Information processing apparatus, Ising unit, and information processing apparatus control method

Patent metadata

Field	Value
Publication number	US-10762415-B2
Application number	US-201715690313-A
Country	US
Kind code	B2
Filing date	Aug 30, 2017
Priority date	Sep 9, 2016
Publication date	Sep 1, 2020
Grant date	Sep 1, 2020

Abstract

Each energy value calculation circuit calculates an energy value, based on a sum total of values obtained by multiplying state values of a plurality of second neurons coupled with a first neuron by corresponding weighting values indicating coupling strengths, and updates the energy value, based on identification information of an updated neuron whose state is updated among the plurality of second neurons and a state value of the updated neuron. Each state transition determination circuit outputs, based on a second energy value and a noise value, a determination signal indicating a determination result of whether a change in a state value of the first neuron is possible. An updated neuron selection circuit selects, based on received determination signals, a first neuron a change in whose state value is possible and outputs identification information of the selected first neuron as identification information of the updated neuron.

First claim

What is claimed is:

1. An information processing apparatus comprising:
a plurality of Ising units each including:
a calculation circuit which calculates a second energy value obtained by updating, based on a piece of first identification information of an updated neuron whose state is updated among a plurality of second neurons coupled with a first neuron of a plurality of neurons and a second state value indicative of a state of the updated neuron, a first energy value calculated based on a sum total of values obtained by multiplying a plurality of first state values indicative of states of the plurality of second neurons by weighting values corresponding to the plurality of second neurons among a plurality of weighting values each indicative of a coupling strength between the first neuron and one of the plurality of second neurons; and
a state transition determination circuit which outputs a determination signal indicative of a determination result of determining, based on the second energy value or a change amount of an energy function of a neural network at the time of a change in a state of the first neuron and a first noise value, whether a change in a third state value indicative of a state of the first neuron is possible, the plurality of Ising units outputting determination signals regarding different first neurons; and

an updated neuron selection circuit which selects a first neuron of the plurality of neurons, for which the change in the third state value is determined to be possible, based on the determination signals outputted by the plurality of Ising units and which outputs a piece of second identification information of the selected first neuron as the piece of first identification information.

2. The information processing apparatus according to claim 1 , wherein the state transition determination circuit determines, based on a change in a comparison result obtained by comparing an addition result obtained by adding together the second energy value and the first noise value with a threshold, whether a change in the third state value is possible.

3. The information processing apparatus according to claim 1 , wherein the state transition determination circuit determines, based on a change in a comparison result obtained by comparing an addition result obtained by adding together the change amount of the energy function of the neural network and the first noise value with a threshold, whether a change in the third state value is possible.

4. The information processing apparatus according to claim 1 , wherein, when there are a plurality of first neurons for which a change in the third state value is determined to be possible, the updated neuron selection circuit selects one of the plurality of first neurons based on a second noise value and outputs a piece of second identification information of the selected first neuron as the piece of first identification information.

5. The information processing apparatus according to claim 4 , wherein:
the updated neuron selection circuit includes a plurality of selection circuits connected to form a plural-stage tree structure;
each of a plurality of first selection circuits at a first stage, of the plurality selection circuits, receives two or more of the determination signals outputted by the plurality of Ising units and generates the piece of second identification information corresponding to a determination signal selected based on the second noise value from the two or more determination signals;
each of a plurality of second selection circuits at second and later stages, of the plurality of selection circuits, receives two or more of determination signals selected by the plurality of first selection circuits or a plurality of third selection circuits at a preceding stage and two or more pieces of second identification information and selects one of the two or more determination signals and one of the two or more pieces of second identification information based on the second noise value; and
a piece of second identification information selected by a third selection circuit at a last stage, of the plurality of second selection circuits, is the piece of first identification information.

6. The information processing apparatus according to claim 1 , wherein the state transition determination circuit calculates, based on the change amount of the energy function of the neural network, a probability value at which update of the state of the first neuron is allowed and outputs the determination signal based on the first noise value and the probability value.

7. An Ising unit comprising:
a calculation circuit which calculates a second energy value obtained by updating, based on a piece of first identification information of an updated neuron whose state is updated among a plurality of second neurons coupled with a first neurons of a plurality of neurons and a second state value indicative of a state of the updated neuron, a first energy value calculated based on a sum total of values obtained by multiplying a plurality of first state values indicative of states of the plurality of second neurons by weighting values corresponding to the plurality of second neurons among a plurality of weighting values each indicative of a coupling strength between the first neuron and one of the plurality of second neurons; and
a state transition determination circuit which outputs a determination signal indicative of a determination result of determining, based on the second energy value or a change amount of an energy function of a neural network at the time of a change in a state of the first neuron and a noise value, whether a change in a third state value indicative of a state of the first neuron is possible.

8. An information processing apparatus control method comprising
selecting, by an updated neuron selection circuit, a first neuron for which a change in a third state value is determined to be possible, from a plurality of neurons, based on determination signals outputted by a plurality of Ising units and supplying a piece of second identification information of the selected first neuron as a piece of first identification information, each of the plurality of Ising units including:
a calculation circuit which calculates a second energy value obtained by updating, based on the piece of first identification information of an updated neuron whose state is updated among a plurality of second neurons coupled with a first neuron of the plurality of neurons and a second state value indicative of a state of the updated neuron, a first energy value calculated based on a sum total of values obtained by multiplying a plurality of first state values indicative of states of the plurality of second neurons by weighting values corresponding to the plurality of second neurons, among a plurality of weighting values each indicative of a coupling strength between the first neuron and one of the plurality of second neurons; and
a state transition determination circuit which outputs a determination signal indicative of a determination result of determining, based on the second energy value or a change amount of an energy function of a neural network at the time of a change in a state of the first neuron and a noise value, whether a change in the third state value indicative of a state of the first neuron is possible, the plurality of Ising units outputting determination signals regarding different first neurons.

Assignees

• Fujitsu Ltd

Inventors

• Tamura Hirotaka
• Matsubara Satoshi

Classifications

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/063 Primary

  using electronic means · CPC title

• G06N3/0445

  Physics · mapped topic

• G06N3/0472 Primary

  Physics · mapped topic