ECCV2020 tutorial - Differentiable Optimization Layers - Hands-on coding

About this lesson

Speaker: Akshay Agrawal, Stanford Tutorial website: http://eccv2020.deepdeclarativenetwor... CVXPY: https://www.cvxpy.org/ CVXPYLayers: https://github.com/cvxgrp/cvxpylayers Notebook: https://github.com/cvxgrp/cvxpylayers/blob/master/examples/torch/signal_denoising.ipynb TL;DR: We show how to train convex optimization layers for structured prediction, with an example in signal denoising. Abstract In this notebook, we show how convex optimization layers can be used for tasks in structured prediction. Specifically, we show how to train a parametrized convex optimization model that depends on the input for predicting the output; the prediction of a convex optimization model is the solution to a convex optimization problem. Convex optimization models provide a powerful and expressive way to add inductive bias to learning pipelines, via the objective function and constraints. As an example, we train a convex optimization model for a signal denoising task.

Original Description