# US 11,308,386 B2 · Signal processing method and apparatus based on spiking neural network

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

- Number: US 11,308,386 B2
- Document ID: US11308386B2
- Status: Granted
- Kind: B2
- Application number: 16/423,750
- Filed: 2019-05-28
- Published: 2019-09-19
- Granted: 2022-04-19
- Earliest claimed: 2016-11-28
- Assignees: Huawei Technologies Co., Ltd.
- Applicants: Huawei Technologies Co., Ltd.
- Inventors: Ming Wang; Zheng Yan; Xijun Xue
- Web page: https://patentgrep.com/patent/US11308386B2

## CPC

- G06N 3/049: Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
- G06N 3/02: Neural networks
- G06N 3/065: Analogue means
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A signal processing method and apparatus includes determining a first signal F1(t) output by a first neuron, processing the first signal F1(t) using q orders of synapse weight parameters wq(t), wq−1(t), . . . , w1(t) to obtain a second signal F2(t), and inputting the second signal F2(t) to a second neuron, where the second neuron is in a layer immediately subsequent to that of the first neuron.

## Claims

### Claim 1

A signal processing method implemented in a spiking neural network and comprising:
receiving, by a memristor, a first signal F1(t) output by a first neuron;
processing, by the memristor, the first signal F1(t) using q orders of synapse weight parameters wq(t), wq−1(t), . . . w1(t) to obtain a second signal F2(t), wherein a speed at which an initial function wx+10(t) met by an (x+1)th-order synapse weight parameter of the q orders of synapse weight parameters attenuates with time t is higher than a speed at which an initial function wx0(t) met by an xth-order synapse weight parameter attenuates with the time t, wherein q is a positive integer greater than 1, and wherein 1≤x≤q−1; and
outputting, by the memristor, the second signal F2(t) to a second neuron, wherein the second neuron is a next-layer neuron of the first neuron,
wherein the first signal F1(t) comprises a first sub-signal F1(t1) output by the first neuron at a moment t1,
wherein a qth-order synapse weight parameter wq(t1) at the moment t1 meets the following condition:
wq(t1)=wq(t1−1)+ΔFq(t1),
wherein wq(t1−1) represents a qth-order synapse weight parameter at a moment t1−1, the moment t1−1 is a previous moment of the moment t1, and ΔFq(t1) is an update quantity based on a learning rule and a stimulation effect at the moment t1 of the first sub-signal F1(t1) on the qth-order synapse weight parameter wq(t1−1) at the moment t1−1,
wherein an xth-order synapse weight parameter wq(t1) at the moment t1 meets the following condition:
wq(t1)=wq(t1−1)+ΔFx(t1)+Δwx,x+1(t1)+Δwx,x+2(t1)+ . . . +Δwx,q(t1),
wherein wx(t1−1) represents an xth-order synapse weight parameter at the moment t1−1 and ΔFx(t1) is an update quantity based on the learning rule and the stimulation effect at the moment t1 of the first sub-signal F1(t1) on the xth-order synapse weight parameter wx(t1−1) at the moment t1−1, and
wherein Δwx,x+1(t1), Δwx,x+2(t1), . . . , Δwx,q(t1) are respectively quantities of impact of an (x+1)th-order synapse weight parameter, an (x+2)th-order synapse weight parameter, . . . , and the qth-order synapse weight parameter that are at the moment t1 on the xth-order synapse weight parameter wx(t1).

### Claim 2 (depends on 1)

The signal processing method of claim 1, wherein an initial function w10(t) of a first-order synapse weight parameter of the q orders of synapse weight parameters does not attenuate with time.

### Claim 3 (depends on 1)

The signal processing method of claim 1, wherein an (x+i)th-order synapse weight parameter wx+1(t1) at the moment t1 is greater than or equal to a threshold of an (x+i)th-order synapse weight, and wherein wx,x+i(t1) is not 0.

### Claim 4 (depends on 1)

The signal processing method of claim 1, wherein an (x+i)th-order synapse weight parameter wx+i(t1) at the moment t1 is less than a threshold of an (x+i)th-order synapse weight, wherein Δwx,x+i(t1) is equal to 0, and wherein i=1, 2, . . . , or q−x.

### Claim 5 (depends on 1)

The signal processing method of claim 1, wherein the learning rule is a learning rule based on a biological feature or a supervised learning rule based on an error back propagation mechanism.

### Claim 6 (depends on 1)

The signal processing method of claim 1, wherein the first signal F1(t) comprises a first sub-signal F1(t1) output by the first neuron at a moment t1, and a qth-order synapse weight parameter wq(t1) at the moment t1 meets the following condition:
wq(t1)=wq(t1−1)+ΔFq(t1),
wherein wq(t1−1) represents a qth-order synapse weight parameter at a moment t1−1, the moment t1−1 is a previous moment of the moment t1, and ΔFq(t1) is an update quantity determined based on a learning rule and a stimulation effect at the moment t1 of the first sub-signal F1(t1) on the qth-order synapse weight parameter wq(t1−1) at the moment t1−1.

### Claim 7 (depends on 1)

The signal processing method of claim 1, wherein an (x+1)th-order synapse weight parameter wx+1(t1) at the moment t1 is greater than or equal to a threshold of an (x+1)th-order synapse weight, wherein an xth-order synapse weight parameter wx(t1) at the moment t1 meets the following condition:
wx(t1)=wx(t1−1)+Δwx,x+1(t1),
wherein wx(t1−1) represents an xth-order synapse weight parameter at the moment t1−1, and wherein Δwx,x+1(t1) is a quantity of impact of the (x+1)th-order synapse weight parameter at the moment t1 on the xth-order synapse weight parameter wx(t1).

### Claim 8 (depends on 7)

The signal processing method of claim 7, wherein an (x+1)th-order synapse weight parameter wx+1(t1) at the moment t1 is less than a threshold of an (x+1)th-order synapse weight, and wherein an xth-order synapse weight parameter wx(t1) at the moment t1 meets an initial function wx0(t1).

### Claim 9 (depends on 1)

The signal processing method of claim 1, wherein processing, by the memristor, the first signal F1(t) using the q orders of synapse weight parameters wq(t), wq−1(t), . . . w1(t) to obtain a second signal F2(t) comprises determining, by the memristor, a product of the first signal F1(t) and a first-order synapse weight parameter w1(t) as the second signal F2(t).

### Claim 10

A spiking neural network comprising:
a first neuron configured to output a first signal F1(t);
a memristor coupled to the first neuron and configured to:
receive the first signal F1(t) output by the first neuron;
process the first signal F1(t) using q orders of synapse weight parameters wq(t), wq−1(t), . . . w1(t) to obtain a second signal F2(t), wherein a speed at which an initial function wx+10(t) met by an (x+1)th-order synapse weight parameter of the q orders of synapse weight parameters attenuates with time t is higher than a speed at which an initial function wx0(t) met by an xth-order synapse weight parameter attenuates with the time t, wherein q is a positive integer greater than 1, and wherein 1≤x≤q−1; and
output the second signal F2(t); and
a second neuron coupled to the memristor and configured to receive the second signal F2(t), wherein the second neuron is a neuron in a layer immediately subsequent to a layer of the first neuron,
wherein the first signal F1(t) comprises a first sub-signal F1(t1) output by the first neuron at a moment t1,
wherein a qth-order synapse weight parameter at the moment meets the following condition:
wq(t1)=wq(t1−1)+ΔFq(t1),
wherein wq(t1−1) represents a qth-order synapse weight parameter at a moment t1−1, the moment t1−1 is a previous moment of the moment t1, and ΔFq(t1) is an update quantity based on a learning rule and a stimulation effect at the moment t1 of the first sub-signal F1(t1) on the qth-order synapse weight parameter wq(t1−1) at the moment t1−1,
wherein an xth-order synapse weight parameter wx(t1) at the moment t1 meets the following condition:
wx(t1)=wx(t1−1)+ΔFx(t1)+Δwx,x+1(t1)+Δwx,x+2(t1)+ . . . +Δwx,q(t1),
wherein wx(t1−1) represents an xth-order synapse weight parameter at the moment t1−1 and ΔFx(t1) is an update quantity based on the learning rule and the stimulation effect at the moment t1 of the first sub-signal F1(t1) on the xth-order synapse weight parameter wx(t1−1) at the moment t1−1, and
wherein Δwx,x+1(t1), Δwx,x+2(t1), . . . , Δwx,q(t1) are respectively quantities of impact of an (x+1)th-order synapse weight parameter, an (x+2)th-order synapse weight parameter, . . . , and the qth-order synapse weight parameter that are at the moment t1 on the xth-order synapse weight parameter wx(t1).

### Claim 11 (depends on 10)

The spiking neural network of claim 10, wherein an initial function w10(t) of a first-order synapse weight parameter of the q orders of synapse weight parameters does not attenuate with time.

### Claim 12 (depends on 10)

The spiking neural network of claim 10, wherein an (x+i)th-order synapse weight parameter wx+i(t1) at the moment t1 is greater than or equal to a threshold of an (x+i)th-order synapse weight, and wherein Δwx,x+i(t1) is not 0.

### Claim 13 (depends on 10)

The spiking neural network of claim 10, wherein an (x+i)th-order synapse weight parameter wx+i(t1) at the moment t1 is less than a threshold of an (x+i)th-order synapse weight, wherein Δwx,x+i(t1) is equal to 0, and wherein i=1, 2, . . . , or q−x.

### Claim 14 (depends on 10)

The spiking neural network of claim 10, wherein the learning rule is a learning rule based on a biological feature or a supervised learning rule based on an error back propagation mechanism.

### Claim 15 (depends on 10)

The spiking neural network of claim 10, wherein the first signal F1(t) comprises a first sub-signal F1(t1) output by the first neuron at a moment t1, and a qth-order synapse weight parameter wq(t1) at the moment t1 meets the following condition:
wq(t1)=wq(t1−1)+ΔFq(t1),
wherein wq(t1−1) represents a qth-order synapse weight parameter at a moment t1−1, the moment t1−1 is a previous moment of the moment t1, and ΔFq(t1) is an update quantity determined based on a learning rule and a stimulation effect at the moment t1 of the first sub-signal F1(t1) on the qth-order synapse weight parameter wq(t1−1) at the moment t1−1.

### Claim 16 (depends on 10)

The spiking neural network of claim 10, wherein an (x+1)th-order synapse weight parameter wx+1(t1) at the moment t1 is greater than or equal to a threshold of an (x+1)th-order synapse weight, wherein an xth-order synapse weight parameter wx(t1) at the moment t1 meets the following condition:
wx(t1)=wx(t1−1)+Δwx,x+1(t1),
wherein wx(t1−1) represents an xth-order synapse weight parameter at the moment t1−1, and wherein Δwx,x+1(t1) is a quantity of impact of the (x+1)th-order synapse weight parameter at the moment t1 on the xth-order synapse weight parameter wx(t1).

### Claim 17 (depends on 16)

The spiking neural network of claim 16, wherein an (x+1)th-order synapse weight parameter wx+1(t1) at the moment t1 is less than a threshold of an (x+1)th-order synapse weight, and wherein an xth-order synapse weight parameter wx(t1) at the moment t1 meets an initial function wx0(t1).

### Claim 18 (depends on 10)

The spiking neural network of claim 10, wherein the memristor is further configured to determine a product of the first signal F1(t) and a first-order synapse weight parameter w1(t) as the second signal F2(t).

### Claim 19

A non-transitory computer-readable medium comprising instructions that, when executed by at least one processor, cause a spiking neural network to:
receive a first signal F1(t) output by a first neuron;
process the first signal F1(t) using q orders of synapse weight parameters wq(t), wq−1(t), . . . w1(t) to obtain a second signal F2(t), wherein a speed at which an initial function wx+10(t) met by an (x+1)th-order synapse weight parameter of the q orders of synapse weight parameters attenuates with time t is higher than a speed at which an initial function wx0(t) met by an xth-order synapse weight parameter attenuates with the time t, wherein q is a positive integer greater than 1, and wherein 1≤x≤q−1; and
output the second signal F2(t) to a second neuron, wherein the second neuron is a neuron in a layer immediately subsequent to a layer of the first neuron,
wherein the first signal F1(t) comprises a first sub-signal F1(t1) output by the first neuron at a moment t1,
wherein a qth-order synapse weight parameter wq(t1) at the moment t1 meets the following condition:
wq(t1)=wq(t1−1)+ΔFq(t1),
wherein wq(t1−1) represents a qth-order synapse weight parameter at a moment t1−1, the moment t1−1 is a previous moment of the moment t1, and ΔFq(t1) is an update quantity based on a learning rule and a stimulation effect at the moment t1 of the first sub-signal F1(t1) on the qth-order synapse weight parameter wq(t1−1) at the moment t1−1,
wherein an xth-order synapse weight parameter wx(t1) at the moment t1 meets the following condition:
wx(t1)=wx(t1−1)+ΔFx(t1)+Δwx,x+1(t1)+Δwx,x+2(t1)+ . . . +Δwx,q(t1),
wherein wx(t1−1) represents an xth-order synapse weight parameter at the moment t1−1 and ΔFF(t1) is an update quantity based on the learning rule and the stimulation effect at the moment t1 of the first sub-signal F1(t1) on the xth-order synapse weight parameter wx(t1−1) at the moment t1−1, and
wherein Δwx,x+1(t1), Δwx,x+2(t1), . . . , Δwx,q(t1) are respectively quantities of impact of an (x+1)th-order synapse weight parameter, an (x+2)th-order synapse weight parameter, . . . , and the qth-order synapse weight parameter that are at the moment t1 on the xth-order synapse weight parameter wx(t1).

### Claim 20 (depends on 19)

The non-transitory computer-readable medium of claim 19, wherein an initial function w10(t) of a first-order synapse weight parameter of the q orders of synapse weight parameters does not attenuate with time.