Back to results
US 11,893,471 B2Granted
Encoding and decoding information and artificial neural networks
AssigneeInait SA
InventorsHenry Markram, Ran Levi, Kathryn Pamela Hess Bellwald, Felix Schuermann
Life of this patent
Estimating the term…
Abstract
In one implementation, a method is implemented by a neural network device and includes inputting a representation of topological structures in patterns of activity in a source neural network, wherein the activity is responsive to an input into the source neural network, processing the representation, and outputting a result of the processing of the representation. The processing is consistent with a training of the neural network to process different such representations of topological structures in patterns of activity in the source neural network.
Claims23
A device comprising:
- a neural network implemented in a data processing apparatus and trained to produce:
- in response to a first input, a first representation of presence or absence of each of a plurality of individual topological patterns of signal transmission activity that would arise between nodes along links in a source recurrent neural network in response to the first input, wherein
- the source recurrent neural network differs from the neural network,
- the first representation only approximates the presence or the absence of each of the individual topological patterns of signal transmission activity that would arise in the source recurrent neural network in response to the first input, and
- the first representation does not encode locations of the nodes and the links in a graph of the source recurrent neural network;
- in response to a second input, a second representation of presence or absence of each of the plurality of the individual topological patterns of signal transmission activity that would arise between the nodes along the links in the source recurrent neural network in response to the second input, wherein the second representation only approximates the presence or the absence of each of the individual topological patterns of signal transmission activity that would arise in the source recurrent neural network in response to the second input and the second representation does not encode locations of the nodes and the links in the graph of the source recurrent neural network;
- and
- in response to a third input, a third representation of presence or absence of each of the plurality of the individual topological patterns of signal transmission activity that would arise between the nodes along the links in the source recurrent neural network in response to the third input, wherein the third representation only approximates the presence or the absence of each of the individual topological patterns of signal transmission activity that would arise in the source recurrent neural network in response to the third input and the second representation does not encode locations of the nodes and the links in the graph of the source recurrent neural network.
5 more elements
A device comprising:
- a neural network implemented in a data processing apparatus and configured to be coupled to input and process representations of the presence or absence of individual topological patterns of signal transmission activity that arise between nodes along links in a source recurrent artificial neural network in which nodes operate as accumulators, wherein the representations do not encode locations of the nodes and the links in a graph of the source recurrent artificial neural network and the topological structures all encompass three or more nodes in the source recurrent artificial neural network and three or more edges between the nodes, wherein the input and processed topological patterns of signal transmission activity arise in the source recurrent artificial neural network in response to a different input, wherein the neural network differs from the source recurrent artificial neural network and is trained to process the representations and produce responsive outputs.
A method implemented by a neural network device, the method comprising:
- inputting a representation of presence or absence of individual topological patterns of signal transmission activity that arise between nodes along links in a source recurrent neural network, wherein the signal transmission activity is responsive to an input into the source recurrent neural network and the representation does not encode locations of the nodes and the links in a graph of the source recurrent neural network;
- processing the representation, wherein the processing is consistent with a training of the neural network to process different such representations of the presence or absence of the individual topological patterns of signal transmission activity in the source recurrent neural network;
- and
- outputting a result of the processing of the representation.
1 more element
Dependencies are inferred from the claim text.