Name: Krishna Narayanan
Pronouns: he/him/his
Biography:
I work at Argonne National Laboratory as a Computer Scientist. I am originally from India and moved to the US for graduate school. I mentored an undergraduate student through the SRP program in Summer 2023. The internship solidified the student’s resolve to pursue research post-graduation.
Institution/Lab: Argonne National Laboratory
Website: www.mcs.anl.gov/~snarayan
SRP Collaboration Topic/Title: AI-Based Adjoints to Diagnose the Sensitivity of Ocean Models to Parameters
Field or research area: AI, Machine Learning, Earth Sciences
Please select all the topical areas that apply to your project:
Computational Science Applications (i.e., bioscience, cosmology, chemistry, environmental science, nanotechnology, climate, etc.); High-Performance Computing; Machine Learning and AI
Brief Abstract:
We are interested in understanding the sensitivities of an ocean model’s output, to the model parameters. Estimating this sensitivity is very time consuming by brute force. Alternatively, adjoints have shown great promise in uncovering sensitivity of the model to its parameters. Yet adjoints are very time consuming to develop manually and very involved to develop via automatic differentiation (AD) for some models. Because neural networks implemented in machine learning frameworks can be differentiated trivially, we have want to explore how to generate an accurate neural network (NN) surrogate for an ocean model. We then use our NN model to generate adjoint versions of the original model.
Desired relevant skills, background, or interests:
Experience in ML frameworks, models.
Other comments:
The project can be tailored to individual interests and expertise
Do any special requirements apply? Permanent Resident OK; International OK
Other, specify:
Keywords:
machine learning; AI; autodiff; climate, ocean; environment; automatic differentiation; Julia; Python
Lightning Talk Title: AI-Based Adjoints