Abin Shakya
ML Researcher | Data Scientist | Algorithmic Trader | Mathematics Enthusiast
Research and Personal Projects
Graph Neural Network Accelerated Molecular Dynamics
Developed a Graph Neural Network model to accelerate Molecular Dynamics simulations for studying atom distribution in Earth's core.
- Achieved significant speedup in simulation time by building a deep potential accelerated simulation
- Implemented coordination module to understand the atomic correlations and local structures.
Active Learning Framework: Iterative Sampling and Training in Interaction Modeling
Designed an active learning framework for molecular dynamics, utilizing iterative data sampling and model training cycles to continuously enhance model accuracy.
- By strategically sampling data from diverse regions of the feature space, we achieved comprehensive coverage, ensuring a well-balanced and representative dataset for model training.
- As a result, we were able to accurately model Nitrogen interactions, which had not been possible with previous methods.
MelodyGen : A transformed based music genertion model
Developed a transformer-based generative AI model that composes original piano music, crafting expressive and dynamic melodies for an immersive listening experience.
- MelodyGen utilizes a Transformer-based architecture to analyze piano patterns and structures, enabling it to compose original piano pieces that sound natural and engaging.
- By leveraging the capabilities of the Transformer model, MelodyGen generates melodies that evolve over time, offering a variety of styles and emotional expressions tailored to different listening preferences.
Listen to a Sample Composition:
Prolog Interpreter In Functional Language SML
Developed a simple Prolog interpreter using the functional programming language SML
- We used an online source for the Abstract Syntax, Parser, and Lexer. Using these components as a foundation, we developed a Prolog Interpreter in the well-known functional programming language, SML.
- The interpreter facilitates user interaction, allowing users to input Prolog facts, rules (Horn clauses), or queries and view the corresponding output.