Tech
Lifelong Deep Neural Network Technology ™: the only AI in the market that never stops learning
In over a decade of work with NASA and DARPA, Neurala has pioneered the first commercially available Continually Learning AI that learns and improve precision seamlessly, one image at a time, on any device.
Today Lifelong-DNN™ (L-DNN) powers 45M+ across cameras, smart phones, drones, and robots.
Developed for NASA to overcome traditional AI bottlenecks
Traditional DNNs require large datasets, servers, and they can’t be easily updated with new knowledge.
NASA asked us to build Lifelong DNN Technology (L-DNN) so that AI could be trained with small datasets, run on all hardware, and become more precise after each new image.
For real-world applications, L-DNN is the 1st commercially available continual AI that is more accurate, customizable, faster, and cheaper than traditional DNNs. [Read White Paper]
L-DNN trains 10,000 faster than DNNs
Faster learning means that
- the cost in time and processing power required to augment the intelligence of the AI are negligible;
- every image can be used to augment AI precision, and
- the precision gains are instantaneous (vs needing to wait for hours-days-weeks for it).
More accurate than traditional DNNs
The patented Lifelong Learning technology emulates the fast learning humans and animals perform on every piece of ‘data’ they encounter.
L-DNN accelerated learning speed means that new learning can occur as new data is available. L-DNN keeps learning where DNNs can’t.
AI learning goes to the Edge
Because of its staggering learning speed, L-DNN learning can happen
- directly on a small compute Edge;
- without requiring Internet connection; and
- without requiring to send images data outside the Edge to the cloud;
With L-DNN, devices can become smarter at every use. [Read White Paper]
LESS DATA + PRIVACY = L-DNN
L-DNN was design with the tough NASA requirement of real-time learning without a server. As such, it requires on average 1000 times less images than DNN to reach its accuracy asymptote.
Also, being able to learn at the Edge and incrementally, you can discard the user data after training, whereas traditional DNN require to keep all data for retraining. [Read White Paper]
Federated Learning built-in
Brain Builder incorporates Lifelong AI technologies to enable rapid AI prototyping and deployment.
This allows users to annotate and train at the same time,just few images, offering instant feedback on performance, and quickly iterating vs. having to wait hours or weeks to understand the results.
L-DNN White Paper
Build, deploy, and refine Custom Visual AI models with Brain Builder
Powered by L-DNN tech, Brain Builder enables users to build and deploy a Custom Visual AI model in minutes, even without AI expertise. Brain Builder uses L-DNN to either speed up dramatically data annotation for traditional DNN building (e.g., in Caffe and Tensor Flow), or directly build and deploy a Custom Vision model either on the cloud, or on a compute Edge. Training occurs instantly, giving immediate feedback to the user, and results can be seen after tens of images rather than thousands.
Enabling hardware agnostic deployment through Brain Builder
