CV
Current academic CV.
Contact Information
| Name | Aaditya L. Kachhadiya |
| Professional Title | Independent Researcher |
| kachhadiyaaaditya@gmail.com |
Professional Summary
High school student and independent researcher working at the intersection of machine learning, physics, and numerical optimization, with interests in scientific machine learning, and principled approaches for scientific discovery.
Experience
-
2026 - present Surat, Gujarat, India
Research Collaborator
Sardar Vallabhbhai National Institute of Technology
Working under the guidance of Dr. Tanmoy Hazra on optimization methods for physics-informed learning and scientific machine learning, in collaboration with researchers at SVNIT Surat.
- First and corresponding author on a manuscript submitted to IEEE Transactions on Neural Networks and Learning Systems.
- Co-authors include Meet Dabgar, Dr. Tanmoy Hazra, and Prof. Anupam Shukla.
- Research focuses on optimization methods and diagnostics for physics-informed scientific learning objectives.
-
2024 - present Surat, Gujarat, India
Independent Researcher
Self-employed
Independent research in scientific machine learning, nonlinear inverse problems, numerical optimization, and learned solvers for PDE-governed systems.
- Solo author of Deceptron Learned Local Inverses for Fast and Stable Physics Inversion, presented at the NeurIPS Machine Learning for Physical Sciences Workshop, 2025.
- Solo Author of Local Inverse Geometry Can Be Amortized, arXiv:2605.13068.
- Developed an open-source PyTorch implementation of Deceptron and D-IPG for nonlinear inverse problems.
Education
-
2026 - 2027 Surat, Gujarat, India
Senior Secondary, Class 12
Shardayatan School
Science
- Gujarat State Board, ongoing.
- Expected completion in March 2027.
-
2024 - 2025 Surat, Gujarat, India
Secondary, Class 10
Shardayatan School
Science
- Gujarat State Board.
Publications
-
2026 Component-Access Subspace Optimization for Heterogeneous Scalarized Learning
Manuscript under review, IEEE Transactions on Neural Networks and Learning Systems
First-author manuscript on subspace optimization methods and diagnostics for physics-informed scientific learning objectives.
-
2026 Local Inverse Geometry Can Be Amortized
arXiv preprint arXiv:2605.13068
Introduces Deceptron Inverse-Preconditioned Gradient, a learned local-inverse optimization method for nonlinear inverse problems, together with a Jacobian Composition Penalty and its runtime inverse-consistency diagnostics.
-
2025 Deceptron Learned Local Inverses for Fast and Stable Physics Inversion
NeurIPS Machine Learning for Physical Sciences Workshop
Workshop paper presenting Deceptron, a learned local-inverse approach for fast and stable physics inversion.
Projects
-
Deceptron / D-IPG
Open-source PyTorch implementation of Deceptron Inverse-Preconditioned Gradient for nonlinear inverse problems.
- Implements learned forward-reverse modules, D-IPG inference, Jacobian-consistency training, runtime diagnostics, and PDE inverse-problem benchmarks.
- Repository includes reproducibility scripts for scientific inverse-problem experiments.
- Code available at github.com/AadityaKachhadiya/deceptron.
-
New York Academy of Sciences Junior Academy
Participant in the New York Academy of Sciences Junior Academy 2025.
- Selected & Participated in a global STEM innovation and research program for high school students. Developed a Hybrid Reinforcement Learning and Forecast-Based Solar Battery Scheduling approach on a real power plant dataset.
Skills
Languages
References
- Dr. Tanmoy Hazra
Assistant Professor and Head, Department of Artificial Intelligence, Sardar Vallabhbhai National Institute of Technology, Surat. Email: tanmoyhazra@aid.svnit.ac.in