and GARCH models for price and volatility Time Series forecasting, respectively. Achieved a 17% performance enhancement over traditional attention mechanisms by integrating the novel MAMBA model. Optimized system performance with LSTM models via Optuna, incorporating XGBoost and leveraging historical data and news sentiments for improved predictive accuracy.

Conference/Talks

Professional Experience

Teaching Assistant - Deep learning (DS 677) | NJIT | Newark, NJ

September 2024 - Present

Research Assistant | NJIT | Newark, NJ

May 2024 - August 2024

Associate Data Engineer | Accenture | Mumbai, IN

July 2021 – April 2023

Publications

A Comparative Study on Dynamic Graph Embedding based on Mamba and Transformers

This paper delves into the exciting field of dynamic graph embedding, which is crucial for understanding the evolving nature of networks in various domains. The key focus of my research was to address the scalability limitations of transformer-based models, which, despite their ability to capture long-range dependencies in temporal graph data, face challenges due to their quadratic computational complexity.

To tackle this, I introduced three novel models: TransformerG2G with graph convolutional networks, DG-Mamba, and DG-Mamba with graph isomorphism network edge convolutions. The DG-Mamba models leverage the state space model architecture, specifically the Mamba model, known for its linear complexity and efficiency in handling long sequences.

Through extensive experiments on multiple benchmark datasets, I demonstrated that my Mamba-based models achieve comparable or even superior performance to transformer-based approaches TransformerG2G in link prediction tasks. Importantly, they offer significant computational advantages, especially when dealing with longer sequences. The DG-Mamba variants consistently outperformed transformer-based models on datasets with high temporal variability, showcasing their ability to capture complex temporal patterns effectively.

Furthermore, I provided insights into the learned temporal dependencies by analyzing attention weights and state matrices, further highlighting the models’ capabilities. By effectively combining state-space models with graph neural networks, my work addresses key limitations of previous approaches and contributes to the advancement of efficient temporal graph representation learning.

In conclusion, my research offers promising directions for scaling dynamic graph embedding to larger, more complex real-world networks. This has the potential to enable new applications in diverse areas such as social network analysis, financial modeling, and biological system dynamics. The Mamba-based models I introduced present a valuable tool for researchers and practitioners working with time-evolving networks, paving the way for more efficient and effective analysis of complex systems.

Xenia – A Smart Tour Planning And Recommendation Using Crowdsourced Data

The paper “Xenia - A Smart Tour Planning And Recommendation Using Crowdsourced Data” presents the design and implementation of a mobile application aimed at enhancing the tour planning experience for users. The app leverages technologies like AI, NLP, and recommender systems to provide personalized tour recommendations and streamline the planning process. The key features of the app include:

The paper concludes that the Xenia app successfully achieves its goal of providing users with a comprehensive and interactive tour planning experience. By incorporating personalized recommendations, a virtual tour guide, shopping suggestions, and crowdsourced data, the app empowers users to make informed decisions and optimize their travel itineraries. Additionally, the integration of local businesses contributes to the growth of the local economy and enhances the overall tourism ecosystem. The app’s user-friendly interface and seamless functionality further contribute to its effectiveness in simplifying the tour planning process and enriching the travel experience for users.

Leadership and Recognitions

Certification

Contact