Firing rate independent spike message passing in large scale neural network modeling

A neural network system comprising:

• a number p of neural network portions i with i=1 to p;
• each neural network portion i comprising a number Ni of hardware implemented pre-synaptic neurons capable each of firing an action potential, wherein Ni can be encoded in a word of n bits;
• each neural network portion i being provided for, upon firing of a number Fi of hardware implemented pre-synaptic neurons in a predetermined period of time:
• if Fi.n<Ni, generating a message of a first type, the message comprising a unique address for each hardware implemented pre-synaptic neuron having fired in said predetermined period of time, each address being encoded as a word of n bits;
• and
• if Fi.n>Ni, generating a message of a second type, the message comprising Ni bits and being encoded in words of n bits, wherein each one of said Ni hardware implemented pre-synaptic neurons is represented by a unique bit, each bit having a first value if the hardware implemented pre-synaptic neuron represented by the bit has fired in said predetermined period of time, and a second value otherwise;
• wherein each neural network portion further comprises a hardware implemented post synaptic neuron associated to at least some of the Ni hardware implemented pre-synaptic neurons, the neural network portion being provided to decode either of said message of the first type and said message of the second type and, in response to said decoding, send to the hardware implemented post synaptic neuron information of which one of the Ni hardware implemented pre-synaptic neurons associated to the hardware implemented post-synaptic neuron have fired in said predetermined period of time;
• the hardware implemented pre-synaptic neurons being arranged in a first node of the neural network portion and the hardware implemented post-synaptic neuron being arranged in a second node of the neural network portion.

A method of transmitting firing information in a neural network portion comprising a number N of pre-synaptic neurons capable each of firing an action potential, wherein N can be encoded in a word of n bits;

• the method comprising, upon firing of a number F of pre-synaptic neurons in a predetermined period of time:
• if F.n<N, generating a message of a first type, the message comprising a unique address for each pre-synaptic neuron having fired in said predetermined period of time, each address being encoded as a word of n bits;
• and
• if F.n>N, generating a message of a second type, the message comprising N bits and being encoded in words of n bits, wherein each one of said N pre-synaptic neurons is represented by a unique bit, each bit having a first value if the pre-synaptic neuron represented by the bit has fired in said predetermined period of time, and a second value otherwise.

A method of transmitting firing information in a neural network system, the neural network system comprising a number p of neural network portions i, with i=1 to p, each neural network portion i comprising:

• a number Ni of pre-synaptic neurons capable each of firing an action potential, wherein Ni can be encoded in a word of n bits;
• a post synaptic neuron associated to at least some of the Ni pre-synaptic neurons, the pre-synaptic neurons being arranged in a first node of the neural network portion i and the post-synaptic neuron being arranged in a second node of the neural network portion i;
• and
• the method comprising:
• for each neural network portion i of the neural network system, generating a message of a first or second type according to the following method:
• upon firing of a number Fi of pre-synaptic neurons in a predetermined period of time:
• if Fi.n<Ni, generating a message of a first type, the message comprising a unique address for each pre-synaptic neuron having fired in said predetermined period of time, each address being encoded as a word of n bits;
• and
• if Fi.n>Ni, generating a message of a second type, the message comprising Ni bits and being encoded in words of n bits, wherein each one of said Ni pre-synaptic neurons is represented by a unique bit, each bit having a first value if the pre-synaptic neuron represented by the bit has fired in said predetermined period of time, and a second value otherwise;
• the method further comprising:
• decoding either of said message of the first type and said message of the second type;
• and
• in response to said decoding, sending to the post synaptic neuron information about which ones of the Ni pre-synaptic neurons associated to the post-synaptic neuron have fired in said predetermined period of time.