Back to results
US 10,650,307 B2Granted
Neuromorphic architecture for unsupervised pattern detection and feature learning
AssigneeInternational Business Machines Corporation
InventorsAngeliki Pantazi, Stanislaw Andrzej Wozniak
Life of this patent
Estimating the term…
Abstract
Embodiments relate to a neuromorphic architecture for unsupervised feature learning using memristive synapses realized using phase-change devices. A spiking neural network architecture for unsupervised pattern learning and a spike-based learning algorithm compatible with phase-change synapses is described, and a feature-learning algorithm capable of performing a sequence of set operations on input patterns is provided. A learning rule for the extraction of certain features of the input that is compatible with spiking neurons and synapses with spike-based plasticity is also provided. The system enables enhanced pattern- and feature-extraction capabilities in neuromorphic systems.
Claims20
An apparatus for computation comprising:
- a first neuronal module comprising a first plurality of neurons, wherein each neuron in the first plurality of neurons operates with a different learning mechanism selected from a first set of learning mechanisms;
- a second neuronal module comprising a second plurality of neurons, wherein each neuron in the second plurality of neurons operates with a different learning mechanism selected from a second set of learning mechanisms;
- and
- an arbitration mechanism operatively connected to the first neuronal module and the second neuronal module;
- wherein the first set of learning mechanisms and the second set of learning mechanisms comprise an application of an inverted learning mechanism following an application of a learning formula comprising spike timing dependent plasticity;
- wherein the spike-timing dependent plasticity comprises an enhancement selected from either selective potentiation or selective depression;
- wherein the enhancement is based on a measure indicative of a learning status, the measure comprising a cumulative strength corresponding to, for the first set of learning mechanisms, a first total post-synaptic potential of activated synapses of the first neuronal module, and for the second set of learning mechanisms, a second total post-synaptic potential of activated synapses of the second neuronal
- module;
- wherein at least one first neuron of the first plurality of neurons of the first neuronal module is decoupled from the first neuronal module, wherein the at least one first decoupled neuron operates independently from the other neurons of the first plurality of neurons, and activates in response to an input matching features learned by the at least one first decoupled neuron;
- and
- wherein at least one second neuron of the second plurality of neurons of the second neuronal module is decoupled from the second neuronal module, wherein the at least one second decoupled neuron operates independently from the other neurons of the second plurality of neurons, and activates in response to an input matching features learned by the at least one second decoupled neuron.
8 more elements
A method for pattern detection and feature learning comprising:
- training a first neuronal module, wherein the first neuronal module comprises a first neuron having a first learning mechanism;
- training a second neuronal module, wherein the second neuronal module comprises a second neuron having a second learning mechanism;
- providing an arbitration mechanism connected to the first neuronal module and the second neuronal module;
- applying an input to the first neuronal module and the second neuronal module;
- wherein the first learning mechanism and the second learning mechanism comprise an application of an inverted learning mechanism following an application of a learning formula comprising spike timing dependent plasticity;
- wherein the spike-timing dependent plasticity comprises an enhancement selected from either selective potentiation or selective depression;
- wherein the enhancement is based on a measure indicative of a learning status, the measure comprising a cumulative strength corresponding to, for the first learning mechanism, a first total post-synaptic potential of activated synapses of the first neuronal module, and for the second learning mechanism, a second total post-synaptic potential of activated synapses of the second neuronal module;
- decoupling at least one first neuron of a first plurality of neurons of the first neuronal module from the first neuronal module, wherein the at least one first decoupled neuron operates independently from the other neurons of the first plurality of neurons, and activates in response to an input matching features learned by the at least one first decoupled neuron;
- and
- decoupling at least one second neuron of a second plurality of neurons of the second neuronal module from the second neuronal module, wherein the at least one second decoupled neuron operates independently from the other neurons of the second plurality of neurons, and activates in response to an input matching features learned by the at least one second decoupled neuron.
7 more elements
An apparatus for computation comprising:
- a first neuronal module comprising a first level tuned neuron, wherein the first level tuned neuron has a first learning mechanism;
- a second neuronal module comprising a second level tuned neuron, wherein the second level tuned neuron has a second learning mechanism;
- and
- an arbitration mechanism operatively connected to the first neuronal module and the second neuronal module and comprising a third neuron that provides a level output that modifies the behavior of the first neuronal module and second neuronal module;
- wherein the first learning mechanism and the second learning mechanism comprise an application of an inverted learning mechanism following an application of a learning formula comprising spike timing dependent plasticity;
- wherein the spike-timing dependent plasticity comprises an enhancement selected from either selective potentiation or selective depression;
- wherein the enhancement is based on a measure indicative of a learning status, the measure comprising a cumulative strength corresponding to, for the first learning mechanism, a first total post-synaptic potential of activated synapses of the first neuronal module, and for the second learning mechanism, a second total post-synaptic potential of activated synapses of the second neuronal module;
- wherein the first level tuned neuron of the first neuronal module is decoupled from the first neuronal module, wherein the first level tuned decoupled neuron operates independently from other neurons of a first plurality of neurons of the first neuronal module, and the first level tuned decoupled neuron activates in response to an input matching features learned by the first level tuned decoupled neuron;
- and
- wherein the second level tuned neuron of the second neuronal module is decoupled from the second neuronal module, wherein the second level tuned decoupled neuron operates independently from other neurons of a second plurality of neurons of the second neuronal module, and the second level tuned decoupled neuron activates in response to an input matching features learned by the second level tuned decoupled neuron.
7 more elements
Dependencies are inferred from the claim text.