Data Science Deep Learning Machine-LearningSee in schedule: Thu, Jul 29, 10:00-10:30 CEST (30 min) Download/View Slides
During my first steps in the field I was promised that machine learning would be automated from the beginning. Unfortunately, once I’ve outsourced looking for the parameters that best matched my data to the machines, I was instead left with having to look for the hyperparameters that define the best model architecture, all by myself. This often ends up being a lengthy manual process. Is there a way to outsource this bit too?
In this talk we will take a look at the automated machine learning libraries Keras Tuner and AutoKeras, which allow the user to create high level templates of deep learning models and use them in automated search for the best hyperparameters. They not only enable speedier development of better models but also make deep learning accessible to a wider pool of people thanks to the abstractions they offer.
In the presentation we will go through several iterations of pretending to know progressively less and less about both our data and machine learning in general, and see how these libraries come to our help in creating highly performant deep learning models with a fraction of the effort. It is aimed at a general audience familiar with Python. Knowledge of Keras is a plus but not a requirement - that is kind of the whole point!
The repository with a notebook with code examples is available at: https://github.com/ondrejiayc/ep2021_automl_keras
Type: Talk (30 mins); Python level: Beginner; Domain level: Beginner
I am currently a data scientist at Trayport, trying to help bring artificial intelligence to energy trading. Before starting to call myself a data scientist a few years ago I was an astronomer. I'd be happy to talk to you about the AI, cool tricks in python, galaxy clusters or interesting books either of us might have read!