CN114781602B - Deep convolution neural network system based on laser array and control method - Google Patents
Deep convolution neural network system based on laser array and control method Download PDFInfo
- Publication number
- CN114781602B CN114781602B CN202210361468.8A CN202210361468A CN114781602B CN 114781602 B CN114781602 B CN 114781602B CN 202210361468 A CN202210361468 A CN 202210361468A CN 114781602 B CN114781602 B CN 114781602B
- Authority
- CN
- China
- Prior art keywords
- output
- signal
- optical
- input
- neural network
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/067—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means
- G06N3/0675—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means using electro-optical, acousto-optical or opto-electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Neurology (AREA)
- Semiconductor Lasers (AREA)
- Optical Communication System (AREA)
Abstract
The invention relates to a depth convolution neural network system based on a laser array, wherein a depth convolution layer comprises a plurality of semiconductor lasers, a plurality of optical couplers and a plurality of feedback loops, which are transversely coupled; an input optical signal is injected into a first semiconductor laser, and a signal output by an nth semiconductor laser is divided into two paths through an nth optical coupler: one path of signal returns through the nth feedback loop and is used as an input signal of the (n + 1) th semiconductor laser; the other path of signal is input to an output layer; the output layer comprises an adder and a programmable gate control array; and the other paths of signals output by the optical couplers are superposed in the adder to form output information, and the programmable gate control array trains the output information to obtain the weight of the output layer. The plurality of semiconductor lasers coupled transversely are arranged, so that the training of interlayer connection can be avoided; only the output of the output layer needs to be trained, so that the dependence on the algorithm is greatly reduced, and the difficulty in hardware implementation is also obviously reduced.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a deep convolutional neural network system based on a laser array and a control method.
Background
Artificial neural networks have become a current subversive computational concept. These systems have had great success in a number of challenging tasks when multiple network layers are cascaded. In such deep neural networks, different layers are used to emphasize particular aspects of the output information, while the input to each network layer is also used as the output signal for the next layer. The continuous hierarchical structure has important significance for improving the computing performance. Deep Convolutional Neural Networks (CNNs) have achieved significant success in various areas over the last decade, including static image recognition.
In a deep Convolutional Neural Network (CNN), each layer convolves its input with a spatial filter. By increasing the width and step size of the filter, the deeper layers focus on more general features, while local features are emphasized in the previous layers.
However, the currently common deep convolutional neural network needs to train inter-layer connections, the requirements on algorithms and hardware are high, and it is difficult to directly map a network topology onto a physical substrate subsequently, which results in a slower information processing rate and also has a significant challenge on hardware integration.
Disclosure of Invention
The invention provides a depth convolution neural network system based on a laser array and a control method, wherein a depth convolution layer of the depth convolution layer comprises a plurality of semiconductor lasers which are transversely coupled, and interlayer coupling of the transverse coupling lasers is instantaneous, so that training of interlayer connection can be avoided; and only the output of an output layer needs to be trained, so that the dependence on the algorithm is greatly reduced, the difficulty is obviously reduced in hardware implementation, the network topology structure is easier to be directly mapped onto a physical substrate, and the convolutional neural network has higher information processing rate compared with the traditional convolutional neural network.
In order to solve the technical problem, the invention provides a depth convolution neural network system based on a laser array, which comprises a depth convolution layer and an output layer; the deep convolutional layer comprises a plurality of semiconductor lasers which are coupled transversely, a plurality of optical couplers and a plurality of feedback loops; injecting an input optical signal into a first semiconductor laser; the output signal of the nth semiconductor laser is divided into two paths by the nth optical coupler: one path of signal returns to the nth semiconductor laser through the nth feedback loop, and the signal returned to the nth semiconductor laser is used as an input signal of the (n + 1) th semiconductor laser; injecting the other signal to an output layer; wherein n is more than or equal to 1; the output layer comprises an adder and a programmable gating array connected with the adder; and the other path of signals output by each optical coupler are superposed in the adder to form output information, and the programmable gate control array trains the output information to obtain the weight of an output layer.
Preferably, the feedback loop comprises: the input end of the delay optical fiber is connected with the output end of the optical coupler; the delay optical fiber receives the two paths of signals output by the optical coupler; a variable optical attenuator having an input end connected to an output end of the delay optical fiber; the variable optical attenuator adjusts the signal output by the delay optical fiber; the input end of the light polarization controller is connected with the output end of the variable optical attenuator, and the output end of the light polarization controller is connected with the input end of the optical coupler; the light polarization controller adjusts the polarization state of the output signal of the variable optical attenuator, and the signal output by the light polarization controller is fed back to the optical coupler.
Preferably, the length of the delay fiber in each of said feedback loops is equal.
Preferably, the feedback loop further comprises an optical filter disposed between the delay optical fiber and the variable optical attenuator.
Preferably, the output layer further comprises a plurality of optical isolators and a plurality of photodetectors; and the other path of signal output by the nth semiconductor laser is injected into the nth optical isolator, the output end of the nth optical isolator is connected with the input end of the nth photodetector, and the output end of each photodetector is connected with the input end of the adder.
Preferably, the deep convolutional neural network system based on the laser array is characterized by further comprising an input layer, wherein the input layer comprises a driving laser, an arbitrary waveform generator and a modulator; the driving laser is connected with the modulator, and a signal output by the driving laser is injected into the modulator; the arbitrary waveform generator is connected with the modulator, and the arbitrary waveform generator and the modulator are used for modulating the signal output by the driving laser to obtain an input optical signal.
Preferably, the modulator is a phase modulator or an intensity modulator.
Preferably, the coupling mode of the plurality of semiconductor lasers is transient coupling.
Preferably, a ridge regression algorithm and a linear regression algorithm are included in the programmable gated array; the ridge regression algorithm and the linear regression algorithm are used for training the information output by the adder to obtain the weight of the output layer.
The control method of the deep convolutional neural network system based on the laser array is used for processing an optical signal by utilizing the deep convolutional neural network system based on the laser array, and is characterized by comprising the following steps of: acquiring input information and modulating the input information to form an input optical signal; performing convolution processing on the input optical signal, wherein: inputting the input optical signal into a first semiconductor laser of the deep convolution layer, and dividing a signal output from an nth semiconductor laser into two paths: controlling the first path of signal to return and serve as an input signal of the (n + 1) th semiconductor laser; controlling the other path of signal output; and superposing the other paths of signals to form output information, and training the output information to obtain an output weight.
Compared with the prior art, the technical scheme of the invention has the following advantages:
1. the deep convolutional layer of the present invention comprises a plurality of laterally coupled semiconductor lasers, in a feedback loop capable of adding light to act as a convolutional kernel: on one hand, the interlayer coupling of the semiconductor lasers is instantaneous and constant in time, so that the training of interlayer connection can be avoided, and the difficulty of hardware integration is reduced; on the other hand, the requirements for algorithms and hardware are low, the network topology is easier to map directly onto a physical substrate subsequently, and compared with the traditional convolutional neural network, the convolutional neural network has a higher information processing rate.
2. The output layer of the invention can obtain the response state of the deep convolutional layer, the output layer comprises the adder and the programmable gate control array, only the output of the output layer needs to be trained, the dependence on the algorithm is greatly reduced, and the difficulty is also obviously reduced in the aspect of hardware implementation.
3. Compared with the traditional convolutional neural network, the convolutional neural network has the advantages of simpler structure, low cost, friendly hardware and easy realization.
Drawings
In order that the present disclosure may be more readily and clearly understood, reference is now made to the following detailed description of the present disclosure taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a schematic diagram of a deep convolutional neural network according to the present invention;
FIG. 2 is a graph of predicted targets and results for the deep convolutional neural network processing time series prediction task of the present invention;
fig. 3 is a schematic diagram of an implementation principle of the deep convolutional neural network of the present invention.
The specification reference numbers indicate: 100-input layer, 200-depth convolution layer, 300-output layer;
11-drive laser, 12-arbitrary waveform generator, 13-modulator; 21-a semiconductor laser array, 22-an optical circulator, 23-an optical polarization controller, 24-a variable optical attenuator, 25-an optical filter, 26-a delay optical fiber and 27-an optical coupler; 31-optical isolator, 32-photodetector, 33-adder, 34-programmable gate array.
Detailed Description
The present invention is further described below in conjunction with the following figures and specific examples so that those skilled in the art may better understand the present invention and practice it, but the examples are not intended to limit the present invention.
The invention discloses a deep convolution neural network system based on a laser array, which is shown in figure 1 and comprises the following components: input layer 100, depth convolution layer 200, and output layer 300.
An input layer 100 comprising a drive laser 11, an arbitrary waveform generator 12 and a modulator 13. The driving laser 11 and the arbitrary waveform generator 12 are both connected to the modulator 13, a signal output by the driving laser 11 is injected into the modulator 13 for modulation, the arbitrary waveform generator 12 is configured to generate mask units, and the mask units modulate the signal input into the modulator 13 according to different mask signals, so as to obtain an input optical signal.
Preferably, the modulator 13 may be a phase modulator or an intensity modulator, or may be directly modulated by the output of the arbitrary waveform generator 12 without using the modulator 13, and the present invention is not limited thereto.
The deep convolutional layer 200 includes a plurality of laterally coupled semiconductor lasers, a plurality of optical circulators 22, an optical coupler 27, and a plurality of feedback loops. The plurality of semiconductor lasers form a semiconductor laser array 21, and the optical circulators 22, the optical couplers 27 and the feedback loops are all arranged in one-to-one correspondence with the semiconductor lasers.
The interlayer coupling mode of the plurality of semiconductor lasers is transient coupling and is constant in time, so that the training of interlayer connection can be avoided, and the difficulty of hardware integration is reduced.
The input optical signal is injected into a first semiconductor laser in the semiconductor laser array 21, and an optical signal output by an nth semiconductor laser is divided into two paths by an nth optical circulator 22 and an nth optical coupler 27 in sequence: the first path of signal returns to the nth semiconductor laser through the nth feedback loop, and the signal returned to the nth semiconductor laser is used as an input signal of the (n + 1) th semiconductor laser; the other signal is input to the output layer 300, wherein n is more than or equal to 1.
The deep convolutional layer 200 of the present invention is added with a feedback loop, so as to realize the construction of a recurrent neural network, and make the deep convolutional layer 200 have a certain memory capacity. Preferably, each feedback loop has the same structure, and the feedback loops include:
(1) And the input end of the delay fiber 26 is connected with the output end of the optical coupler 27, and the delay fiber 26 is used for receiving two paths of signals output by the optical coupler 27. (2) And an optical filter 25, wherein an input end of the optical filter 25 is connected to an output end of the delay fiber 26, and the optical filter 25 is used for filtering a signal output by the delay fiber 26. (3) And a variable optical attenuator 24 connected to the optical filter 25 for adjusting the feedback intensity of the output signal of the optical filter 25. (4) And the optical polarization controller 23 has an input end connected to the variable optical attenuator 24 and an output end connected to an input end of the corresponding optical coupler 27, the optical polarization controller 23 is configured to control a polarization state of a signal output from the variable optical attenuator 24, and the signal output from the optical polarization controller 23 is fed back to the optical coupler 27.
The optical filter 25 also adjusts the dynamic characteristics of the semiconductor laser, realizes different reservoir states, serves as a convolution kernel, can realize important features in generalization, and highlights common features, thereby improving the performance of the deep convolutional layer 200.
An output layer 300 that includes a plurality of optical isolators 31, a plurality of photodetectors 32, an adder 33, and a programmable gate array.
Specifically, the other signal output by the nth optical coupler 27 is injected into the nth optical isolator 31, the output end of the nth optical isolator 31 is connected to the input end of the nth photodetector 32, the nth photodetector 32 converts the signal output by the nth optical isolator 31 into an electrical signal, and the output end of each photodetector 32 is connected to the input end of the adder 33. The electrical signals output by each photodetector 32 are input to an adder 33 and are superimposed in the adder 33 to form output information, which is calculated in a programmable gate array 34 to obtain the weight of the output layer 300.
The adder 33 is connected to the programmable gate array 34, and the programmable gate array 34 includes a ridge regression algorithm and a linear regression algorithm, so as to train the information output from the adder 33. Only need train in output layer 300, greatly reduced the reliance to the algorithm, also obviously reduced the degree of difficulty in hardware implementation simultaneously, be easier directly to map network topology structure on the physical base plate, compare traditional convolution neural network have faster information processing rate.
Further preferably, the feedback strength of each feedback loop may be the same or different. The length of the delay fiber 26 in each feedback loop is equal, and the delay fiber 26 is preferably used to transmit signals due to the advantages of the delay fiber 26, such as low dispersion, large transmission bandwidth, and low attenuation. In addition, a spatial light transmission signal may also be used, and in practical applications, a suitable transmission mode is selected according to the situation, which is not limited in the present invention.
Further, simulating the deep convolutional neural network system based on the laser array according to numerical simulation, and establishing a rate equation as follows:
wherein 1, m and n respectively represent the 1 st semiconductor laser, the m (m is more than or equal to 2 and less than or equal to n-1) th semiconductor laser and the last semiconductor laser.
Represents the electric field of the semiconductor laser, and>
represents the carrier concentration of a semiconductor laser and
indicating its threshold. Γ represents the light field confinement factor, c represents the speed of light, n g Denotes the group index, a diff Representing a difference factor, alpha H The line width is the line width enhancement factor->
Representing the cavity resonance frequency, P the pumping rate, γ N the carrier decay rate, N the reflectivity, and the threshold gain g th Denoted Γ g th ＝n g /(cτ p )，τ p Representing the photon lifetime. Injected electric field->
I d Representing the steady state intensity of the drive laser 11 and u (t) representing the input optical signal.
Referring to FIG. 2, a diagram of the predicted target and the predicted result of the deep convolutional neural network processing Santa Fe time series prediction task is shown. It can be seen from fig. 2 that the time series prediction result is close to the target, the NMSE is about 0.0008 by adjusting the filtering bandwidth of the optical filter 25, and the deep convolutional layer 200 can more efficiently implement information processing, thereby implementing a deep convolutional neural network system with simple structure, low cost, easy integration, and high information processing speed.
The invention further provides a deep convolutional neural network system control method based on the laser array, so as to process optical signals.
Referring to fig. 3, the method for controlling the deep convolutional neural network system includes the following steps:
step one, acquiring input information and modulating the input information to form an input optical signal u (t).
Step two, convolution processing is carried out on the input optical signal u (t), wherein: an input optical signal u (t) is input into the first semiconductor laser of the deep convolutional layer 200, and a signal output from the nth semiconductor laser is divided into two paths: and controlling the first path of signal to return and serve as an input signal of the (n + 1) th semiconductor laser, and controlling the second path of signal to be output.
And step three, superposing all the other paths of signals to form output information, and training the output information to obtain an output weight.
The nth optical coupler 27 in the deep convolutional layer 200 divides the signal output by the nth semiconductor laser into two paths, the other path of signal output by each optical coupler 27 forms an electrical signal Xn (t) through the photodetector 32, and the outputs of the photodetectors 32 are superimposed in the adder 33 to form output information X = X1 (t) + X2 (t) + \ 8230 \ 8230; (Xn (t)). The programmable gating array 34 in the output layer 300 trains the output information to derive the weights of the output layer 300.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications of the invention may be made without departing from the spirit or scope of the invention.
Claims (10)
1. The deep convolution neural network system based on the laser array is characterized by comprising:
a depth convolution layer and an output layer;
the deep convolutional layer comprises a plurality of semiconductor lasers which are coupled transversely, a plurality of optical couplers and a plurality of feedback loops; the optical couplers and the feedback loops are in one-to-one correspondence with the semiconductor lasers;
injecting an input optical signal into a first semiconductor laser; the signal output by the nth semiconductor laser is divided into two paths by the nth optical coupler: one path of signal returns through the nth feedback loop and is used as an input signal of the (n + 1) th semiconductor laser; the other path of signal is input to the output layer; wherein n is more than or equal to 1;
the output layer comprises an adder and a programmable gating array connected with the adder; and the other path of signals output by each optical coupler are superposed in the adder to form output information, and the programmable gate control array trains the output information to obtain the weight of an output layer.
2. The laser array based deep convolutional neural network system of claim 1, wherein the feedback loop comprises:
the input end of the delay optical fiber is connected with the output end of the optical coupler; the delay optical fiber receives two paths of signals output by the optical coupler;
a variable optical attenuator having an input end connected to an output end of the delay optical fiber; the variable optical attenuator adjusts the signal output by the delay optical fiber;
the input end of the light polarization controller is connected with the output end of the variable optical attenuator, and the output end of the light polarization controller is connected with the input end of the optical coupler; the light polarization controller adjusts the polarization state of the output signal of the variable optical attenuator, and the signal output by the light polarization controller is fed back to the optical coupler.
3. The laser array based deep convolutional neural network system of claim 2, wherein the length of the delay fiber in each of the feedback loops is equal.
4. The laser array based deep convolutional neural network system of claim 2, wherein the feedback loop further comprises an optical filter disposed between the delay fiber and the variable optical attenuator.
5. The laser array based deep convolutional neural network system of claim 1, wherein the output layer further comprises a plurality of optical isolators and a plurality of photodetectors;
and the other path of signal output by the nth optical coupler is injected into the nth optical isolator, the output end of the nth optical isolator is connected with the input end of the nth photodetector, and the output end of each photodetector is connected with the input end of the adder.
6. The laser array based deep convolutional neural network system of claim 1, further comprising an input layer comprising a drive laser, an arbitrary waveform generator, and a modulator;
the driving laser is connected with the modulator, and a signal output by the driving laser is injected into the modulator; the arbitrary waveform generator is connected with the modulator, and the arbitrary waveform generator and the modulator are used for modulating the signal output by the driving laser to obtain the input optical signal.
7. The laser array based deep convolutional neural network system of claim 6, wherein the modulator is a phase modulator or an intensity modulator.
8. The deep convolutional neural network system as claimed in claim 1, wherein the coupling mode of the plurality of semiconductor lasers is transient coupling.
9. The laser array based deep convolutional neural network system of claim 1, comprising a ridge regression algorithm and a linear regression algorithm within the programmable gated array; the ridge regression algorithm and the linear regression algorithm are used for training the information output by the adder to obtain the weight of the output layer.
10. The control method of the deep convolutional neural network system based on the laser array, which utilizes the deep convolutional neural network system based on the laser array as claimed in any one of claims 1 to 9 to process the optical signal, and is characterized by comprising the following steps:
acquiring input information and modulating the input information to form an input optical signal;
performing convolution processing on the input optical signal, wherein: inputting the input optical signal into a first semiconductor laser of the deep convolution layer, and dividing a signal output from an nth semiconductor laser into two paths: controlling one path of signal to return and serve as an input signal of the (n + 1) th semiconductor laser, and controlling the other path of signal to be output;
and superposing the other paths of signals to form output information, and training the output information to obtain an output weight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210361468.8A CN114781602B (en) | 2022-04-07 | 2022-04-07 | Deep convolution neural network system based on laser array and control method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210361468.8A CN114781602B (en) | 2022-04-07 | 2022-04-07 | Deep convolution neural network system based on laser array and control method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114781602A CN114781602A (en) | 2022-07-22 |
| CN114781602B true CN114781602B (en) | 2023-03-24 |
Family
ID=82427894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210361468.8A Active CN114781602B (en) | 2022-04-07 | 2022-04-07 | Deep convolution neural network system based on laser array and control method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114781602B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115169542B (en) * | 2022-09-02 | 2022-11-22 | 之江实验室 | Two-dimensional photon convolution acceleration system and device for convolution neural network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113792870A (en) * | 2021-10-29 | 2021-12-14 | 广西大学 | All-optical control optical neural network system based on acousto-optic technology |
| CN114037070A (en) * | 2022-01-07 | 2022-02-11 | 苏州浪潮智能科技有限公司 | Optical signal processing method, photonic neural network chip and design method thereof |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488565A (en) * | 2015-11-17 | 2016-04-13 | 中国科学院计算技术研究所 | Calculation apparatus and method for accelerator chip accelerating deep neural network algorithm |
| CN109444912B (en) * | 2018-10-31 | 2020-08-04 | 电子科技大学 | Driving environment sensing system and method based on cooperative control and deep learning |
| CN113759345A (en) * | 2021-09-07 | 2021-12-07 | 苏州大学 | Laser radar based on polarization modulation light injection laser and regulation and control method thereof |
-
2022
- 2022-04-07 CN CN202210361468.8A patent/CN114781602B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113792870A (en) * | 2021-10-29 | 2021-12-14 | 广西大学 | All-optical control optical neural network system based on acousto-optic technology |
| CN114037070A (en) * | 2022-01-07 | 2022-02-11 | 苏州浪潮智能科技有限公司 | Optical signal processing method, photonic neural network chip and design method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114781602A (en) | 2022-07-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sawyer | Speed-up of neutrino transformations in a supernova environment | |
| CN114781602B (en) | Deep convolution neural network system based on laser array and control method | |
| CN111753977B (en) | Optical neural network convolution layer chip, convolution calculation method and electronic equipment | |
| Zhang et al. | The winner-take-all mechanism for all-optical systems of pattern recognition and max-pooling operation | |
| CN111338423A (en) | Optical device | |
| US20220414442A1 (en) | Optical computing apparatus and system, and computing method | |
| CN112101540B (en) | Optical neural network chip and calculation method thereof | |
| CN110247299A (en) | A kind of parallel double light feedback semiconductor laser reserve pool computing systems | |
| CN114325932B (en) | On-chip integrated all-optical neural network optical computing chip | |
| CN114519403B (en) | Optical diagram neural classification network and method based on-chip diffraction neural network | |
| Zhu et al. | Neural piecewise-constant delay differential equations | |
| JP6701247B2 (en) | Optical signal processor | |
| Xiang et al. | Photonic integrated neuro-synaptic core for convolutional spiking neural network | |
| CN112560288A (en) | Reserve pool computing device based on optical pump spin VCSEL | |
| CN111680796A (en) | On-chip photonic neural network device based on cascade optical coupler, chip and application method thereof | |
| CN114757344B (en) | System and method for simulating biological neuron dynamics to realize logic operation | |
| Zhao et al. | Experimental implementation of spike-based neuromorphic XOR operation based on polarization-mode competition in a single VCSOA | |
| CN114925838A (en) | Raman transition method and device, and method and device for constructing quantum logic gate | |
| CN115145535B (en) | Depth reserve pool computing system and method based on spin vertical cavity surface emitting laser | |
| CN117669649B (en) | Depth delay reserve pool computing system | |
| CN108919589B (en) | Reconfigurable dynamic all-optical chaotic logic gate capable of storing in delayed mode | |
| CN112232487A (en) | Optical neural network chip and calculation method thereof | |
| Feng et al. | Spike-based all-optical fast XOR operation in a VCSEL under dual-polarization optical injections | |
| CN112350781B (en) | Low-voltage-driven high-speed photon quantum state preparation device and method | |
| Zhang et al. | BP-based supervised learning algorithm for multilayer photonic spiking neural network and hardware implementation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |