AutoML Selected Topics (#4)

After some time, we are happy to continue with our next AutoML meetup.
The meetup will include 3 sessions with speakers from Explorium and Whatify.
Hope you'll join us!

• link to zoom meeting will be available few days before the meetup.

Session 1: Erez Sali - CTO at Whatify

Title: On AutoML, business questions and slot machines

AutoML has become an essential element in the way machine learning is used. In the last International conference on machine learning (ICML), the paper that laid the foundation to the field was selected as the most influential paper on ML in the decade that has passed.
In this talk we’ll review the approaches underlying AutoML, their link to slot machines and describe how it is used to generate predictions and recommended actions

Session 2: Shai Yanovski - VP Data at Explorium

Title: Automatic Data Discovery and Augmentation

While many recent works in the field of AutoML are focused on various parts of machine learning pipelines such as model selection, hyperparameter tuning, and feature selection, most of those works start with a defined dataset as an input; leaving the initial step of the dataset creation task to the human operator.
In this talk we’ll present an overview to the relatively unexplored area of automatic data discovery and augmentation. Discussing the existing approaches, challenges and opportunities in this emerging area of AutoML landscape.

Session 3: Noam Maloletnik - Head of Applied Research at Whatify

Title: Meta Learning for faster search and optimization

Many different search strategies can be used to explore the space of neural architectures, including random search, Bayesian optimization, evolutionary methods, reinforcement learning (RL), and gradient-based methods. All of these approaches require large computational resources as they train a large number of networks. In addition they all suffer from the cold start problem.
In this session we will review the use of meta-features and meta-learning to greatly accelerate the search.