Francois Chollet: What’s Next for ML? Trends from Powering Google's Production ML

The presentation discusses the need to make machine learning (ML) universally accessible to developers who are not ML experts. The speaker emphasizes the importance of UX design, componentry use, and automation in empowering developers to use ML effectively.

• ML is moving at an incredible speed and has the potential to transform various industries.
• To realize the full potential of ML, it needs to be accessible to anyone with an idea and some coding skills.
• The Responsible AI Toolkit is a suite of tools and resources that can help developers address safety concerns at each stage of the model development process.
• Progressive disclosure of complexity is a key design principle in making ML accessible to developers.
• Keras API provides a range of workflows from the very high level to the very low level, corresponding to different user profiles.
• Avoiding API silos is important in making ML accessible to developers.
• The speaker encourages developers to check out the Model Remediation Package to address bias in Keras models.

The speaker gives an example of how ML can be used to optimize fish farming in Norway or monitor the Amazon rainforest for illegal logging. However, to realize the full potential of ML, it needs to be accessible to developers who are not ML experts.

Francois Chollet is a Software Engineer and AI Researcher at Google, created the Keras deep-learning library, and is a primary contributor to the TensorFlow machine learning framework. TensorFlow is the most popular machine learning platform, adopted by over 3 million developers globally. It is the third most downloaded repository on GitHub, runs on 4 billion devices, and powers machine learning at Google, Apple, Netflix, Twitter, and others. Keras is behind YouTube's recommendations and Waymo's self-driving cars. Join Chollet as he shares what he considers to be the most important machine learning trends and their broader implications. Among the trends he will discuss is machine learning’s transition from an expert craft to a democratized utility, creating an ecosystem of reusable parts with premade and pre-trained models where all you need to do is bring your data. Another trend he will cover is the increasing scale of large models built on more data; this allows for use cases that are quickly changing our world. Simultaneously, the growing number of edge devices that can run ML models, including web browsers, is lowering costs, increasing privacy, and dramatically expanding the set of machine learning applications. Chollet will also talk about how, given all these other trends, it’s more important than ever to properly navigate privacy, bias, and safety issues to ensure beneficial applications.