Ann: December 2018 - Quarter Update, page-41

Good question tigers.

The CIFAR-10 dataset contains 60,000 32x32 color images in 10 different classes

The 82 % accuracy reflects Akida processing the above dataset at 11,000 frames per second. Also the Co only used a small portion of the power of Akida to come up with this accuracy that was comparable to the benchmark standard, ie 517,000 not 1.2 million neurons, 616 million synapse against 10 billion synapse that are available.

I initially thought that Louis had made mention that they had succeeded in tweaking an improvement in the CIFAR 10 results but I couldn't find that reference.  While I appreciate jtardif's response, I too have questions about the 92.5% result.

The video shows an accuracy of 925 out of only 1000 images and it runs for 3min.20s.

Questions

Is the video slowed down so we can actually see the demo?

Or is the video showing improved accuracy resulting from slowed down processing time?

Or maybe the result is 92% accurate based on these 1000 images. But it has a lower average accuracy when it processes the full 60,000 images of CIFAR 10.
Or is the 92% accuracy running on the full 1.2 million neurons?   

Have passed these questions on to LDN to clarify.  Keep you posted.