Everything to know about tiny AI

594
Tiny AI | iTMunch

The world, today, is dominated by Artificial Intelligence. From automated human resource management to automated streamlined manufacturing and from virtual assistants to self-driving cars, AI has become indispensable in almost all sectors. According to “The Global Artificial Intelligence Trends 2020” survey, conducted by Analytics Insight, 37% of Artificial Intelligence Technologies are adopted by the high tech industry. But, an ever-increasing usage of this technology has led to the overuse of vast amounts of data and computing power while depending on centralized cloud systems. Since extensive energy is utilized in training the modern AI, the process results in large amounts of carbon emissions. Researchers at the University of Massachusetts, Amherst, found out that training a single AI can emit as much as 284 tonnes of carbon dioxide which is equivalent to five times the lifetime carbon emission of an average car. This is not only hazardous for the environment but is also obstructing the speed and privacy of AI applications. In order to counter this problem, Artificial Intelligence researchers, along with tech giants, are focusing on developing Tiny AI.

What is Tiny AI?

Tiny AI is also known as Tiny ML, that is Tiny Machine Learning, which runs on lesser energy. It is nothing but an attempt by academic researchers in developing compressed AI algorithms to reduce the size of existing machine-learning models that utilize large amounts of datasets and computational power. Tiny AI is a step towards ‘Green Computing’ that involves not only shrinking the size of AI models but also accelerating their inference while maintaining their capabilities. The methods used to develop these compressed algorithms are known as distillation methods and can be used to scale down a model 10 times its existing size. 

This can be best understood with the example of Google Assistant software which was previously a program, approximately, 100 gigabytes in size. But, in May 2019, Google CEO – Sundar Pichai announced that the Google Assistant had now been reduced to roughly half a gigabyte in size and that users need not send requests to a remote server. This also drastically reduced the software’s carbon footprint. Another example is Apple’s virtual assistant, Siri, whose speech recognition capabilities are run locally on the iPhone.     

The reduced size of models enables programs to be directly installed on the device itself and does not require users to send data to the cloud or a remote server. This proves that Tiny AI will play a major role in reducing AI technology’s environmental footprints.

SEE ALSO: Artificial Intelligence Develops a Whole New Sport Named ‘Speedgate’

What are the Benefits of Tiny AI?

Being the next AI revolution, Tiny Machine Learning can offer many potential benefits. Let us take a look at a few of them.

  • Energy Efficiency

As mentioned earlier, Artificial Intelligence involves the transmission of large amounts of datasets to and from giant data cloud centers which, ultimately, consumes a lot of energy. With Tiny AI, conventional AI environments can be decentralized, enabling the systems to perform their own data processing. This will also prove to be an energy-efficient method.

  • Privacy

Transmission of data is subject to a possible violation of privacy. As data leaves the device, it can be intercepted by malware and also becomes less secure when stored at a singular location such as the cloud. With the use of Tiny AI, the data will primarily stay on the device, improving privacy and security by minimizing external communication. 

  • Latency

Devices that deploy Artificial Intelligence, transmit data to the cloud for processing and receive a response based on the algorithm’s output. This entire process is dependent on an external factory and that is the speed of the internet. If the internet is slow, the process will also slow down. However, with Tiny AI, as the data would never have to leave the device, there would neither be any external communication nor any dependency. This will result in faster reaction time and reduced latency. 

How Tiny AI can help?

Tiny AI will help in functioning hyper-efficient machine-learning systems. Below is how it can achieve this.

The first sign of a competent Tiny AI system is smarter usage of data. This can be achieved through data reduction techniques or through alternative data resources. Compression strategies such as network pruning can also result in smarter data usage.
Tiny AI can efficiently produce new architectures, new materials, and new structures with the help of 3D integrated systems as a result of the technological advances in nanotechnology. This way, Tiny ML will help in achieving more with less.
Tiny AI will help meet all of the technology’s endpoints as compressed AI algorithms can be easily delivered ‘on-chip’. Energy-efficient processing for edge or extreme edge devices can assist in achieving new learning methodologies such as joint and distributed learning, adaptive inference techniques, and sensor data fusion.

SEE ALSO: Australia joins global artificial intelligence forum as a founding member

Applications of Tiny AI

Tiny AI can find its application in a number of fields. Let us take a look at a few of them mentioned below.

  • Manufacturing Industry

With the innovations in Tiny AI, it might be possible in the future to operate robots on the factory floor that autonomously collaborate with humans. Tiny ML can also help in improving the efficacy of maintenance and design of a company by analyzing and visualizing its sensor data.

  • Mobility and logistics

Autonomous or self-driving cars can be the future with the help of Tiny AI. In order to ensure safety, capacitive sensors in the seat and radar systems in the dashboard can be continuously checked. Other operations inside a car, such as operating the music system or manipulating the AC system, can be controlled with gesture recognition technology.

  • Healthcare

Personalized medicine, as well as treatment, can be achieved with the help of Tiny AI. This is possible through continuous monitoring of health conditions through wearables or by comfortably gathering medical-grade data. 

Potential Challenges

With the invention of Tiny AI, the benefits of Artificial Intelligence would get distributed. This can create multiple challenges. For example, it would be difficult to counter deepfake videos that use AI to fabricate images and sounds that appear to be real. Combating surveillance systems might also prove to be a challenge. And, discriminatory algorithms may also rise. It would be essential to consider such challenges and develop policies to keep a check on them.         

Final Word

As Artificial Intelligence continues to become an integral part of our lives, it has become more than important to note the technology’s effects on the environment. With the help of Tiny AI, we would not only be looking at a greener future but also at limitless possibilities. 

Subscribe to our Newsletter!