FPTalks on June 4: Jeffrey Sarnoff (IEEE)

[Pavel Panchekha]

Hello all!

The next FPTalks Seminar is Thursday, June 4 from 9–10am Pacific Time on Zoom:

https://washington.zoom.us/j/99708186928?pwd=HbhpebAtCWvoP4VQahYb8G1QpQnTgm.1

I'm super excited for Jeffrey to talk about the fast floats for AI in the IEEE P3109 family. Here's the abstract:

This talk introduces the work of IEEE SA P3109, a group drafting a proposed new standard: Arithmetic Formats for Machine Learning. The presentation provides familiarity and conversancy with the group's purpose, direction, and approach. P3109 defines a versatile arithmetic framework and structured family of small floating-point formats designed specifically for machine learning. The standard covers formats from 3-bit to 16-bit, including the 4-bit and 8-bit formats increasingly used in ML training and inference hardware. This framework incorporates a consistent approach to the design of operations, which, together with rigorous formal verification, support computational reliability. We offer a dedicated arithmetic framework for machine learning that provides the required consistency and flexibility for training and inference in deep learning.

Jeffrey is the Secretary of P3109, and points folks to the Github repository for more.

Also, we are getting reading for our annual workshop. Please register for the event or even submit a talk. We're excited to see you all there.

—Pavel Panchekha

[Pavel Panchekha]

Reminder: Jeffrey Sarnoff is speaking about Fast floats for AI: the IEEE P3109 family at the FPTalks seminar, this Thursday, June 4, from 9–10am Pacific Time, on Zoom:

https://washington.zoom.us/j/99708186928?pwd=HbhpebAtCWvoP4VQahYb8G1QpQnTgm.1

—Pavel Panchekha