Release notes AutoML v.0.3.2: Neural architecture search, new data science modules and UI improvements

Claudio Bruderer

Written by:
Claudio Bruderer (Product Manager at Modulos)

We’re very happy to ship the latest version of AutoML: v.0.3.2. With this version, we’re making the user interface more intuitive and we’re adding some exciting new data science features including a new AI approach we’re very excited about: neural architecture search!

Data Science Updates

New Models: Neural Architecture Search 

Illustration of neural architecture search with network morphism.
Illustration of neural architecture search with network morphism.

Neural architecture search (NAS) is an approach to efficiently search for better neural network architectures. There are different flavors of NAS and the first one we have implemented is NAS with network morphism. Network morphism starts by training a neural network. Then, it modifies the trained neural network in the next iteration by, for example, inserting additional layers or adding skip connections. Finally, the new architecture is re-trained and the process starts again. This is significantly more efficient than simply re-training the entire model.

We’re including NAS with network morphism for tabular data in this release. Since NAS is a computationally intensive process, we currently offer NAS as an “experimental” feature and it is de-selected by default. If you want to use it, you can manually enable it during workflow creation.

New Objectives

We added three new objectives that make it easier to train an AI solution for your problem. The new objectives are:

  • Mean Absolute Error (MAE): This objective is a standard error to be minimized, when quantifying the performance of models in a regression setting.
  • Precision and Recall: these two standard measures help you optimize models for different error tolerances. Precision is useful when a false positive is costly. Recall is useful when you care about finding as many positive results as possible and false positives are less of an issue for you.

User Interface Improvements

Redesign of the Datasets and Workflows Pages

The new workflow overview page.
The new workflow overview page.

We have redesigned the page displaying all the datasets uploaded onto AutoML making the overall look and feel lighter and more intuitive. The different status of the dataset to be uploaded and validated are now displayed in a dynamic way. We also applied similar changes to the page displaying all the created workflows, using which ML models are trained.

Redesign of the Workflow Status Page

The new workflow detail page.
The new workflow detail page.

The page showing the status of each workflow has been restructured and redesigned as well, giving it a more elegant touch. The metadata has been reorganized to allow for a quicker overview. By adding additional tabs, the navigation has furthermore been made more intuitive.

Other Improvements and Bug Fixes

We additionally made a range of other fixes and improvements to AutoML, which include:

  • Refined the navigation and scrollbar in the tutorials section on the platform.
  • Standardized the primary and secondary buttons. 
  • Improved the design and placement of the error messages.
  • Polished the UX of the User Management section.
  • Further streamlined the installation procedure.