Deep learning, artificial intelligence, and neural networks are challenging new concepts for many, but an intensive short course from ICHEC aims to plug the knowledge gap.
The Irish Centre for High-End Computing (ICHEC) is getting ready to run a first-of-its-kind training course on deep learning. Starting on 17 September and running for five days, the course aims to give participants from industry, SMEs and large public sector bodies rapid hands-on training in deep learning methods.
Peter Woods, business development and public sector lead at ICHEC, believes in the center's vision to extend the reach of high-performance computing to all Irish companies, especially the SME sector. Here, he answers some questions on what to expect from the course and why short-term training solutions are needed to address emerging technology skills gaps.
What do you define as 'deep learning'?
There are a number of deep learning definitions out there, but I prefer to think about what makes deep learning work: artificial neural networks. Artificial neural networks mimic how the human brain works and come in a range of flavors, but they all operate in a similar manner: being fed data and slowly tuned through trial and error until the network can reliably recognize patterns such as a vehicle in an image.
At its core, deep learning is extending that approach to bigger, more complex networks trained in the same way. This allows us to model and recognize many more complex patterns, such as telling whether a vehicle is a sports car, a family car, a van or a truck, or whether it looks like it is stationary or moving. While the idea isn't new, the depth and complexity of the networks mean that training them requires vast amounts of data and tremendous computational power. The current combination of fast networking, smart devices and increased computational power (notably through graphics processing units, or GPUs) means that deep learning is now not only a possibility, but also relatively affordable.
What can someone learn about deep learning in five days?
This is a very intensive course where the participants will go through all deep learning concepts and relevant frameworks from a number of partners. The course will equip participants with the knowledge to begin developing some of their own deep learning projects once they leave.
Nevertheless, we are not under the illusion that once someone completes the course that they will be a deep learning expert. But they will definitely have the required baseline knowledge and experience to enable them to become more proficient with the right commitment.
The programming side of implementing deep learning is actually not that difficult itself, relatively speaking. The hard part is how you prepare the data and design the right approach and parameters for the specific problem at hand. This can only come with further practice and experiments. Our course enables participants to embark on that journey.
Will the course be welcoming to someone completely in the dark on machine learning, or is there a certain level of understanding recommended?
The prerequisite for participants is that they have a certain level of programming experience and know some basic maths like algebra. While it is indeed possible for someone who is completely in the dark about machine learning to benefit from the course, it does help if that person has an aptitude for numerical and statistical methods, along with a willingness to learn. So, while a mobile app developer might find the course overly challenging, a data analyst with programming experience who knows statistics (but not machine learning) would benefit greatly.
What kind of hands-on experience will they get?
The course is extremely hands-on. The last thing we wanted was an entirely theory-based course. From the very first session, students will be at their laptops doing programming tutorials. Throughout the five days, they will gain experience with a variety of environments (or frameworks) that allow you to conduct and deploy deep learning, such as PyTorch, Matlab, Caffe2, TensorFlow and Microsoft Cognitive Toolkit (formerly CNTK).
A common example (in this case, CIFAR-10 image classification) will be demonstrated over all the frameworks in order to highlight similarities and differences. Following on from that, there will be examples specific to each framework to highlight unique features and exemplar applications.
How can you see participants applying what they learn on this course in their work?
The goal of this course is that, after five days, each participant will have understood enough deep learning concepts and frameworks to allow them to begin their own projects. They may not get all the way through a project without further self-study, but we can assist them when necessary.
As a follow-up, ICHEC will be running a next-level course in deep learning next year, subject to demand. This course will run one afternoon per week over eight weeks and will be more project-driven. It is anticipated that participants on this Level 2 course will develop their own specific projects related to their organizations and, over the eight weeks, will gain the relevant knowledge to successfully see these through.
The University of Limerick is also launching a master's in artificial intelligence in collaboration with ICHEC this month. Do you think that educational institutes and curricula are struggling to keep up with these emerging and rapidly developing technologies?
ICHEC is delighted to be collaborating with the University of Limerick on the MSc in AI, which we have no doubt will be a big success. However, an MSc takes time, and the feedback we were receiving from our industry partners is that they need something to fill the short-term gap. That's why we're running these short deep learning courses. Once the students from the MSc start to graduate, then the need for these intensive courses will possibly dissipate.
Technology moves at such a rapid pace that it is almost impossible to continually change curricula to keep pace. Indeed, our short courses are very flexible to keep pace with change, but educational institutes are increasingly coming to terms with this and becoming more flexible with the curricula design and update process. The MSc in AI was established in a matter of months via the Skillnet programme, from conception in early 2018 to the first cohort of 100-plus students who started the preparatory modules in September 2018.
What I would advise is that we never sit on our laurels as we will get left behind very quickly. Businesses should already be anticipating what the next MSc or intensive course should be so that we can be prepared even before the need arises. May I suggest quantum computing?