diff --git a/contents/core/ai_for_good/ai_for_good.qmd b/contents/core/ai_for_good/ai_for_good.qmd index d2228fdf..b4c5d4e2 100644 --- a/contents/core/ai_for_good/ai_for_good.qmd +++ b/contents/core/ai_for_good/ai_for_good.qmd @@ -40,7 +40,11 @@ What is special about the SDGs is that they are a collection of interlinked obje A recent study [@vinuesa2020role] highlights the influence of AI on all aspects of sustainable development, particularly on the 17 Sustainable Development Goals (SDGs) and 169 targets internationally defined in the 2030 Agenda for Sustainable Development. The study shows that AI can act as an enabler for 134 targets through technological improvements, but it also highlights the challenges of AI on some targets. The study shows that AI can benefit 67 targets when considering AI and societal outcomes. Still, it also warns about the issues related to the implementation of AI in countries with different cultural values and wealth. -In our book's context, TinyML could help advance at least some of these SDG goals. +While all forms of AI and machine learning have the potential to contribute to the SDGs, the focus of this chapter is on TinyML because of its unique ability to address challenges faced in resource-limited settings. ML systems, particularly those reliant on cloud infrastructure, often require significant computational power, consistent internet connectivity, and substantial financial investment, which can limit their adoption in developing regions or remote areas. In contrast, TinyML enables localized, low-cost, and low-power solutions by running efficient machine learning models directly on microcontrollers. These qualities make it particularly effective for tackling issues such as agricultural monitoring, health diagnostics in underserved areas, and environmental conservation where infrastructure may be minimal. + +By focusing on TinyML, this chapter highlights a branch of AI that provides practical, localized solutions capable of functioning independently of the energy and connectivity demands typically associated with larger-scale ML deployments. TinyML aligns well with the SDG's emphasis on sustainability and accessibility by offering scalable innovations that address global challenges in resource-constrained settings. + +In the context of this book, TinyML could contribute to advancing the following SDG goals: * **Goal 1 - No Poverty:** TinyML could help provide low-cost solutions for crop monitoring to improve agricultural yields in developing countries. @@ -48,7 +52,7 @@ In our book's context, TinyML could help advance at least some of these SDG goal * **Goal 3 - Good Health and Wellbeing:** TinyML could help enable low-cost medical diagnosis tools for early detection and prevention of diseases in remote areas. -* **Goal 6 - Clean Water and Sanitation:** TinyML could monitor water quality and detect contaminants to ensure Access to clean drinking water. +* **Goal 6 - Clean Water and Sanitation:** TinyML could monitor water quality and detect contaminants to ensure access to clean drinking water. * **Goal 7 - Affordable and Clean Energy:** TinyML could optimize energy consumption and enable predictive maintenance for renewable energy infrastructure. @@ -56,13 +60,13 @@ In our book's context, TinyML could help advance at least some of these SDG goal * **Goal 13 - Climate Action:** TinyML could monitor deforestation and track reforestation efforts. It could also help predict extreme weather events. -The portability, lower power requirements, and real-time analytics enabled by TinyML make it well-suited for addressing several sustainability challenges developing regions face. The widespread deployment of power solutions has the potential to provide localized and cost-effective monitoring to help achieve some of the UN's SDGs. In the rest of the sections, we will dive into how TinyML is useful across many sectors that can address the UN SDGs. +The portability, lower power requirements, and real-time analytics enabled by TinyML make it well-suited for addressing several sustainability challenges that developing regions face. The widespread deployment of power solutions has the potential to provide localized and cost-effective monitoring to help achieve some of the UN's SDGs. In the rest of the sections, we will dive into how TinyML is useful across many sectors that can address the UN SDGs. ## Agriculture -Agriculture is essential to achieving many of the UN Sustainable Development Goals, including eradicating Hunger and malnutrition, promoting economic growth, and using natural resources sustainably. TinyML can be a valuable tool to help advance sustainable agriculture, especially for smallholder farmers in developing regions. +Agriculture is essential to achieving many of the UN Sustainable Development Goals, including eradicating hunger and malnutrition, promoting economic growth, and using natural resources sustainably. TinyML can be a valuable tool to help advance sustainable agriculture, especially for smallholder farmers in developing regions. -TinyML solutions can provide real-time monitoring and data analytics for crop health and growing conditions - all without reliance on connectivity infrastructure. For example, low-cost camera modules connected to microcontrollers can monitor for disease, pests, and nutritional deficiencies. TinyML algorithms can analyze the images to detect issues early before they spread and damage yields. Precision monitoring can optimize inputs like water, fertilizer, and pesticides - improving efficiency and sustainability. +TinyML solutions can provide real-time monitoring and data analytics for crop health and growing conditions---all without reliance on connectivity infrastructure. For example, low-cost camera modules connected to microcontrollers can monitor for disease, pests, and nutritional deficiencies. TinyML algorithms can analyze the images to detect issues early before they spread and damage yields. Precision monitoring can optimize inputs like water, fertilizer, and pesticides---improving efficiency and sustainability. Other sensors, such as GPS units and accelerometers, can track microclimate conditions, soil humidity, and livestock wellbeing. Local real-time data helps farmers respond and adapt better to changes in the field. TinyML analytics at the edge avoids lag, network disruptions, and the high data costs of cloud-based systems. Localized systems allow customization of specific crops, diseases, and regional issues. @@ -93,7 +97,7 @@ This exercise teaches you how to predict crop yields in Nepal by combining satel Universal health coverage and quality care remain out of reach for millions worldwide. In many regions, more medical professionals are required to access basic diagnosis and treatment. Additionally, healthcare infrastructure like clinics, hospitals, and utilities to power complex equipment needs to be improved. These gaps disproportionately impact marginalized communities, exacerbating health disparities. -TinyML offers a promising technological solution to help expand Access to quality healthcare globally. TinyML refers to the ability to deploy machine learning algorithms on microcontrollers, tiny chips with limited processing power, memory, and connectivity. TinyML enables real-time data analysis and intelligence in low-powered, compact devices. +TinyML offers a promising technological solution to help expand access to quality healthcare globally. TinyML refers to the ability to deploy machine learning algorithms on microcontrollers, tiny chips with limited processing power, memory, and connectivity. TinyML enables real-time data analysis and intelligence in low-powered, compact devices. This creates opportunities for transformative medical tools that are portable, affordable, and accessible. TinyML software and hardware can be optimized to run even in resource-constrained environments. For example, a TinyML system could analyze symptoms or make diagnostic predictions using minimal computing power, no continuous internet connectivity, and a battery or solar power source. These capabilities can bring medical-grade screening and monitoring directly to underserved patients. @@ -101,9 +105,9 @@ This creates opportunities for transformative medical tools that are portable, a Early detection of diseases is one major application. Small sensors paired with TinyML software can identify symptoms before conditions escalate or visible signs appear. For instance, [cough monitors](https://stradoslabs.com/cough-monitoring-and-respiratory-trial-data-collection-landing) with embedded machine learning can pick up on acoustic patterns indicative of respiratory illness, malaria, or tuberculosis. Detecting diseases at onset improves outcomes and reduces healthcare costs. -A detailed example could be given for TinyML monitoring pneumonia in children. Pneumonia is a leading cause of death for children under 5, and detecting it early is critical. A startup called [Respira xColabs](https://www.samayhealth.com/) has developed a low-cost wearable audio sensor that uses TinyML algorithms to analyze coughs and identify symptoms of respiratory illnesses like pneumonia. The device contains a microphone sensor and microcontroller that runs a neural network model trained to classify respiratory sounds. It can identify features like wheezing, crackling, and stridor that may indicate pneumonia. The device is designed to be highly accessible - it has a simple strap, requires no battery or charging, and results are provided through LED lights and audio cues. +A detailed example could be given for TinyML monitoring pneumonia in children. Pneumonia is a leading cause of death for children under 5, and detecting it early is critical. A startup called [Respira xColabs](https://www.samayhealth.com/) has developed a low-cost wearable audio sensor that uses TinyML algorithms to analyze coughs and identify symptoms of respiratory illnesses like pneumonia. The device contains a microphone sensor and microcontroller that runs a neural network model trained to classify respiratory sounds. It can identify features like wheezing, crackling, and stridor that may indicate pneumonia. The device is designed to be highly accessible, featuring a simple strap, requiring no battery or charging, and providing results through LED lights and audio cues. -Another example involves researchers at UNIFEI in Brazil who have developed a low-cost device that leverages TinyML to monitor heart rhythms. Their innovative solution addresses a critical need - atrial fibrillation and other heart rhythm abnormalities often go undiagnosed due to the prohibitive cost and limited availability of screening tools. The device overcomes these barriers through its ingenious design. It uses an off-the-shelf microcontroller that costs only a few dollars, along with a basic pulse sensor. By minimizing complexity, the device becomes accessible to under-resourced populations. The TinyML algorithm running locally on the microcontroller analyzes pulse data in real-time to detect irregular heart rhythms. This life-saving heart monitoring device demonstrates how TinyML enables powerful AI capabilities to be deployed in cost-effective, user-friendly designs. +Another example involves researchers at UNIFEI in Brazil who have developed a low-cost device that leverages TinyML to monitor heart rhythms. Their solution addresses a critical need by tackling the issue of atrial fibrillation and other heart rhythm abnormalities, which often go undiagnosed due to the prohibitive cost and limited availability of screening tools. It uses an off-the-shelf microcontroller that costs only a few dollars, along with a basic pulse sensor. By minimizing complexity, the device becomes accessible to under-resourced populations. The TinyML algorithm running locally on the microcontroller analyzes pulse data in real-time to detect irregular heart rhythms. This life-saving heart monitoring device demonstrates how TinyML enables powerful AI capabilities to be deployed in cost-effective, user-friendly designs. TinyML's versatility also shows promise for tackling infectious diseases. Researchers have proposed applying TinyML to identify malaria-spreading mosquitoes by their wingbeat sounds. When equipped with microphones, small microcontrollers can run advanced audio classification models to determine mosquito species. This compact, low-power solution produces results in real time, suitable for remote field use. By making entomology analytics affordable and accessible, TinyML could revolutionize monitoring insects that endanger human health. TinyML is expanding healthcare access for vulnerable communities from heart disease to malaria. @@ -121,7 +125,7 @@ This portable, self-contained system shows great promise for entomology. The res The first TinyML contest in healthcare, TDC'22 [@jia2023life], was held in 2022 to motivate participating teams to design AI/ML algorithms for detecting life-threatening ventricular arrhythmias (VAs) and deploy them on Implantable Cardioverter Defibrillators (ICDs). VAs are the main cause of sudden cardiac death (SCD). People at high risk of SCD rely on the ICD to deliver proper and timely defibrillation treatment (i.e., shocking the heart back into normal rhythm) when experiencing life-threatening VAs. -An on-device algorithm for early and timely life-threatening VA detection will increase the chances of survival. The proposed AI/ML algorithm needed to be deployed and executed on an extremely low-power and resource-constrained microcontroller (MCU) (a $10 development board with an ARM Cortex-M4 core at 80 MHz, 256 kB of flash memory and 64 kB of SRAM). The submitted designs were evaluated by metrics measured on the MCU for (1) detection performance, (2) inference latency, and (3) memory occupation by the program of AI/ML algorithms. +An on-device algorithm for early and timely life-threatening VA detection will increase the chances of survival. The proposed AI/ML algorithm needed to be deployed and executed on an extremely low-power and resource-constrained microcontroller (a $10 development board with an ARM Cortex-M4 core at 80 MHz, 256 kB of flash memory and 64 kB of SRAM). The submitted designs were evaluated by metrics measured on the microcontroller for (1) detection performance, (2) inference latency, and (3) memory occupation by the program of AI/ML algorithms. The champion, GaTech EIC Lab, obtained 0.972 in $F_\beta$ (F1 score with a higher weight to recall), 1.747 ms in latency, and 26.39 kB in memory footprint with a deep neural network. An ICD with an on-device VA detection algorithm was [implanted in a clinical trial](https://youtu.be/vx2gWzAr85A?t=2359). @@ -160,7 +164,7 @@ Researchers from Moulay Ismail University of Meknes in Morocco [@bamoumen2022tin ## Disaster Response -In disaster response, speed and safety are paramount. But rubble and wreckage create hazardous, confined environments that impede human search efforts. TinyML enables nimble drones to assist rescue teams in these dangerous scenarios. Processing data locally using TinyML allows for quick interpretation to guide rescue efforts. +In disaster response, speed and safety are paramount, yet rubble and wreckage create hazardous, confined environments that impede human search efforts. TinyML enables nimble drones to assist rescue teams in these dangerous scenarios. Processing data locally using TinyML allows for quick interpretation to guide rescue efforts. When buildings collapse after earthquakes, small drones can prove invaluable. Equipped with TinyML navigation algorithms, micro-sized drones like the CrazyFlie with less than 200 KB of RAM and only 168 MHz CPU clock frequency can safely traverse cramped voids and map pathways beyond human reach [@duisterhof2019learning]. Obstacle avoidance allows these drones to weave through unstable debris. This autonomous mobility lets them rapidly sweep areas humans cannot access. Onboard sensors and TinyML processors analyze real-time data to identify signs of survivors. Thermal cameras can detect body heat, microphones can pick up calls for help, and gas sensors can warn of leaks [@duisterhof2021sniffy].