Short answer: Yes.
Long answer: See below.
Quotes from BRN Prospectus:
"Generally, ANNs are configured for their specific application via a learning process, with learning typically taking place during a specified training phase."
"BrainChip’s technology, known as ‘Spiking Neuron Adaptive Processor (SNAP), is aiming to redefine how processing works within ANNs."
"SNAP technology learns autonomously by emulating the learning behaviour of biological neural networks. Massive feedback enables the learning process, known as Synaptic Time Dependent Plasticity (STDP) learning. This learning process is directly related to the way biological neural networks learn. Like the brain, SNAP technology learns from usage patterns, evolving its learning through experience."
"BrainChip has successfully demonstrated that SNAP technology learns rapidly from usage patterns by comparing SNAP technology to a software ANN that learnt through a traditional learning method."
"As mentioned above, SNAP technology learns autonomously by emulating the learning behaviour of a human brain.
Testing to date demonstrates the current learning methods of existing technologies such as backpropagation and genetic algorithm are very slow compared to SNAP technology's STDP learning method."
"BrainChip has already successfully demonstrated SNAP technology on a computer with a simulated car that negotiates a random track. Intense spikes are generated by the simulated sensors when the simulated car touches the edge of the track. These spikes train the hardware neural network to keep the car in the middle of the track. Each time the simulated car touches the edge of the track it does not reset itself back to the starting line, rather, the simulated car learns autonomously to navigate the track and keeps going."
Quote from 2015-10-30 Milestone 2 Achieved announcement:
"SNAP autonomously learned to avoid the edges of the track in 0.89 seconds, whilst
executing 3,571,430 neural update cycles."
"True Autonomous Learning
The complexity of the SNAP neuron and synapse design enables true autonomous learning capabilities unlike current deep learning solutions. This is a truly significant feature. The ability to autonomously learn and extract features is just one opportunity BrainChip can exploit."
Quote from 2015-11-05 Milestone 2 Independent Validation announcement:
"Dr Jeffrey Krichmar, Professor of Cognitive Science of University of California Irvine (UCI), has witnessed SNAP working in real time and reviewed the data associated with this significant technical advancement. He has subsequently provided BrainChip with his written confirmation of the achievement of Milestone 2."
It should be clear by now, but just in case it isn't: Brainchip demo car learns every time it touches a wall, all on its own, without any human intervention. This is stated in several ways in the prospectus and in subsequent announcements and in my reading there is no ambiguity to give room for alternative interpretations.
What is clear from the CogniMem demonstration is that training was conducted via the human operator making an intervention to avoid the wall when the simulated car reached the unknown input scenario of a close wall.
I have not researched Cognimem in any detail, but in the demonstration linked above it is plain as day that the process of how the ANN learns what to do with the input of being close to a wall is handled manually by a human in Cognimem whereas BRN clearly claims, and has apparently demonstrated, that this process is performed autonomously in Brainchip.
Expand
(20min delay)