Here is what we are trying to achieve :
- a human holds a pendulum
- pendulum amplifies hand’s micro-movements
- a machine watches a pendulum swinging
- data is analyzed to recover the human thought signature from the brain induced micro-movements of the hand
However, instead of hard coding the algorithm, we teach the computer to recognize by itself what the pendulum is doing,
As a proof of concept, we initially limit our scope to a binary output : the computer has to guess between simple thoughts : a YES and a NO.
So from 10s of a VGA input stream : 640×480 x 30 x 10 = 92.16 MegaPixels the machine has to come with a single bit : 1 or 0 !
That is colossal among data which normally would require millions of neurons for its processing. by using special filters and FFTs, the Fourier analysis of the input frames allows us to extract and track shapes in a simple manner compared to doing it in the spacial domain.
We believe we can reduce the neural network to a manageable size so that we can train our machine with a set of videos before releasing it into the wild for some amazing applications… ranging from simple brain-to-machine interface to artificial telepathy!
A glance at the data flow is depicted below:
Stay tuned for coming results 🙂