The Arduino Nano 33 BLE Sense. (Image Credit: Arduino)
Developing machine learning (ML) algorithms for hardware involves complex mathematical models based on training data. Using this approach enabled it to make predictions without being programmed for that task, which is expensive and complex. Additionally, ML tasks were usually translated to the cloud. Combining these two made computing slower, more expensive, and less predictable. That’s where TinyML comes in.
Tiny Machine Learning (TinyML) is a technology that brings machine and deep learning models on tiny hardware. TinyML performs on-device data analytics for audio, vision, motion, and more. It enables smart devices to make decisions without transmitting data to the cloud. Thanks to its size, TinyML can be utilized in any environment.
Companies are using TinyML to develop product intelligence. For example, Arduino is pushing out TinyML for millions of developers. The company is also collaborating with Edge Impulse to transform the Arduino board into a powerful embedded ML platform, similar to the Arduino Nano 33 BLE Sense and various 32-bit boards. This partnership allows users to run learning models based on artificial neural networks reaching and testing small sensors and low-powered microcontrollers.
Advancements allowed deep learning models to be scaled-down, faster, and operable on embedded hardware via TensorFlow Lite for Microcontrollers, uTensor, and Arm’s CMSIS-NN. However, it’s still complex to develop a quality dataset, extract the correct features, train and deploy the models. This can now be achieved with the emergence of TinyML.
There are some significant implications when it comes to TinyML accessibility, especially in today’s world. For instance, drug development trials typically take five years since millions of design decisions must be made before FDA approval. That can easily be solved by using TinyML’s power and hardware instead of relying on live animals for testing models. Ultimately, this solution would only take 12 months to complete trials.
Additionally, TinyML can listen in on beehives, serving as an alert system for farmers. It’s also capable of detecting anomalies or distress caused by insects as small as a wasp. A small sensor prompts an alert based on a sound model that detects when a hive is being attacked. Farmers can then secure and save the hive.
The demand for affordable, easily deployable COVID-19 solutions is present for humans. Detecting early symptoms could have an immediate impact on millions of lives. Using an Arduino board with TinyML allows a user to identify and alert odd coughing so they can start a defense mechanism to contain the virus.
Edge Impulse and Arduino developed a project that runs TinyML on an Arduino Nano BLE Sense capable of detecting certain coughing sounds in real-time audio. This includes background noise samples with a coughing dataset. A cough detection system, which operates under 20KB of RAM on the Nano BLE Sense, was developed using an optimized TinyML model. The project was designed by Kartik Thakore to help fight COVID-19.
This technique can be used for other embedded audio pattern matching applications in fields like machine monitoring, safety, elderly care, and childcare.
Software developers and engineers can use this technology, allowing it to increase in popularity while unlocking TinyML’s potential. Developers creating embedded systems with ML could develop a model by using their smartphone as the edge device. Its sensors can then collect the data.
There is potential for TinyML to make an impact in healthcare, transportation, retail, wellness, fitness, agriculture, and manufacturing. That’s because billions of tiny devices could serve as an extension of human brains, feelings, and emotions in everyday life.
Have a story tip? Message me at: http://twitter.com/Cabe_Atwell