JPH05233859A - Optical neural network system - Google Patents

Optical neural network system

Info

Publication number
JPH05233859A
JPH05233859A JP3665092A JP3665092A JPH05233859A JP H05233859 A JPH05233859 A JP H05233859A JP 3665092 A JP3665092 A JP 3665092A JP 3665092 A JP3665092 A JP 3665092A JP H05233859 A JPH05233859 A JP H05233859A
Authority
JP
Japan
Prior art keywords
element array
light
weight
plural
spatial light
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
JP3665092A
Other languages
Japanese (ja)
Other versions
JP2530404B2 (en
Inventor
Katsumi Kaizu
勝美 海津
Satoru Yoshida
悟 吉田
Tadashi Sone
正 曽根
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nippon Telegraph and Telephone Corp
Original Assignee
Nippon Telegraph and Telephone Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nippon Telegraph and Telephone Corp filed Critical Nippon Telegraph and Telephone Corp
Priority to JP4036650A priority Critical patent/JP2530404B2/en
Publication of JPH05233859A publication Critical patent/JPH05233859A/en
Application granted granted Critical
Publication of JP2530404B2 publication Critical patent/JP2530404B2/en
Anticipated expiration legal-status Critical
Expired - Fee Related legal-status Critical Current

Links

Abstract

PURPOSE:To provide the optical neural network system which can considerably shorten time required for a learning processing. CONSTITUTION:In this optical neural network system consisting of an input part to generate plural beams corresponding to plural input signals, an intermediate part to execute prescribed arithmetic by changing the transmissivity of light corresponding to the plural beams from the input part, and an output part to convert the plural beams transmitted through the intermediate part to plural output signals, the intermediate part is constituted by using a spatial optical modulation element array 6 composed of plural spatial optical modulation elements to set error between an output signal and a teaching signal corresponding to the plural beams from the input part, and a weight setting element array 7 consisting of plural weight setting elements provided with photodetection parts to photodetect one part of light transmitted through the spatial optical modulation element array 6 and to convert it to an electric signal, calculation parts to calculate weight based on the electric signal, and spatial optical modulation parts to set the calculated weight as the transmissivity of light.

Description

【発明の詳細な説明】Detailed Description of the Invention

【0001】[0001]

【産業上の利用分野】本発明は、光ニューラルネットワ
ークシステムの改良に関するものである。
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement of an optical neural network system.

【0002】[0002]

【従来の技術】ニューラルネットワークシステムは、神
経細胞間の情報伝達の動作を模擬することにより、従来
の計算機では不得意とされる認識処理や連想処理を行わ
せようとするもので、現在、そのための研究が活発に行
われている。
2. Description of the Related Art A neural network system attempts to perform recognition processing and associative processing, which are weak in conventional computers, by simulating the operation of information transmission between nerve cells. Is actively researched.

【0003】このようなニューラルネットワークシステ
ムの一つとして、信号の伝搬及び信号同士の積和演算を
光の空間伝搬特性を用いて行う光ニューラルネットワー
クシステムがある。
As one of such neural network systems, there is an optical neural network system that performs signal propagation and product-sum calculation of signals using the spatial propagation characteristics of light.

【0004】図2は従来の光ニューラルネットワークシ
ステムの一例、例えばOptics Letters,Vol.12,No.6
に示された従来の学習形光ニューロコンピュータシステ
ムの構成例を示すもので、図中、1は発光素子アレイ、
2は空間光変調素子アレイ、3は受光素子アレイ、4は
コンピュータである。
FIG. 2 shows an example of a conventional optical neural network system, for example, Optics Letters, Vol.12, No.6.
FIG. 1 shows a configuration example of the conventional learning type optical neurocomputer system shown in FIG.
Reference numeral 2 is a spatial light modulation element array, 3 is a light receiving element array, and 4 is a computer.

【0005】次に、このシステムの学習動作について説
明するが、発光素子アレイ1のi番目の素子から発した
光は空間光変調素子アレイ2のi行目を均一に照射する
ものとし、また、空間光変調素子アレイ2のj列目を透
過した光は受光素子アレイ3のj番目の素子に全て集光
されるものとする。
Next, the learning operation of this system will be described. It is assumed that the light emitted from the i-th element of the light emitting element array 1 irradiates the i-th row of the spatial light modulation element array 2 uniformly. It is assumed that the light transmitted through the j-th column of the spatial light modulation element array 2 is all focused on the j-th element of the light-receiving element array 3.

【0006】コンピュータ4は、始めに学習させようと
するデータOi を発光素子アレイ1の発光強度として、
また、神経細胞間の結合強度に相当する重みWijを空間
光変調素子アレイ2の光の透過率として設定する。これ
により、受光素子アレイ3から出力されるデータI
j は、 となる。
The computer 4 first sets the data O i to be learned as the light emission intensity of the light emitting element array 1,
Further, a weight W ij corresponding to the coupling strength between nerve cells is set as the light transmittance of the spatial light modulator array 2. As a result, the data I output from the light receiving element array 3
j is Becomes

【0007】次に、コンピュータ4は受光素子アレイ3
からの出力データIj と、発光素子アレイ1に入力した
データOi に対する出力の期待値(教師データ)とを比
較し、その誤差δj を求めるとともに、受光素子アレイ
3からの出力データをこれに近づけるよう、空間光変調
素子アレイ2の各セル(素子)に設定する新たな重みW
ij(new) を、 Wij(new) =Wij+ηOi δj ……(2) (但し、ηは重みの更新の度合いを決める係数)により
計算し、この値を空間光変調素子アレイ2の各セルへ転
送する。
Next, the computer 4 uses the light receiving element array 3
Output data I j is compared with the expected value (teacher data) of the output with respect to the data O i input to the light emitting element array 1, the error δ j is obtained, and the output data from the light receiving element array 3 is calculated. New weight W to be set for each cell (element) of the spatial light modulation element array 2 so as to approach
ij (new) is calculated by W ij (new) = W ij + ηO i δ j (2) (where η is a coefficient that determines the degree of update of the weight), and this value is calculated. To each cell of.

【0008】前述した操作を学習すべき複数のデータに
対して繰返し実施することにより、最終的に出力データ
を教師データとほぼ一致させることが可能となる。
By repeating the above-mentioned operation for a plurality of data to be learned, it becomes possible to finally make the output data substantially coincide with the teacher data.

【0009】[0009]

【発明が解決しようとする課題】しかしながら、前述し
た光ニューラルネットワークシステムでは、重みWij
算出が外部のコンピュータ4で行われるため、システム
内の結合数が増加すると、これに比例して計算時間が増
加する。また、計算した重みを空間光変調素子アレイの
各セルへ転送する処理に要する時間も必要となる。
However, in the above-described optical neural network system, since the weight W ij is calculated by the external computer 4, when the number of connections in the system increases, the calculation time increases in proportion to this. Will increase. In addition, the time required to transfer the calculated weight to each cell of the spatial light modulation element array is also required.

【0010】例えば、250個の重みからなる光ニュー
ラルネットワークシステムで8個のデータを学習する場
合、1回の学習で2000回の重みの演算と転送が必要
となる。一般に、学習が集束するためには104 程度の
学習回数が必要となるため、この場合の演算回数は2×
107 回となる。1回の重みの演算と転送に必要な時間
を2msとすると、総学習時間は4×104 秒(約11時
間)となり、必ずしも光の並列演算性、並列伝達性を有
効に活用しているとはいえず、効率的な学習が不可能で
あった。
For example, when eight data are learned by an optical neural network system having 250 weights, it is necessary to calculate and transfer the weights 2000 times by one learning. Generally, about 10 4 learning times are required for the learning to converge, so the number of calculations in this case is 2 ×.
10 7 times. Assuming that the time required for one weight calculation and transfer is 2 ms, the total learning time is 4 × 10 4 seconds (about 11 hours), and the optical parallel computing performance and parallel transmissibility are not always effectively utilized. However, efficient learning was impossible.

【0011】本発明は前述した従来の問題点に鑑み、光
の並列演算性、並列伝達性を有効に活用し、学習処理に
要する時間を大幅に短縮し得る光ニューラルネットワー
クシステムを提供することを目的とする。
In view of the above-mentioned conventional problems, the present invention provides an optical neural network system which can effectively reduce the time required for learning processing by effectively utilizing the parallel computing property and parallel transmissibility of light. To aim.

【0012】[0012]

【課題を解決するための手段】本発明では前記目的を達
成するため、複数の入力信号に対応した複数の光を発生
する入力部と、該入力部からの複数の光に対して光の透
過率を変化させることにより所定の演算を行わせる中間
部と、該中間部を透過した複数の光を複数の出力信号に
変換する出力部とからなる光ニューラルネットワークシ
ステムにおいて、前記中間部を、入力部からの複数の光
に対して出力信号と教師信号との誤差を設定する複数の
空間光変調素子からなる空間光変調素子アレイと、該空
間光変調素子アレイを透過した光の一部を受光し電気信
号に変換する受光部、該電気信号に基いて重みを計算す
る計算部及び該計算した重みを光の透過率として設定す
る空間光変調部を備えた複数の重み設定素子からなる重
み設定素子アレイとを用いて構成した光ニューラルネッ
トワークシステムを提案する。
In order to achieve the above-mentioned object, the present invention achieves the above object, and an input section for generating a plurality of lights corresponding to a plurality of input signals, and a transmission of light for the plurality of lights from the input section. In an optical neural network system comprising an intermediate section for performing a predetermined calculation by changing a rate and an output section for converting a plurality of lights transmitted through the intermediate section into a plurality of output signals, the intermediate section is input. Spatial light modulation element array including a plurality of spatial light modulation elements for setting an error between an output signal and a teacher signal for a plurality of light from the spatial light modulator, and receiving a part of light transmitted through the spatial light modulation element array A weight setting unit including a plurality of weight setting elements including a light receiving unit for converting into an electric signal, a calculation unit for calculating a weight based on the electric signal, and a spatial light modulator for setting the calculated weight as a light transmittance. Element array We propose an optical neural network system using the.

【0013】[0013]

【作用】本発明によれば、入力部からの複数の光に対し
て出力信号と教師信号との誤差を設定する複数の空間光
変調素子からなる空間光変調素子アレイにより、重みの
更新に必要な誤差情報を光を用いて伝達でき、また、空
間光変調素子アレイを透過した光の一部を受光し電気信
号に変換する受光部、該電気信号に基いて重みを計算す
る計算部及び該計算した重みを光の透過率として設定す
る空間光変調部を備えた複数の重み設定素子からなる重
み設定素子アレイにより、光による誤差情報に基いて重
みの更新を全セルで同時に実行できる。
According to the present invention, it is necessary to update the weight by the spatial light modulation element array composed of a plurality of spatial light modulation elements for setting the error between the output signal and the teacher signal for a plurality of light from the input section. Error information can be transmitted using light, and a light receiving unit that receives a part of the light transmitted through the spatial light modulation element array and converts it into an electric signal, a calculation unit that calculates a weight based on the electric signal, and With the weight setting element array including a plurality of weight setting elements having the spatial light modulator that sets the calculated weight as the light transmittance, the weight can be updated simultaneously in all cells based on the error information due to the light.

【0014】[0014]

【実施例】図1は本発明の光ニューラルネットワークシ
ステムの一実施例を示すもので、図中、5は入力データ
を設定する空間光変調素子アレイ、6は出力データと教
師データとの誤差を設定する空間光変調素子アレイ、7
は重み設定素子アレイ、8は演算結果を得る受光素子ア
レイ、9は出力データと教師データとの誤差を求める外
部の演算回路、例えばコンピュータである。
FIG. 1 shows an embodiment of an optical neural network system of the present invention. In the figure, 5 is a spatial light modulator array for setting input data, and 6 is an error between output data and teacher data. Spatial light modulator array to be set, 7
Is a weight setting element array, 8 is a light receiving element array for obtaining a calculation result, and 9 is an external calculation circuit for obtaining an error between output data and teacher data, for example, a computer.

【0015】空間光変調素子アレイ5は入力光に対して
縦方向に均一な光変調ができる複数のストライプ状の空
間光変調素子の集合体であり、また、空間光変調素子ア
レイ6は入力光に対して横方向に均一な光変調ができる
複数のストライプ状の空間光変調素子の集合体である。
受光素子アレイ8は横1列の光を同時に受光し加算して
出力する複数のストライプ状の光電変換素子からなって
いる。
The spatial light modulator array 5 is an assembly of a plurality of stripe-shaped spatial light modulators capable of uniformly modulating the input light in the vertical direction, and the spatial light modulator array 6 is an input light. On the other hand, it is an assembly of a plurality of stripe-shaped spatial light modulators capable of uniform light modulation in the lateral direction.
The light-receiving element array 8 is composed of a plurality of stripe-shaped photoelectric conversion elements that simultaneously receive light in one horizontal row, add it, and output it.

【0016】重み設定素子アレイ7は、空間光変調素子
アレイ5,6を透過した光の一部を受光し電気信号に変
換する受光部と、該電気信号に基いて重みを計算する計
算部と、該計算した重みを光の透過率として設定する空
間光変調部とを備えた複数の重み設定素子からなってい
る。
The weight setting element array 7 includes a light receiving section for receiving a part of the light transmitted through the spatial light modulation element arrays 5, 6 and converting it into an electric signal, and a calculation section for calculating a weight based on the electric signal. , A plurality of weight setting elements having a spatial light modulator that sets the calculated weight as a light transmittance.

【0017】次に、本システムをニューラルネットワー
クにおける学習方法の一つであるバックプロパゲーショ
ンによる学習に適用した場合の動作について説明する。
説明に当って、光源(図示せず)の光量は「1」に正規
化されているものとする。
Next, the operation when the present system is applied to learning by backpropagation, which is one of learning methods in a neural network, will be described.
In the description, it is assumed that the light amount of the light source (not shown) is normalized to "1".

【0018】図1に示す如く、入力データにOi 、これ
に対する教師データとしてRj が設定されていると仮定
する。また、重み設定素子アレイ7の位置(i,j)の
セルには初期値としてWijなる重みが設定されているも
のとする。
As shown in FIG. 1, it is assumed that O i is set as input data and R j is set as teacher data for the input data. Further, it is assumed that the cell at the position (i, j) of the weight setting element array 7 has a weight W ij set as an initial value.

【0019】バックプロパゲーションによる学習処理
は、(1) ニューラルネットワーク演算、(2) 演算結果に
対する誤差の逆伝搬、(3) 誤差の値を基にした新しい重
みの生成、の3つの処理に分けられる。
The learning process by back propagation is divided into three processes: (1) neural network calculation, (2) back propagation of error with respect to calculation result, and (3) generation of new weight based on error value. Be done.

【0020】(1) ニューラルネットワーク演算 空間光変調素子アレイ5には入力データOi 、空間光変
調素子アレイ6には誤差伝搬のタイミングではないので
「1」が設定される。従って、受光素子アレイ8のj番
目のセルからは、 なる値が出力される。
(1) Neural network operation: Input data O i is input to the spatial light modulation element array 5, and "1" is set to the spatial light modulation element array 6 because it is not the timing of error propagation. Therefore, from the j-th cell of the light-receiving element array 8, Is output.

【0021】(2) 誤差の逆伝搬 誤差δj は、コンピュータ9により δj =Ij −Rj ……(4) で計算され、その結果が空間光変調素子アレイ6に設定
される。この結果、重み設定素子アレイ7の位置(i,
j)のセルで受光される値はOi δj となる。
(2) Error Back Propagation The error δ j is calculated by the computer 9 by δ j = I j -R j (4) and the result is set in the spatial light modulator array 6. As a result, the position (i,
The value received by the cell of j) is O i δ j .

【0022】(3) 重みの更新 重み設定素子アレイ7の位置(i,j)のセル内では受
光された値Oi δj を基にして、 Wij(new) =Wij+ηOi δj ……(4) (但し、ηは重みの更新の度合いを決める係数)により
新しい重みWij(new) の演算が行われる。この演算は重
み設定素子アレイ7の各セルでそれぞれが受光した値を
基にして同時に実行される。
(3) Update of weight In the cell at the position (i, j) of the weight setting element array 7, based on the value O i δ j received, W ij (new) = W ij + ηO i δ j .. (4) (where η is a coefficient that determines the degree of update of the weight), the new weight W ij (new) is calculated. This calculation is simultaneously executed based on the values received by the cells of the weight setting element array 7.

【0023】前述した演算を実行する重み設定素子アレ
イ7中の1つのセルの詳細を図3に示す。図中、71は
空間光変調素子アレイ6を透過した光を受光し電気信号
に変換する受光素子、72は乗算回路、73は加算回
路、74は計算した新しい重みを記憶する記憶回路、7
5は1つ前の重みを記憶する記憶回路、76は計算した
重みを光の透過率として設定する空間光変調素子であ
る。なお、1つのセルはその大部分の領域が空間光変調
素子76で占められ、受光素子71は例えば、その中央
部の一部分に設けられる。また、他の回路は各セル間も
しくはアレイ周辺に設けられる。
FIG. 3 shows the details of one cell in the weight setting element array 7 for executing the above-mentioned calculation. In the figure, 71 is a light receiving element that receives the light transmitted through the spatial light modulation element array 6 and converts it into an electric signal, 72 is a multiplication circuit, 73 is an addition circuit, 74 is a storage circuit for storing the calculated new weight, 7
Reference numeral 5 is a storage circuit for storing the previous weight, and reference numeral 76 is a spatial light modulator for setting the calculated weight as the light transmittance. Most of the area of one cell is occupied by the spatial light modulation element 76, and the light receiving element 71 is provided, for example, in a part of the central portion thereof. Other circuits are provided between cells or around the array.

【0024】ここまでは説明を簡単にするため、正の誤
差による重みの正方向への更新について述べたが、空間
光変調素子アレイ6の列を二分し、正負それぞれの誤差
を設定し、重み設定素子アレイ7にも正負の誤差信号を
独立して受光する受光部を設けることにより、前記構成
を適用した重みの加減算が可能となる。
Up to this point, in order to simplify the explanation, the updating of the weight in the positive direction by the positive error has been described. However, the columns of the spatial light modulator array 6 are divided into two, and the positive and negative errors are set and the weight is set. By providing the setting element array 7 with a light receiving section that independently receives positive and negative error signals, it is possible to add and subtract the weights to which the above configuration is applied.

【0025】このように前記実施例によれば、重みの更
新は全てのセルで同時に実行されるため、重み更新時間
はニューラルネットワークの規模によらず一定となり、
高速処理が実現できる。また、外部のコンピュータで実
施される誤差の演算数は重み設定素子アレイ7のセルの
数、即ちニューラルネットワークの規模の1/2乗とな
るため、コンピュータにおける計算時間も短時間で済
む。さらに、前記誤差の演算は非常に単純であることか
ら、各セル毎に専用の演算回路を設け並列処理を行うこ
とにより、さらに高速化を図ることも可能である。
As described above, according to the above-described embodiment, since the weight update is executed simultaneously in all cells, the weight update time becomes constant regardless of the scale of the neural network,
High-speed processing can be realized. Further, since the number of errors calculated by an external computer is the number of cells of the weight setting element array 7, that is, the power of 1/2 of the scale of the neural network, the calculation time in the computer can be short. Further, since the calculation of the error is very simple, it is possible to further increase the speed by providing a dedicated calculation circuit for each cell and performing parallel processing.

【0026】[0026]

【発明の効果】以上説明したように本発明によれば、入
力部からの複数の光に対して出力信号と教師信号との誤
差を設定する複数の空間光変調素子からなる空間光変調
素子アレイと、該空間光変調素子アレイを透過した光の
一部を受光し電気信号に変換する受光部、該電気信号に
基いて重みを計算する計算部及び該計算した重みを光の
透過率として設定する空間光変調部を備えた複数の重み
設定素子からなる重み設定素子アレイとを用いたため、
重みの更新に必要な誤差情報を光を用いて伝達でき、ま
た、該光による誤差情報に基いて重みの更新を全セルで
同時に実行できるので、従来のように外部で重みの演算
を行い、これを電気的に転送する必要がなくなり、学習
に要する時間を飛躍的に短縮させることが可能となる。
As described above, according to the present invention, a spatial light modulator array including a plurality of spatial light modulators for setting an error between an output signal and a teacher signal for a plurality of lights from an input section. And a light receiving unit that receives a part of the light transmitted through the spatial light modulation element array and converts it into an electric signal, a calculation unit that calculates a weight based on the electric signal, and the calculated weight is set as the light transmittance. Since a weight setting element array composed of a plurality of weight setting elements having a spatial light modulator to be used,
The error information necessary for updating the weight can be transmitted using light, and the weight can be updated simultaneously in all cells based on the error information due to the light, so that the weight is calculated externally as in the conventional case, There is no need to transfer this electrically, and it is possible to dramatically reduce the time required for learning.

【図面の簡単な説明】[Brief description of drawings]

【図1】本発明の光ニューラルネットワークシステムの
一実施例を示す構成図
FIG. 1 is a configuration diagram showing an embodiment of an optical neural network system of the present invention.

【図2】従来の光ニューラルネットワークシステムの一
例を示す構成図
FIG. 2 is a configuration diagram showing an example of a conventional optical neural network system.

【図3】重み設定素子アレイ中の1つのセルの詳細を示
す構成図
FIG. 3 is a configuration diagram showing details of one cell in the weight setting element array.

【符号の説明】[Explanation of symbols]

5,6…空間光変調素子アレイ、7…重み設定素子アレ
イ、8…受光素子アレイ、9…コンピュータ、71…受
光素子、72…乗算回路、73…加算回路、74,75
…記憶回路、76…空間光変調素子。
5, 6 ... Spatial light modulation element array, 7 ... Weight setting element array, 8 ... Light receiving element array, 9 ... Computer, 71 ... Light receiving element, 72 ... Multiplication circuit, 73 ... Addition circuit, 74, 75
... memory circuit, 76 ... spatial light modulator.

Claims (1)

【特許請求の範囲】[Claims] 【請求項1】 複数の入力信号に対応した複数の光を発
生する入力部と、該入力部からの複数の光に対して光の
透過率を変化させることにより所定の演算を行わせる中
間部と、該中間部を透過した複数の光を複数の出力信号
に変換する出力部とからなる光ニューラルネットワーク
システムにおいて、 前記中間部を、 入力部からの複数の光に対して出力信号と教師信号との
誤差を設定する複数の空間光変調素子からなる空間光変
調素子アレイと、 該空間光変調素子アレイを透過した光の一部を受光し電
気信号に変換する受光部、該電気信号に基いて重みを計
算する計算部及び該計算した重みを光の透過率として設
定する空間光変調部を備えた複数の重み設定素子からな
る重み設定素子アレイとを用いて構成したことを特徴と
する光ニューラルネットワークシステム。
1. An input section for generating a plurality of lights corresponding to a plurality of input signals, and an intermediate section for performing a predetermined calculation by changing the light transmittance of the plurality of lights from the input section. And an output unit for converting a plurality of lights transmitted through the intermediate unit into a plurality of output signals, the intermediate unit including an output signal and a teacher signal for the plurality of lights from the input unit. A spatial light modulation element array including a plurality of spatial light modulation elements for setting an error between the spatial light modulation element, and a light receiving section for receiving a part of light transmitted through the spatial light modulation element array and converting the light into an electric signal; And a weight setting element array including a plurality of weight setting elements including a spatial light modulator that sets the calculated weight as a light transmittance. Neural network Network system.
JP4036650A 1992-02-24 1992-02-24 Optical neural network system Expired - Fee Related JP2530404B2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP4036650A JP2530404B2 (en) 1992-02-24 1992-02-24 Optical neural network system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP4036650A JP2530404B2 (en) 1992-02-24 1992-02-24 Optical neural network system

Publications (2)

Publication Number Publication Date
JPH05233859A true JPH05233859A (en) 1993-09-10
JP2530404B2 JP2530404B2 (en) 1996-09-04

Family

ID=12475736

Family Applications (1)

Application Number Title Priority Date Filing Date
JP4036650A Expired - Fee Related JP2530404B2 (en) 1992-02-24 1992-02-24 Optical neural network system

Country Status (1)

Country Link
JP (1) JP2530404B2 (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11209594B2 (en) 2018-09-07 2021-12-28 Corning Incorporated Cable with overcoated non-coplanar groups of fusion spliced optical fibers, and fabrication method
US11360265B2 (en) 2019-07-31 2022-06-14 Corning Research & Development Corporation Fiber optic cable assembly with overlapping bundled strength members, and fabrication method and apparatus
US11561344B2 (en) 2017-03-21 2023-01-24 Corning Research & Development Corporation Fiber optic cable assembly with thermoplastically overcoated fusion splice, and related method and apparatus
US11808983B2 (en) 2020-11-24 2023-11-07 Corning Research & Development Corporation Multi-fiber splice protector with compact splice-on furcation housing
US11867947B2 (en) 2021-04-30 2024-01-09 Corning Research & Development Corporation Cable assembly having routable splice protectors
US11886009B2 (en) 2020-10-01 2024-01-30 Corning Research & Development Corporation Coating fusion spliced optical fibers and subsequent processing methods thereof

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11561344B2 (en) 2017-03-21 2023-01-24 Corning Research & Development Corporation Fiber optic cable assembly with thermoplastically overcoated fusion splice, and related method and apparatus
US11209594B2 (en) 2018-09-07 2021-12-28 Corning Incorporated Cable with overcoated non-coplanar groups of fusion spliced optical fibers, and fabrication method
US11360265B2 (en) 2019-07-31 2022-06-14 Corning Research & Development Corporation Fiber optic cable assembly with overlapping bundled strength members, and fabrication method and apparatus
US11774677B2 (en) 2019-07-31 2023-10-03 Corning Research & Development Corporation Fiber optic cable assembly with overlapping bundled strength members, and fabrication method and apparatus
US11886009B2 (en) 2020-10-01 2024-01-30 Corning Research & Development Corporation Coating fusion spliced optical fibers and subsequent processing methods thereof
US11808983B2 (en) 2020-11-24 2023-11-07 Corning Research & Development Corporation Multi-fiber splice protector with compact splice-on furcation housing
US11867947B2 (en) 2021-04-30 2024-01-09 Corning Research & Development Corporation Cable assembly having routable splice protectors

Also Published As

Publication number Publication date
JP2530404B2 (en) 1996-09-04

Similar Documents

Publication Publication Date Title
CN109784486B (en) Optical neural network processor and training method thereof
Li et al. Class-specific differential detection in diffractive optical neural networks improves inference accuracy
CN109784485B (en) Optical neural network processor and calculation method thereof
Kosko Adaptive bidirectional associative memories
CN109376855B (en) Optical neuron structure and neural network processing system comprising same
TWI819368B (en) Optoelectronic computing system
US20220164634A1 (en) Optical diffractive processing unit
JPH0643957A (en) Optical information processor
CN111723337B (en) Photon tensor core integrated circuit architecture for neural network training and neural network training method thereof
US20240078421A1 (en) Two-dimensional photonic convolutional acceleration system and device for convolutional neural network
US20210264241A1 (en) Optical multiply and accumulate unit
CN115545173A (en) Optical modulation neuron for signal processing and all-optical diffraction neural network method
CN114970836B (en) Reservoir neural network implementation method and system, electronic device and storage medium
US20210294874A1 (en) Quantization method based on hardware of in-memory computing and system thereof
JP2530404B2 (en) Optical neural network system
CN111898316A (en) Construction method and application of super-surface structure design model
US20210287078A1 (en) Artificial Neural Network Optical Hardware Accelerator
Song et al. A hybrid-integrated photonic spiking neural network framework based on an MZI array and VCSELs-SA
US5220642A (en) Optical neurocomputer with dynamic weight matrix
US11527059B2 (en) Reservoir computing
JPH0490015A (en) Optical neurocomputer
Simpkins Design, modeling, and simulation of a compact optoelectronic neural coprocessor
Héroux et al. High Density Multi-Chip Module for Photonic Reservoir Computing
Kashefi Rapidly training device for fiber-optic neural network
Antonik et al. Chaotic time series prediction using a photonic reservoir computer with output feedback

Legal Events

Date Code Title Description
LAPS Cancellation because of no payment of annual fees