AI Best Practice and Open Source Case Studies - Open Source SG

Speakers

Dr William Jones
Dr Jeremy Bennett

Synopsis

Best practice for AI in the Electronics Industry: 5 questions you need to ask.

TechWorks, the trade body for the UK Electronics Systems industry, is developing best practice guidelines for engineers developing AI/ML systems, with the first version announced at Bletchley Park earlier this year.

This standard is being developed openly and collaboratively, and in this talk we will present the key ideas behind the standard.

At the end of this talk, the attendees will have a clear understanding of the 5 major questions to be addressed before progressing with an AI project in the tech sector.

Bringing up an AI platform on novel hardware: An open source case study.

The major AI platforms, such as PyTorch, TensorFlow and Triton are all designed to work on standard hardware comprising a general purpose compute engine and a specialist AI accelerator. For example an Intel x86_64 platform combined with an Nvidia GPGPU.

Numerous companies are now designing chips with specialist AI accelerators built in, particularly for embedded use with edge inference. But how do you get AI platforms to work efficiently on this new chip. For example PyTorch has around 4,000 standard operations, all of which need to be built efficiently to split their load been general processor and accelerator.

In this talk we will present a project run jointly with Southampton University to bring up PyTorch on an embedded RISC-V core. The project used the open source OneAPI Construction Kit created by the Unified Acceleration Foundation.

We will show the approach used, and see what demands it makes on PyTorch, and how it also demonstrated portability by swapping in an alternative core as accelerator.

At the end of this talk, attendees will have a clearer understanding of the complexity of the AI infrastructure we increasingly rely on. They will also understand how to approach bringing up an AI system in a way that is efficient and flexible.

About the speakers

Dr William Jones is head of AI/ML at Embecosm and previous co-founded the crypto-currency company ARWeave. He leads the TechWorks initiative developing best practice guidelines for AI/ML.

Dr Jeremy Bennett is Chief Executive of Embecosm and a specialist in tooling for embedded systems. He is a past Chair of the BCS Open Source Specialist Group.

Our events are for adults aged 16 years and over.

BCS is a membership organisation. If you enjoy this event, please consider joining BCS. You’ll be very welcome. You’ll receive access to many exclusive career development tools, an introduction to a thriving professional community and also help us Make IT Good For Society. Join BCS today: https://www.bcs.org/membership/events

Please familiarise yourself with the Visitor Instructions for the BCS London Office.

Please note, if you have any accessibility needs, please let us know via groups@bcs.uk and we’ll work with you to make suitable arrangements.

For overseas delegates who wish to attend the event, please note that BCS does not issue invitation letters.

This event is brought to you by: Open Source SG