Big dreams, big possibilities, ... and big challenges. That’s the state of artificial intelligence today. Colossal data sets are pumped into gigantic cloud data centers, to be analyzed by endless algorithms. It’s a process that:
- relies on a tremendous amount of energy
- raises considerable privacy concerns
- requires huge bandwidths
This model is economically and ecologically unsustainable. Big AI is unsuited for offline and realtime decision-making – such as we need for autonomous driving or connected health solutions. And it’s unable to bring us closer to the dream of autonomous, human-like systems.
In order to dream big, we must think small. Or even tiny.
The era of cloud dominance is ending: future AI environments will be decentralized. Edge and extreme edge devices will do their own processing. They will send a minimum amount of data to a central hub. And they will work – and learn – together.
That doesn’t mean that cloud centers will become a thing of the past. They will play a key role, for instance in high-performance computing applications such as DNA analysis. But they need to become drastically more efficient to deal with petabytes of data in hours instead of days.
Tiny AI is imec’s answer towards reaching those goals. It’s an integrated approach that involves co-development of data, hardware and algorithms.
Smarter data usage is the first step towards hyper-efficient AI systems. Imec is looking into solutions such as:
- data reduction through techniques such as surrogate modeling
- alternative data sources such as radar sensing systems
- unsupervised learning methods such as GAN and LSTM
- compression strategies such as network pruning
- AI-assisted data processing
Through advances in nanotechnology, imec keeps pushing the limits of doing more with less:
- new architectures
- new structures such as 3D integrated systems
- new materials
- new packaging solutions
In order to enable energy-efficient processing at the (extreme) edge, imec delivers on-chip AI algorithms for almost all its endpoints. We make them smarter by using:
- new edge-learning methodologies such as distributed and joint learning
- alternative ANN architectures such as spiking
- sensor fusion strategies
- adaptive inference techniques, used for example in the LEAPS project
- transfer learning approaches
Applications of Tiny AI
Soon, just about every business will be a technology business. Here are some key sectors in which Tiny AI will have a massive impact.
Tiny AI in healthcare
The big dream in healthcare is personalized medicine. It will be driven by our improved ability to gather data and turn it into actionable insights. Some examples:
- In genomics, improvements in data usage, algorithms and hardware lead to faster results – demonstrated in our ExaScience Life Lab and the Genome Analytics Platform project.
- Connected health solutions comfortably gather medical-grade data that’s used for clinical research (e.g. neurotechnology) or continuous monitoring through wearable, implantable, ingestible or non-contact technologies.
- A project such as ROBO-CURE uses artificial intelligence for personalized treatments of children with type 1 diabetes.
Tiny AI in Industry 4.0
One day, we will see cobots that autonomously collaborate with humans on the factory floor. And it will be thanks to the innovations in Tiny AI. Already today, the way we make and grow things is being revolutionized by artificial intelligence:
- A project like DyVerSIFy mines, analyzes and visualizes companies’ sensor data in order to detect errors and anomalies, and to improve the efficacy of maintenance and design.
- In agriculture, a combination of affordable sensing and smart algorithms helps to improve yields by means of precision farming – e.g. the MoniCow project or automatic crop disease detection through hyperspectral imaging.
- Real-time and context-aware anomaly detection – explored in the RADIANCE project, for instance – paves the road for automatic decisioning and other efficiency gains.
Tiny AI in mobility and logistics
The world is anxiously waiting for autonomous and connected cars to appear on our roads. In several ways, Tiny AI will play a key role in realizing those possibilities:
- To enhance safety, the driver’s health will be continuously checked with capacitive sensors in the seat and radar systems in the dashboard.
- We will control our in-car entertainment system with a flick of the wrist thanks to gesture recognition technology.
- Augmented insights through cooperative sensor fusion are essential for self-driving cars that rely on several sensors to get a complete picture of their surroundings.
Want to use Tiny AI to make smart solutions?
Imec helps you with technical AI challenges such as:
- extraction of quality data at the (extreme) edge
- adaptive processing of multi-dimensional signals from distributed sensors
- analysis and understanding of anomalies in combined data streams
- automated decisioning in specific environments