We believe that the mass adoption of machine learning will be unlocked by a wide array of technologies: from those that help data scientists to be more effective to those that make the deployment and monitoring of machine learning systems more robust.
It is the point of view of this show however, and the community around it, that one key to this unlocking will be the full realization of the dream of AutoML: the complete automation of the entirety of the machine learning life cycle – from the possible framings of a problem, to the model building and model construction, model hosting, model monitoring, and much more.
It is this expanded view that we analyze on the show – through conversations with the world’s leading researchers and practitioners.
One theme of the show is that of meta learning, or Learning To Learn. Meta learning is a subfield of machine learning that emphasizes the need to learn general principles of learning, so as to design and engineer better systems capable of learning on their own.
In this show, we advocate for a wide definition of metalearning, one that includes any possible experience that can influence the design and implementation of automatically learning systems, hence, the tight connection between metalearning and AutoML.
Traditional metalearning uses metadata, like properties of the learning problem, algorithm properties (like performance measures), or patterns previously derived from the data, to learn, select, alter or combine different learning algorithms to effectively solve a given learning problem. This more traditional metadata learning is nested within our wide definition of metalearning, but other, even analytical approaches for learning how to learn, are to be included as well.
We take the stance that, to best advance the prospect of systems autonomously learning, we have to engage with a wide segment of researchers that explore all forms of learning to learn. This show is our effort to bring those conversations closer to you.