Developing TinyML University Programs

ICTP runs workshop on scaling university research and teaching on embedded machine learning
Developing TinyML University Programs
Charlotte Phillips

The recent ICTP Workshop on Widening Access to TinyML Network by Establishing Best Practices in Education aimed to develop sustainable university programs that both teach and research at the intersection of embedded systems and machine learning, or TinyML. The program combined a hands-on mixture of experience sessions and group work with keynote lectures from leaders in the field, and talks from researchers who are currently using TinyML in a diverse range of applications.

The workshop grew out of connections between a network of TinyML researchers. “The whole goal of the workshop was to share experiences of teaching TinyML,” says Marco Zennaro, local organiser, research scientist, and coordinator of STI. “We created a working group, an ICTP network of universities in developing countries working with TinyML during the COVID-19 pandemic.”  Members of this network kept in touch weekly online. “I think we’ve met almost 100 times now on Zoom,” adds Zennaro. 

“This workshop is really to bring people together.” -  Evgeni Gousev, TinyML Foundation

“TinyML is a new, very fast, emerging field of machine learning that has quite a bit of potential for many different applications,” adds Evgeni Gousev of the TinyML Foundation, which was formed to connect TinyML researchers and private enterprise. “Many companies are driving this from the industrial perspective, and the educational part of this is very important.”

“This specific workshop is really to bring people together: people who have been working in this field for the past several years and can share their experience, and also people interested in implementing this in education and really brainstorming and sharing all this knowledge,” says Gousev. 

The majority of the workshop participants teach semester-long or shorter university-level TinyML courses, and came to share their experience of teaching in the field. Marcus Rub from Hahn-Schickard-Gesellschaft für angewandte Forschung e.V. and Thomas Amberg from the University of Applied Sciences and Arts Northwestern gave a talk on their experience of developing open training materials for the internet of things.

 “TinyML is a huge field.” - Alessandro Grande, Edge Impulse

The research talks showcased a broad range of TinyML uses, such as in health monitoring and animal tracking, and challenged the TinyML community to solve these critical issues in a sustainable and privacy preserving manner. Paul Kucera from UCAR/COMET in Boulder, USA, proposed the use of TinyML for weather stations. “His point was that we need more weather stations, and we need to have more weatherization in developing countries,” adds Zennaro. “Could we use TinyML for this? The hope is that someone in the community will pick up the challenge. Cyril Caminade from ICTP also proposed the possibility of detecting mosquitoes with this technology. People could come up with new ideas to solve these issues.”

Alessandro Grande from Edge Impulse also provided a keynote lecture. “I discussed how Edge Impulse can enable the next generation of engineers and students to learn about machine learning and apply these concepts to actually build the next generation of products,” says Grande. “TinyML is a huge field and two of the biggest application spaces are digital health, where people are developing new wearable devices that can detect and inform users on their health and activity, and on the other hand, actual industrial productivity devices that are helping to make industries and production lines more efficient and effective.”

David Cuartielles, co-founder of the open-source hardware and software company Arduino, discussed different examples of teaching at the workshop. “In the last 20 years, I've been working on designing different courses, from programming to electronics and everything in between, so I looked at the history of different ways of teaching in formal and informal settings,” says Cuartielles. “Right now, TinyML is used to improve sensors and make decisions on-site. That's what the technology allows you to do. When it comes to education, we're using it to teach students how to design cameras, accelerometers, or other complex sensors, where it's really hard to find a programmatic kind of solution. We use TinyML that allows you to capture data, make decisions based on that data, and simplify the whole workflow from the concept to final part.”

Eric Pan, founder and CEO of Seeed Studio, gave a keynote lecture on real-world sensing and data collection in relation to TinyML. “I hope participants now have more understanding and more confidence about adopting TinyML, and also to include it in their courses, to involve more young people and realize its potential,” says Pan. “We have 1 million customers and makers all over the world. The key thing we want to do is to enable them to really deliver solutions using their ingenuity, from innovations, to resolve real-time issues. I think TinyML will become more pervasive. It can be embedded into a lot of applications to become a new way of fetching data by compressing the raw data and making inferences. It could be cheaper, more powerful, and more people could use it.”

Marcelo Rovai, professor at the Federal University of Itajubá in Minas Gerais, Brazil, gave a hardware demonstration at the workshop to give participants an idea of TinyML-based devices. “Seeed Studio and Arduino manufactured these devices, and they donated one set to each of the participants. We used the opportunity to give a hands-on session,” adds Rovai. “I think that people left here very excited about the possibilities. They saw that we are providing real courses, what we are doing, and what types of projects the students, professors and researchers are doing at universities.”

“Participants at this meeting got a chance to get to know each other in person, because they've been collaborating throughout the years now, especially after the pandemic,” adds Cuartielles. “They get a chance to build a network, so they can continue to create a shared curriculum for TinyML education.”

“As this is a completely new topic, people have different issues in different environments. We wanted to share experiences and produce a white paper about the best way to teach TinyML.” - Marco Zennaro, ICTP

Brian Plancher from Barnard College, Columbia University, USA, acted as the facilitator for the white paper sessions. These were held at the end of the workshop, with the aim of producing a document that lays out a roadmap for developing integrated university programs that teach TinyML, providing real-world impact through academic research. “The white paper will be provided online by our network,” says Zennaro. “The TinyML Foundation is also interested in the output of the paper. All these people have been teaching TinyML, and they have their own slides. So, one of the outputs that I care about is how we can make these open.” 

Topics discussed in the white paper development sessions included the translation of teaching materials, and ways of making these as open as possible, to widen access to a broader community. “We’re expecting not just the production of this paper, but also a network,” adds Rovai. “This was one of the most important things we’ve done here, create a network. So people from Africa and Asia, but also invited people from universities here and in America, brought experiences from where they already run courses. They can help others to implement courses in their areas, mainly in developing countries. That is our goal.” 


Publishing Date