February 2020 Summaries

Distributed applications have been around for over 50 years but are becoming more prevalent due to the end of Moore's Law and the increasing computational demands of machine learning applications. The growth in these demands far exceeds the capabilities of even specialized processors, leading system architects to distribute computations across multiple nodes. Google and Nvidia have released distributed AI systems, while companies like OpenAI are using thousands of CPU cores for reinforcement learning simulations. To bridge this gap between application demands and hardware capabilities, new software tools, frameworks, and curricula are needed to empower developers to build such applications in the era of big data and AI.

The January 30, 2020 Ray Meetup in San Francisco featured several talks and presentations from industry experts. The event included a welcome introduction by Dean Wampler, followed by talks on various topics such as fast scheduling in Ray 0.8, training sparse neural networks with Ray, and using Ray for machine teaching services. Additionally, a demo of the Ray Dashboard was showcased, although it was not recorded due to technical issues. A new scalable machine learning model serving library called `Ray.Serve` was also introduced by Simon Mo from Anyscale. The event concluded with an invitation to learn more about Ray and sign up for the free Ray Summit this Fall.