Hi Stuart,Help needed please on Akida version 2. Seems like we need to focus on the Use Cases for Akida with LSTM (Long-Short Term Memory). I assume this version will also use the Spike-Timing (STDP). While it is all fancy-math under the hood, understanding at a high level what LSTM for Brainchip means to our customers would be great learning.
I would love to hear the new Use Cases. Meaning, there must be companies that want more than Akida 1000 (STDP focused I believe, with less complicated use cases).
I would like to be able to answer the questions:
What are the things version-2 can do, which version-1 cannot?
How does Akida with LSTM change the Total Addressable Market?
What companies are waiting for version-2?
Because of our new marketing team, I assume when this next version comes out, it is going to get a lot of attention. I am trying to get prepared for that and learn more.
View attachment 12623
Akida 1000 does not retain the results of a classification/identification in the SNN. In AV applications it passes these on to a CPU which is programmed to act on the information received from Akida. The CPU may, for example, use a time series of data from Akida to determine the speed and direction of an object in the field of view to track an object.
LSTM will enable Akida to retain a short sequence of previous classifications, and this can be used to determine the speed and direction of objects in the field of view. That is, Akida will be able to track objects. The benefit is that Akida performs these functions much faster and with much less power than a CPU.