# US 2019/0057301 A1 · Neuromorphic processing apparatus

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0057301 A1
- Document ID: US20190057301A1
- Status: Published application
- Kind: A1
- Application number: 15/680,140
- Filed: 2017-08-17
- Published: 2019-02-21
- Earliest claimed: 2017-08-17
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Angeliki Pantazi; Severin Sidler; Stanislaw A. Wozniak
- Web page: https://patentgrep.com/patent/US20190057301A1

## CPC

- G06N 3/049: Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
- G06N 3/065: Analogue means
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Neuromorphic processing apparatus is provided. The present invention may include a spiking neural network comprising a set of input spiking neurons each connected to each of a set of output spiking neurons via a respective synapse for storing a synaptic weight which is adjusted for that synapse in dependence on network operation in a learning mode of the apparatus, and each synapse is operable to provide a post-synaptic signal, dependent on its synaptic weight, to its respective output neuron. The present invention may further include a pre-processor unit adapted to process input data, defining a pattern of data points, to produce a first set of input spike signals which encode values representing respective data points, and a second set of input spike signals which encode values complementary to respective said values representing data points, and to supply the input spike signals to respective predetermined input neurons of the network.

## Claims

### Claim 1

An apparatus, comprising:
a spiking neural network comprising a set of input spiking neurons each connected to each of a set of output spiking neurons via a respective synapse for storing a synaptic weight which is adjusted for that synapse in dependence on network operation in a learning mode of the apparatus, each synapse being operable, in response to receipt of a spike from its respective input neuron, to provide a post-synaptic signal, dependent on said weight, to its respective output neuron; and
a preprocessor unit adapted to process input data, defining a pattern of data points, to produce a first set of input spike signals which encode values representing respective data points, and a second set of input spike signals which encode values complementary to respective said values representing data points, and to supply the input spike signals to respective predetermined input neurons of said network.

### Claim 2 (depends on 1)

The apparatus according to claim 1, wherein the apparatus is operable in said learning mode and an application mode, wherein said learning mode, the preprocessor unit is operable to process at least one set of training data, defining a training pattern of data points, to produce said first and second sets of input spike signals for the training pattern and to supply the input spike signals to respective input neurons of the network in accordance with a predetermined mapping, and the network is adapted such that the synaptic weights of respective synapses are adjusted in dependence on network operation in response to the input spike signals in the learning mode, and wherein said application mode, the preprocessor unit is operable to process application data, defining an application pattern of data points, to produce said first and second sets of input spike signals for the application pattern and to supply the input spike signals to respective input neurons of the network in accordance with said predetermined mapping, and the network is adapted such that the synaptic weights of the synapses remain as adjusted in the learning mode.

### Claim 3 (depends on 2)

The apparatus according to claim 2, wherein in the learning mode, the network is adapted such that the synaptic weight of each synapse is adjusted in dependence on timing of at least one of spikes produced by the input neuron of that synapse and spikes produced by the output neuron of that synapse.

### Claim 4 (depends on 3)

The apparatus according to claim 3, wherein in the learning mode, the synaptic weight of each synapse is adjusted in dependence on relative timing of spikes produced by the input and output neurons of that synapse.

### Claim 5 (depends on 2)

The apparatus according to claim 2, comprising:
a postprocessor unit connected to the output neurons of the network and adapted to classify said pattern of data points in dependence on spikes produced by the output neurons in response to said input spike signals in the application mode.

### Claim 6 (depends on 2)

The apparatus according to claim 2, wherein each synapse comprises at least one memristive device for storing said synaptic weight.

### Claim 7 (depends on 6)

The apparatus according to claim 6, wherein said memristive device comprises a phase-change memory cell.

### Claim 8 (depends on 1)

The apparatus according to claim 1, wherein the preprocessor unit is adapted such that the second set of input spike signals comprises signals encoding complementary values for all of the data points.

### Claim 9 (depends on 1)

The apparatus according to claim 1, wherein the preprocessor unit is adapted such that the second set of input spike signals comprises signals encoding complementary values for a predetermined subset of the data points.

### Claim 10 (depends on 1)

The apparatus according to claim 1, wherein the preprocessor unit is further adapted to identify active data points from said input data, and to produce the second set of input spike signals to comprise signals encoding complementary values for data points in a portion of the pattern containing said active data points.

### Claim 11 (depends on 1)

The apparatus according to claim 1, wherein the preprocessor unit is adapted to produce the first set of input spike signals by encoding the input data corresponding to respective data points via a predetermined encoding scheme, and to produce the second set of input spike signals by producing data complementary to the input data corresponding to respective data points, and encoding the complementary data via said predetermined encoding scheme.

### Claim 12 (depends on 1)

The apparatus according to claim 1, wherein the preprocessor unit is adapted to produce the first set of input spike signals by encoding the input data corresponding to respective data points via a predetermined encoding scheme, and to produce the second set of input spike signals by encoding the input data corresponding to respective data points via the inverse of said predetermined encoding scheme.

### Claim 13 (depends on 1)

The apparatus according to claim 1, wherein the preprocessor unit is adapted to produce the input spike signals using a rate encoding scheme.

### Claim 14 (depends on 1)

The apparatus according to claim 1, wherein the preprocessor unit is adapted to produce the input spike signals using a spike correlation encoding scheme.

### Claim 15 (depends on 1)

The apparatus according to claim 1, comprising:
a postprocessor unit connected to the output neurons of the network and adapted to classify said pattern of data points in dependence on spikes produced by the output neurons in response to said input spike signals.

### Claim 16 (depends on 1)

The apparatus according to claim 1, wherein each synapse comprises at least one memristive device for storing said synaptic weight.

### Claim 17 (depends on 16)

The apparatus according to claim 16, wherein said memristive device comprises a phase-change memory cell.

### Claim 18 (depends on 1)

The apparatus according to claim 1, wherein said pattern of data points comprises an image and said input data comprises image data.
