We’re a tech startup, doing R&D in Real-time Machine Learning
People learn a lot from very few examples, without forgetting similar things they already knew. But machine “deep learning” can’t do that: it requires offline training with tons of data, and can’t handle updates without retraining. Why?
Why it matters
As companies reorganize their data infrastructure around event streams, they’ll need AI models that can adapt in real time, for a wide range of applications such as analytics, recommender systems, and anomaly detection.
Ideally robots would learn a lot on their own, like kids do, with adults providing safety and pointing out good and bad behavior. However, state-of-the-art reinforcement learning for AI can’t come close to “learning a lesson” in real time.
Our brains learn in real time. If we knew how, it would be a step toward treating brain disorders, improving educational methods, and simply understanding who we are.
“We can’t do this with machines today, which means that we’re missing an essential piece which evolution has figured out but … we haven’t figured out yet.”
— YANN LeCUN, facebook AI Research and NYU, on how does the brain learn so much so quickly
“… there is no model that is capable of assimilating new information while simultaneously and efficiently preserving the old.”
— kemker et al., measuring catastrophic forgetting in neural networks
We’re in the patenting phase now, and can’t say much yet. Please check back here for updates, or follow us on Twitter below. Thanks for your patience!