Quantum deformed binary neural networks

Patent metadata

Field	Value
Publication number	US-12481865-B2
Application number	US-202117491426-A
Country	US
Kind code	B2
Filing date	Sep 30, 2021
Priority date	Oct 1, 2020
Publication date	Nov 25, 2025
Grant date	Nov 25, 2025

Abstract

Certain aspects of the present disclosure provide techniques for processing data in a quantum deformed binary neural network, including: determining an input state for a layer of the quantum deformed binary neural network; computing a mean and variance for one or more observables in the layer; and returning an output activation probability based on the mean and variance for the one or more observables in the layer.

First claim

What is claimed is:

1 . A method for processing data in a quantum deformed binary neural network, comprising:
determining an input state for a layer of the quantum deformed binary neural network based on input image data, wherein the quantum deformed binary neural network is configured to perform the determining of the input state based on the input image data and a quantum phase estimation; computing statistics for one or more observables in the layer, wherein the quantum deformed binary neural network is configured to perform the computing of the statistics based on using a quantum circuit with one or more entangled qubits comprising activation bits and weight bits that are entangled with one another to process the input state; returning, by the quantum deformed binary neural network, an output activation probability based on the statistics for the one or more observables in the layer; and assigning a classification to the input image data based on the output activation probability.

2 . The method of claim 1 , wherein returning the output activation probability comprises performing an additional quantum phase estimation.

3 . The method of claim 1 , wherein determining the input state is performed according to



ψ
j

〉

=



ψ
H

〉

⊗


ψ
〉

W

j
,
:

,



ψ
H

〉

=

⊗
N

i
=
1

⁢

[

p
i

⁡

(

h
i

=
0

)

⁢


0
〉

+

p
i

⁡

(

h
i

=
1

)

⁢


1
〉

]

,

where |ψ j is the input state for the jth layer of the quantum deformed binary neural network, where p i is an ith probability, where h i is an ith activation, where H is a Hamiltonian, and where W j is a jth row of a weight matrix for the quantum deformed binary neural network.

4 . The method of claim 1 , wherein:
the statistics for the one or more observables in the layer comprise a mean and a variance, and computing the statistics is performed according to μ= ψ|DHD -1 |ψ , σ 2 = ψ|DH 2 D -1 |ψ −μ 2 , where μ is the mean and σ 2 is the variance, where D is the quantum circuit, and where H is a Hamiltonian.

5 . The method of claim 1 , wherein returning the output activation probability is performed according to

Φ
⁡

(

-

2
⁢

μ
j

-
N

2
⁢

σ
j

)

,

where Φ is a cumulative distribution function of a standard normal distribution, where μ j is a mean for a jth layer of the quantum deformed binary neural network and σ j is a variance for the jth layer of the quantum deformed binary neural network, and where N is a number of input neurons.

6 . A method for training a quantum deformed binary neural network, comprising:
providing input image data to the quantum deformed binary neural network; receiving output data from the quantum deformed binary neural network, wherein the output data comprises a plurality of class probabilities for the input image data and wherein the quantum deformed binary neural network is configured to generate the output data based on using a quantum circuit with one or more entangled qubits comprising activation bits and weight bits that are entangled with one another to process an input state determined based on the input image data and a quantum phase estimation; calculating a loss based on the plurality of class probabilities and a known class of the input image data; and adjusting a plurality of quantum weights of the quantum deformed binary neural network based on the loss.

7 . The method of claim 6 , wherein the quantum weights comprise quantum wave functions.

8 . The method of claim 6 , wherein the loss comprises a cross-entropy loss.

9 . A processing system, comprising:
at least one memory comprising computer-executable instructions; and at le

Assignees

• Qualcomm Inc

Inventors

• Bondesan Roberto
• Welling Max

Classifications

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06V10/95

  structured as a network, e.g. client-server architectures · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06N10/00

  Quantum computing, i.e. information processing based on quantum-mechanical phenomena · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title