Meet one of the key players helping accelerate neural network models.
The Artificial Intelligence Products Group (AIPG) is responsible for the creation of many innovations, including the hardware accelerators critical to providing the training and inference that will help solve the main challenge to advanced deep learning: slow training for neural network models.
One person involved in accelerating neural network models? Michal Lukaszewski. He joined Intel over four years ago as a software developer and is now a deep learning engineering manager in Gdansk. As part of AIPG, he works on the team that focuses on benchmarking and measurement; providing a wide range of reliable data and recommendations to customers to enable the AI revolution of the future.
We recently spoke with Michal about his work, its impact, and what AIPG is looking for in future team members.
What does your group do?
In short, we create hardware and software accelerators that enable our customers to more easily and affordably work with deep learning or machine learning technologies and solutions.
We have several specialized teams based here in Poland. Our software architecture team supports the hardware team based in the United States to design new hardware dedicated to accelerating deep learning solutions. We also have teams that produce the kernels, the drivers for the hardware produced in the other sites. We create software we call the engine; the dedicated low-level library that optimizes mathematical operations for the hardware architecture we use to run neural networks. Though Intel doesn’t create its own framework for deep learning, we are participating in external projects to accelerate these as well.
I like to compare our job to vehicle mechanics. You can drive a car all your life and have no idea how it works. When the car breaks down, you go to the mechanic because he knows how the car really works. We are the designers; we are creating, we are looking under the hood, we are fixing things, optimizing things.
Not only do we upgrade existing solutions, we create new solutions and approaches and implement them into the hardware and software. We believe that these new ideas really change deep learning and machine learning. This is not only evolution, this really is revolution.
What is the purpose and impact of this technology?
A neural network model is a series of several highly complex, sophisticated mathematical operations to process any given data. The mathematical operations within these models are dedicated to finding the context of data and understanding what it means. We have hundreds of models dedicated to different operations, including face image recognition, image classification, and object detection.
We also have models dedicated to processing sound for better natural language understanding and written text comprehension. Some of these trained models can be used to create new things—new music, new paintings—or to process incredibly fast deep learning simulations to help in research that would be cost prohibitive using classic simulation methods.
One of the coolest projects we are working on here in Poland is a cloud-based platform that gives non-technical people access to an easy-to-use environment in which to run deep learning and machine learning tools at accessible prices.
What makes your group successful?
We are small, agile, and really focused on the customer. We work closely together and are constantly pushing to find opportunities, not just react to market needs. The main difference I find here in AIPG is the level of communication within the group. For instance, on a recent project, we communicated directly with the vice president about the technical details we were proposing. This spirit of cooperation can be seen at all levels. Each sub-group specializes in different areas of software development, and we talk to each other to keep sight of our wider goals as a group, and to focus on our customers’ needs.
What would the ideal candidate joining your team bring to the table?
AI is a vast world and some people are afraid they don’t have the right experience to fit in. But Intel Poland is a very welcoming place. When we look for new talent, we focus on the programming skills and we’re willing to teach skilled programmers the AI perspective. So, if somebody is a Python engineer or C++ engineer or C engineer, they’re warmly welcome in AIPG. Mathematical knowledge, while not required at the start, can really help the daily work.
The most important quality we look for, though, is the desire to learn. People who are focused on learning and who constantly want to learn new things, to go deeper to understand things.
Curious about Intel Poland? Find out more here: https://www.intel.com/content/www/us/en/jobs/locations/poland.html