Deep learning R&D engineer Kyle McAdoo talks about a typical day for him, the coolest parts of his job and the skills needed to work in edge computing.
One of the best parts of working in the tech sector may be getting to focus on emerging technologies such as edge computing and deep learning.
That’s what Kyle McAdoo does as a deep learning R&D engineer in Intel’s versatile processing unit (VPU) team.
After graduating from Queen’s University Belfast, he worked with start-up Amphion Semiconductor for 10 years. He then joined Irish chip company Movidius in March 2015, before it was acquired by Intel in 2016.
In an interview with SiliconRepublic.com, McAdoo said the VPU is focused on bringing deep learning inference acceleration to the edge, right onto your laptop.
“The key goals are performance and power. No conversation could be had about VPU without including those two words.”
‘Keeping motivation high in such a fast-paced arena can at times be difficult’
– KYLE MCADOO
What does a typical day look like for you?
My particular team is focused on performance – looking at existing VPU products that we may have in our hands but primarily at future products at various stages of development. The skillset within the team is huge and often extends way beyond my particular comfort zone!
It needs to be, as so many aspects of architecture and software plus hardware execution need to come together to even understand performance targets at the VPU level, let alone plan ways to deliver it.
Often, we will be looking at network topologies with a view to understanding how they would best map to the VPU. We compare that to what is happening right now and try to understand any differences.
The information we learn here could go to software teams for current generations and to architecture and hardware teams for next-generation products as appropriate. I would hope that the working culture we have grown means to the outsider the lines between those groups should look very blurred.
What are some of the coolest elements of your job?
While many things in this field are unpredictable, a growing requirement for AI acceleration in many aspects of life is certainly not. VPU can have a voice in this development, which will be magnified by the ubiquitous Intel presence in computing.
The thought that many people will be leveraging your work and be totally oblivious to it in use cases ranging in complexity from trivial to barely fathomable – I think that’s a huge motivation.
We know the pressure to innovate and improve from within the team and externally will be enormous. This represents the challenge but also the excitement of VPU and its pathway over the next few years.
What are some of the skills you use on a daily basis?
Thankfully I still have the opportunity to get involved on the front line, so to speak, now and then. I enjoy the architectural work around our solutions and engaging with other teams to make sure we have alignment and can learn from and enable each other.
At the other end of the spectrum, I have always loved to debug. Debugging functionally and from a performance viewpoint really engages me and has a requirement to maintain a high-level view of all areas while deep-diving into some.
What personality traits make you particularly suited to your role?
The trait which has served me best is the confidence to say ‘I don’t know’. Realising there are often people around me better placed to provide an answer or even to help understand what the question should be!
What are some of the biggest challenges you face?
As is the case for many technical people, the transition from individual contributor, entirely focused on your own tasks, to taking on some more managerial activities is challenging.
Accepting that you can’t understand everything about everything but hopefully helping to enable others to achieve the solutions is a learning process for me.
More generally though, keeping motivation and personal drive high in such a fast-paced arena can at times be difficult. That Kipling line about triumph and disaster and “treating those two imposters just the same” comes to mind.
Sometimes the hardest thing to do is to experience a success, then decide to revisit it and tear it apart to decide how it can be improved. Likewise, we also need to accept that in the performance arena not everything you try will work.
Sometimes even if it does, a step forward elsewhere can render some really good work redundant. Accepting this and coming back the next day to start again, enthusiasm intact, is a learned skill.
What advice would you give to someone curious about working on the VPU side of things?
As the need for AI acceleration has increased, dedicated accelerators specialised for the role are essential to ensure power can be kept low while maintaining sufficient performance to service existing use cases.
Of course, new use cases get dreamed up by the day – imagine what will happen when this functionality is available to everyone with a laptop! The space for exploration seems totally unlimited right now. VPU is and must remain a really fertile environment for new ideas.
What kind of experience and educational backgrounds are relevant for your team?
People really interested in problem solving, first and foremost. Those with engineering and computer science backgrounds will have more of the toolset required to do the job immediately.
However, these tools such as programming languages and so on can of course be learned on the job. Understanding how to break problems down, solve them as part of a team and remain driven to improve seem to me to be the common traits of the best people I have the fortune to work with.
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.
