Yujom® Success Stories
Young Indian Researcher Presents Independent AI Research at NeurIPS 2025
A student research achievement in machine learning and computational physics, presented at an international research workshop in San Diego, USA.
Aaditya Kachhadiya with his research poster at NeurIPS 2025.
About the Achievement
Aaditya Lalitbhai Kachhadiya presented his independent research at the Machine Learning for the Physical Sciences (ML4PS) Workshop, held during NeurIPS 2025 at the San Diego Convention Center, USA. His research was presented as a poster on 6 December 2025.
The work explores a learning-based approach to inverse problems in physical systems—problems in which unknown physical parameters must be estimated from observed data.
The Research
Deceptron: Learned Local Inverses for Fast and Stable Physics Inversion
“Deceptron: Learned Local Inverses for Fast and Stable Physics Inversion”
The Challenge
Inverse problems involve recovering unknown parameters from measurements or observations. These problems can be difficult to solve because of instability and sensitivity to optimization choices.
The Proposed Approach
The paper introduces a lightweight learning-based module called Deceptron, which learns a local inverse of a differentiable forward model. This is used to precondition gradient-based optimization.
The proposed method, called Deceptron Inverse-Preconditioned Gradient (D-IPG), incorporates consistency constraints and Jacobian-based regularization. The paper reports faster convergence on benchmark inverse problems compared with standard projected gradient methods, while remaining competitive with classical second-order approaches such as Gauss–Newton.
Explore the Research
Visit the official conference and workshop pages for further information about the research presentation.
View NeurIPS Research Page Visit ML4PS WorkshopEncouraging Curiosity and Research
This achievement highlights the participation of a high-school student in advanced scientific research and the value of curiosity, independent learning and sustained effort.
Disclosure: Aaditya Kachhadiya is the son of Yujom's founder. This page shares his research achievement for informational purposes; it does not imply that the research was conducted, funded, or validated by Yujom® unless separately documented.