MIT SIAM Seminar
Wednesday, November 28, 2018 at 4:30pm
Ali Ramadhan, Graduate student, EAPS (Climate Science)
Title: Reducing the error bars on climate predictions
Abstract: Starting with the first computational weather forecasts, a ridiculously crazy idea a hundred years ago, we'll see how modern climate models work and why uncertainties in climate predictions are so high despite their sophistication. Then I'll talk about how we're trying to reduce uncertainties in climate predictions by developing a new climate model in Julia that runs on massively parallel GPU accelerators and learns from observations and high-resolution simulations.
Sam Raymond, Graduate student, Civil and Environmental Engineering
Title: Teaching a Neural Network Physics to help us design complex devices.
Abstract: Machine learning structures like neural networks are especially good at finding complex transfer functions to connect various inputs and outputs. This can be extended to creating connections that we as scientists struggle to envision. With enough good data, a network can learn complex physics and be used to invert equations to infer the initial or boundary conditions for a system. This work presents an application of this methodology that leverages numerically simulated, synthetic, data to enable the design of custom microfluidic devices.