myASX, as I am asking questions I am prepared to answer them.
As cafa rightly points out, most people here know virtually nothing about Brainchip, and I include myself in that number which is why I am asking my questions.
Unfortunately, when it comes to technical exposition of the product Peter produces convoluted vagaries which do not help - we need a little light in this darkness. I have read his book, Higher Intelligence, which is written with a simplicity and clarity which would help here.
I had read the above and spent some time trying to untangle what he means, but alas it eludes my mensa mind.
The bit you highlight is particularly interesting as, on its face, it is a nonsense; it runs right up against the "Tay problem" you mentioned in an earlier post. Without any instruction / direction / programming, and left to simply learn - you cannot be sure of what you will get, whether it will be what you want, or if it will even be useful. For example, given a video steam from a camera trained on the highway, and left to simply learn, it must be as likely to calculate the percentage and frequency of red in the image as it is to identify and count cars.
I do understand, and have had an interest in neural networks, for over 20 years, I do understand the basics. The problem is that Peter is possibly not using language as it is understood in the machine learning community. It is unclear if his favorite phrase of "autonomous learning" is synonymous with "unsupervised learning" or he means something different. If he means something different, it needs definition and clarification to communicate his claimed point of difference to a community with an established understanding of what learning is. Even the use of "feature extraction" and "unsupervised learning" are used in ways that could suggest something other than the meaning generally understood by those practiced in the art. Certainly the majority of day traders here demonstrate from their comments that they do not understand the technical meaning of those industry standard terms.
I was also going to explain in simple terms how the sentence you highlighted ignores the fact that a neural network must be customised for each particular application, but that would make this post way too long, but I will take the time in a separate post if somebody asks and really wants to learn.
So there you have it; short answer and long answer. I understand the generalities of the field, but not the specifics of Brainchip - I would love to if I could be provided with relevant technical information using industry standard terminology used in industry standard ways.
Expand
(20min delay)