tinyML: Pioneering sustainable solutions in resource-limited environments

Go back to programme

tinyML: Pioneering sustainable solutions in resource-limited environments

The tinyML workshop at the AI for Good Summit 2024 will delve into the groundbreaking field of tiny machine learning (tinyML). In today’s rapidly evolving technological landscape, tinyML stands out as a revolutionary approach with far-reaching implications across various sectors including education, health, environment and so on. This workshop is aimed to explore the transformative power of tiny machine learning, the use of artificial intelligence in addressing critical challenges, and the great potential in making positive contributions to the United Nations Sustainable Development Goals.  

The transformative power of TinyML lies in its ability to operate in resource-constrained environments by empowering on-device sensor data analytics with minimal energy consumption, opening up new avenues for innovation and progress in different sectors under extreme conditions. In addition, tinyML could be applied to challenges that are aligned with SDGs, such as detecting plant disease and monitoring wildlife, which aim for a zero hunger and bio-diversified future. 

Throughout the workshop, participants will have the opportunity to engage with leaders and experts from diverse fields such as education, healthcare, environment, and development. They will share use cases of tinyML in action, demonstrating its real-world applications and impact on society. 

Whether you are new to the field or an experienced professional, this workshop will provide valuable insights into the boundless opportunities and applications of tinyML, demonstrating, especially in emerging markets, its significant role in shaping a more sustainable and equitable future.  

tinyML for Good: Cases Studies in Education, Sustainability, and Healthcare 

Didem Un Ates 

This keynote at the UN AI for Good Summit explores the transformative potential of tiny machine learning (tinyML) across global challenges. The session will introduce the TinyML Foundation, showcasing how its members employ tinyML in education, sustainability, and healthcare. Through these case studies, attendees will see tinyML's role in enhancing learning, promoting environmental sustainability, and improving healthcare by enabling efficient, real-time data processing at the edge. Additionally, the presentation will touch on the importance of Responsible AI and the necessity for a proactive approach to talent development, particularly in developing countries. This discussion aims to highlight the need for ethical deployment and local capacity building in tinyML technologies, fostering a balanced development of AI that benefits society globally. 

 

Widening access to TinyML by establishing best practices in education 

Marco Zennaro



TinyML can significantly contribute to achieving the Sustainable Development Goals (SDGs) and advancing scientific research in fields like environmental monitoring and energy management. To enhance access, participation, and the impact of this emerging technology, we introduce an initiative aimed at establishing and supporting a global network of academic institutions in developing countries focused on TinyML. 

 

Small is Beautiful: Making space for tiny sustainable solutions’. 

Fran Baker 

Climate crisis, growing inequality, and rapid developments in technology. When we talk about ‘AI for Good’ we often imagine enormous scalable solutions that affect billions of people worldwide to address the biggest challenges facing humanity  – both in terms of the technologies themselves and the reach they can attain. While these solutions are necessary and important, they are not the only path to progress for the Sustainable Development Goals. In this talk, the focus will be on showcasing how small is beautiful, recognizing EdgeAI and tinyML as valuable contributors to global sustainability challenges. 

 

An introduction to TinyML, Software, Hardware and Applications. 

Marcus Rüb 

This presentation provides an introduction to Tiny Machine Learning (TinyML), covering its software, hardware, and applications. TinyML combines machine learning with embedded systems, enabling ML models to run on low-power devices. The presentation begins with an overview of TinyML's significance and potential impact. It then explores software frameworks and tools for optimizing ML models on microcontrollers, followed by an examination of the hardware components, including microcontrollers and sensors, that support efficient ML execution at the edge. Real-world applications in sectors such as healthcare, agriculture, and consumer electronics illustrate TinyML's transformative capabilities. The presentation concludes with a discussion of challenges and future directions, emphasizing the ongoing innovation needed in this field. This overview equips the audience with a foundational understanding of TinyML and its potential to revolutionize embedded intelligence. 

 

What can be done with tinyML? 

Martin Croome 

GreenWaves Technologies is a European fabless semiconductor startup designing highly efficient and easy to program ultra-low-power application processors for tinyML applications that interpret and transform rich data sources such as images, sounds and radar signals using AI and signal processing. In this talk, Martin Croome, VP Marketing will present a quick tour of the elements that distinguish tinyML from other AI applications, where tinyML is getting used and some of the challenges and opportunities in developing tinyML applications. 

 

MedTech@CSEM: R&D Strategies, Innovative Algorithms, and Medical Certifications 

Mathieu Lemay 

CSEM S.A., the Swiss Centre of Microelectronics and Microtechnology, is a public-private non-for-profit Swiss RR&D Centre and acts as a technology provider for the industrial sector. CSEM develops innovative technology platforms and offers services from contract R&D to development of solutions for publications including precision manufacturing, energy, security, life sciences, and digital health. CSEM has been active in the (medical) wearable technology domain for 20+ years and its position as a leading innovation center in digital health has been federated by its heritage in the Swiss watchmaking industry. CSEM Digital Health research activity is certified ISO 13485 and operates on the Insel-Campus in Bern. It has long-term and widespread competencies in the development of low-power wireless sensors for human-monitoring applications comprising sensor electronics, signal processing and advanced feature extraction at the sensor level or in the cloud, data analytics and machine learning, as well as low-power chip design and wireless communication chipsets. In this presentation Dr. Mathieu Lemay, Signal Processing and AI Group Leader, will present CSEM’s current technologies with a particular angle on how R&D strategies and medical certification are driven with respect to the development and validation of algorithm solutions. 

 

Harnessing TinyML for Sustainable and Human-Centered Agriculture 

Sebastian Bosse 

The presentation discusses how TinyML technology can transform farming by improving sustainability and focusing on human needs. It covers the application of TinyML in optimizing farming practices, managing resources efficiently, and enhancing livestock care, while addressing challenges like scalability and data security. This exploration shows TinyML's crucial role in advancing agricultural innovation. 

 

What TinyML has to offer the education sector and the future generation of employees 

Paul Sant 

In this presentation we will discuss the tole that TinyML and artificial intelligence has to offer those within the education sector, particularly future generations of students. We know that technology devices are becoming ubiquitous, and content is being ingested on a variety of devices, with mobile devices being very popular. We will touch upon how we can utilise TinyML is a range of educational settings, explore some of the opportunities and challenges it may provide, and look at how we can develop and educate future generations to develop sustainable solutions that can provide high quality solutions even when there are restrictions around performance metrics such as power and available memory/processor power. 

 

Accessible and Actionable TinyML for a Better Future 

Violet Su 

The presentation will shed some light on how TinyML can be applied widely and extensively. It will share some efforts and approaches to make AI especially TinyML more accessible, followed by innovative projects made by enthusiastic & creative community members. These projects, ranging from waste management to early detection and warning for flood, and tree disease recognition, target specific issues in the real world, hoping to inspire more people to embark on the journey and dive deeper into the wonder world of adopting AI for Good! 

 

Development and Performance of a Static Pluviometer System 

Parth Saxena 

As the frequency and severity of climate-related events such as droughts, floods, and water scarcity continue to escalate, accurate rainfall monitoring becomes increasingly critical. The TinyML organization is involved in solutions for accurate rainfall mapping in developing nations that lack infrastructure and satellite technology. This paper covers various industry methods of measuring rainfall as well as our ground pluviometer system. Our system consists of an inexpensive static rain gauge that can operate for approximately six to twelve months without maintenance. It utilizes resistive sensing technology accompanied by an Arduino Uno to measure the water level depth from its measurement vessel, recording inches of rain hourly. Due to its low maintenance and large size, it has a resolution of 1 mm, making it better suited for areas with heavier rain. This study also provides a side-by-side comparison of our pluviometer system and an industry-standard rain gauge, the MeteoRain 200 Compact, from Barani Systems, with the differences in data being statistically insignificant. By prioritizing cost, sustainability, simplicity, ease of maintenance, and assembly, this research contributes to essential rainfall monitoring solutions, specifically for developing countries. This technology can be used to collect sufficient data for the development of advanced machine-learning models capable of making accurate predictions for agricultural practices, urban planning decisions, water resource management strategies, and disaster preparedness measures, in areas where rainfall monitoring is a challenge. This can be achieved through statistical analysis models such as autoregressive integrated moving average (ARIMA) or deep learning models such as recurrent neural networks, which effectively predict future values based on past data. 

Share this session