Optical information processing system and training method and apparatus

Abstract

The present disclosure provides an optical information processing system and a training method and device, and relates to the fields of optical computing, artificial intelligence and optoelectronic hybrid computing. The system can include an encoder, a mode selection transmission module and a decoder; the encoder is configured to modulate an input light field to generate a continuous complex light field; the mode selection transmission module is configured to perform discrete mapping of the continuous complex light field in a low-dimensional latent space, and obtain an output light field through nonlinear evolution; and the decoder is configured to generate a task processing result corresponding to the input light field according to the output light field. The scheme disclosed in the present disclosure can improve the nonlinear representation capability of the system.

Description

Technical Field

[0001] This disclosure relates to fields such as optical computing, artificial intelligence, and optoelectronic hybrid computing, and in particular to optical information processing systems and training methods and apparatus. Background Technology

[0002] Studies have shown that current optical neural networks or optical generation models, under pure free space angular spectrum propagation or diffraction structures, can still be reduced to a cascade of linear operators, which significantly limits their expressive power. Summary of the Invention

[0003] This disclosure provides an optical information processing system, training method, and apparatus.

[0004] An optical information processing system includes: an encoder, a mode selection transmission module, and a decoder;

[0005] The encoder is used to modulate the input optical field to generate a continuous complex optical field; The mode selection transmission module is used to perform low-dimensional latent space discretization mapping on the continuous complex optical field and obtain the output optical field through nonlinear evolution. The decoder is used to generate the task processing result corresponding to the input light field based on the output light field.

[0006] A training method for an optical information processing system, comprising: Acquire training samples, which include the sample input light field and the real label; The sample input light field is input into an optical information processing system to obtain the output task processing result, which is then determined as the prediction label. The optical information processing system includes an encoder, a mode selection transmission module, and a decoder. The encoder is used to modulate the sample input light field to generate a continuous complex light field. The mode selection transmission module is used to discretize the continuous complex light field into a low-dimensional latent space and obtain the output light field through nonlinear evolution. The decoder is used to generate the prediction label based on the output light field. The loss value is determined based on the predicted label and the real label in the training sample, where the real label is the actual task processing result corresponding to the input light field of the sample. The configuration parameters of the optical information processing system are updated based on the loss value.

[0007] A training device for an optical information processing system includes: a sample acquisition module and a system training module; The sample acquisition module is used to acquire training samples, which include the sample input light field and the real label. The system training module is used to input the sample input light field into the optical information processing system, obtain the output task processing result, and determine it as the prediction label. The optical information processing system includes an encoder, a mode selection transmission module, and a decoder. The encoder is used to modulate the sample input light field to generate a continuous complex light field. The mode selection transmission module is used to discretize the continuous complex light field into a low-dimensional latent space and obtain the output light field through nonlinear evolution. The decoder is used to generate the prediction label based on the output light field. A loss value is determined based on the prediction label and the real label in the training samples. The real label is the actual task processing result corresponding to the sample input light field. The configuration parameters of the optical information processing system are updated based on the loss value.

[0008] An electronic device, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described above.

[0009] A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the methods described above.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a schematic diagram of the composition structure of the first embodiment of the optical information processing system described in this disclosure; Figure 2 This is a schematic diagram of the composition structure of the mode selection transmission module described in this disclosure; Figure 3 This is a schematic diagram of the composition structure of the second embodiment of the optical information processing system described in this disclosure; Figure 4 This is a schematic diagram of the composition structure of the third embodiment of the optical information processing system described in this disclosure; Figure 5 This is a flowchart of an embodiment of the training method for the optical information processing system described in this disclosure; Figure 6 This is a schematic diagram of the composition structure of an embodiment of the training device of the optical information processing system described in this disclosure; Figure 7 A schematic block diagram of an electronic device that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0012] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0013] Furthermore, it should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0014] Figure 1 This is a schematic diagram of the structural composition of the first embodiment 100 of the optical information processing system described in this disclosure. Figure 1 As shown, it includes: encoder 101, mode selection transmission module 102 and decoder 103.

[0015] Encoder 101 is used to modulate the input light field to generate a continuous complex light field.

[0016] The mode selection transmission module 102 is used to discretize the continuous complex optical field into a low-dimensional latent space and obtain the output optical field through nonlinear evolution.

[0017] Decoder 103 is used to generate the task processing result corresponding to the input light field based on the output light field.

[0018] Using the scheme described in the above system embodiment, the input light field can be encoded into a continuous complex light field by the encoder 101. The core relies on the mode selection transmission module 102 to realize the discretization mapping and nonlinear evolution of the optical latent space, and complete the nonlinear calculation at the optical level. Then, the decoder generates the task processing result corresponding to the input light field, thereby breaking the limitation of linear operator cascade, enabling the system to obtain complex nonlinear representation capabilities, fit high-order functions and complex patterns, and have hierarchical structure expression capabilities similar to deep neural networks or variational autoencoders (VAEs). As a result, the performance of cutting-edge artificial intelligence (AI) tasks such as generation, reasoning, and representation learning is significantly improved.

[0019] In this embodiment of the disclosure, a continuous complex optical field refers to an optical field distribution continuously described in the spatial and / or temporal domains by a complex-valued function. The amplitude of this complex-valued function characterizes the intensity distribution of the optical field, and the phase characterizes the wavefront shape of the optical field. Unlike a real optical field that only records intensity, a continuous complex optical field fully preserves information such as light interference and diffraction, and is the fundamental representation for optical coherence processing and latent space mapping.

[0020] A low-dimensional latent space refers to an abstract space whose dimension (degrees of freedom) is much smaller than that of the input light field state space. In this space, the complex information of the light field is compressed into a small number of key features, which facilitates subsequent processing. The dimension of the latent space is determined by the number of modes supported by the mode selection and transmission module 102.

[0021] Discretization refers to the process of assigning continuously changing light field states to a finite number of discrete modes in a low-dimensional latent space according to preset rules. Each mode corresponds to an independent dimension of the latent space and can be realized through at least one of the following physical forms: Channel: A physically separate transmission path (such as a waveguide array or a multi-core fiber).

[0022] Mode: Eigenstates of the optical field (such as spatial mode, polarization mode, wavelength mode, and time mode).

[0023] Category: Discrete response states of devices (such as on / off states of saturable absorbers, resonant / non-resonant states of resonant cavities, TE / TM states of polarization beam splitters, and crystalline / amorphous states of phase change materials), where TE refers to transverse electric waves and TM refers to transverse magnetic waves.

[0024] Discretization enables the stable representation and processing of optical field information with limited resources. Discretization mapping refers to projecting or allocating the high-dimensional continuous complex optical field output by encoder 101 into a low-dimensional latent space through mode selection and transmission module 102, and discretizing the optical field state into a finite number of distinguishable modes or channels in the latent space during this process. This mapping establishes a definite correspondence between the input continuous complex optical field and the discrete state in the latent space, which can be linear (e.g., through mode orthogonal projection) or nonlinear (e.g., through mode competition in a nonlinear medium), determined by the physical structure of mode selection and transmission module 102.

[0025] Nonlinear evolution refers to the phenomenon where, during propagation, the optical field is affected by the nonlinear response of the medium, and its state (including amplitude, phase, spectrum, spatial mode distribution, or polarization state) changes with propagation distance or time according to nonlinear differential equations, resulting in a nonlinear dependence between the output and input optical fields. In this embodiment, nonlinear evolution includes: intrinsic nonlinear evolution, which utilizes the inherent nonlinear optical effects of waveguide materials (including Kerr effect, saturable absorption, nonlinear mode coupling, parametric processes, etc.) to perform nonlinear transformations and inter-channel coupling on the signals of each channel; and programmable nonlinear evolution, which dynamically controls the evolution process through integrated active modulation mechanisms (such as electro-optic, thermo-optic, and acousto-optic effects). Nonlinear evolution occurs in media such as nonlinear crystals, highly nonlinear optical fibers, and semiconductor optical amplifiers, or in structures such as metasurfaces and saturable absorbers. The key role of nonlinear evolution is to enable the system to fit higher-order functions and complex modes, breaking the limitations of linear operator cascading.

[0026] In summary, the continuous complex optical field is the information carrier processed by the optical information processing system. Discretization mapping compresses it into a finite discrete mode in a low-dimensional latent space, while nonlinear evolution completes the key nonlinear transformation of information in this space, thereby enabling the system to obtain complex representation capabilities beyond linear operators.

[0027] In some embodiments of this disclosure, encoder 101 may include an optical encoder consisting of M layers of metasurfaces or diffractive optical elements, where M is a positive integer.

[0028] The specific value of M can be determined according to actual needs; for example, M=1 or M>1. When M>1, layer-by-layer, refined high-dimensional mapping of the input light field can be achieved, effectively improving the capacity and abstraction level of optical feature extraction, and enhancing the overall system's ability to represent and process complex tasks. In addition, the M-layer metasurface or diffractive optical element can be a designable metasurface or diffractive optical element.

[0029] Encoder 101 can receive an input optical field, such as an image optical field carrying predetermined information, a modulated coherent beam, or encoded data from a previous stage system. Furthermore, encoder 101 can perform physical domain feature extraction and high-dimensional transformation on the input optical field, thereby converting the input optical field into a continuous complex optical field. This can be achieved by precisely controlling the input optical field locally and subwavelength (e.g., changing the phase, amplitude, or polarization) to complete a complex forward transformation during light propagation, thus mapping the input optical field into a new, higher-dimensional, and more task-relevant continuous complex optical field. The continuous complex optical field already carries the information extracted through "optical feature extraction" and can be considered as a feature map of the optical domain.

[0030] In practical applications, encoder 101 can correspond to an encoder in deep learning, which can extract latent features from the input light field. The mode selection transmission module 102 can perform low-dimensional latent space discretization mapping on the continuous complex light field from encoder 101, and obtain the output light field through nonlinear evolution processing.

[0031] The mode selection transmission module 102 may include several different components. For example... Figure 2 As shown, Figure 2 This is a schematic diagram of the composition of the mode selection transmission module 102 described in this disclosure, which may include: an input coupling interface 1021, a waveguide constraint and nonlinear processing unit 1022, and an output coupling interface 1023.

[0032] An input coupling interface 1021 is used to couple the received continuous complex optical field to a waveguide constraint and nonlinear processing unit 1022. The waveguide constraint and nonlinear processing unit 1022 is used to discretize the continuous complex optical field and map it to multiple discrete modes in a low-dimensional latent space, and to apply nonlinear evolution to each discrete mode signal, wherein each discrete mode corresponds to an independent dimension of the latent space. An output coupling interface 1023 is used to synthesize the nonlinearly evolved signals into an output optical field. In some embodiments of this disclosure, the discrete modes are implemented through at least one of the following physical forms: physically separated transmission channels, eigenmodes of the optical field, and discrete response categories of optical devices.

[0033] Among them, physically separated transmission channels include, but are not limited to: few-mode fiber, multimode fiber, multi-core fiber, planar optical waveguide array, ridge waveguide array, channel waveguide array, photonic crystal waveguide, slit waveguide array, metamaterial waveguide array, and spatial channels of multi-planar optical converters. The intrinsic modes of the optical field include, but are not limited to: spatial modes (LP mode, TE / TM mode, HE / EH mode), orbital angular momentum modes (OAM modes carrying different topological charges l), polarization modes (linear polarization, circular polarization, cylindrical vector polarization), wavelength modes (discrete wavelength channels), and time modes (discrete time slots); among them, LP mode refers to linear polarized mode, HE mode refers to hybrid electric mode, EH mode refers to hybrid magnetic mode, and OAM mode refers to orbital angular momentum mode.

[0034] The discrete response categories of optical devices include, but are not limited to: the on / off state of a saturable absorber, the resonant / non-resonant state of an optical resonator, the orthogonal polarization output state of a polarization beam splitter, and the crystalline / amorphous state of a phase change material.

[0035] In a specific example, the mode selection transmission module 102 can be a physical computing unit composed of a waveguide array and an embedded nonlinear processing mechanism. Its core function is to discretize and map the input continuous complex optical field to multiple independent channels (i.e., discrete modes) in a low-dimensional latent space for parallel transmission, and to perform in-situ nonlinear calculations on it. Each waveguide channel (transmission channel, i.e., each independent waveguide) corresponds to a basis dimension in the low-dimensional latent space, and the light intensity and phase inside it are the scalar values ​​of that dimension. The final output is a characteristic optical field after nonlinear transformation and mixing.

[0036] The input coupling interface 1021 can efficiently couple the continuous complex optical field propagating in free space to the subsequent waveguide confinement and nonlinear processing unit 1022 by means of precise mode matching and medium transition design, while minimizing reflection loss and mode mismatch loss.

[0037] Typical implementations of the input coupling interface 1021 may include: direct end-face coupling, integrated mode converter, or discrete optical coupler.

[0038] The waveguide constraint and nonlinear processing unit 1022 is a core component of the mode selection transmission module 102, and mainly undertakes the following two key tasks: Waveguide constraint and latent space formation: Each independent waveguide channel serves as a physical carrier, corresponding to and constraining a basis dimension. All channels together constitute a hardware-implemented, discrete optical latent space.

[0039] Embedded nonlinear computing: Nonlinear processing mechanisms are integrated into or tightly coupled to waveguide arrays to apply controllable nonlinear optical transformations to discrete optical latent signals in transmission.

[0040] In some embodiments of this disclosure, the nonlinear evolution may include: nonlinear transformation and intermode coupling of the mode signal using the inherent nonlinear optical effects of the waveguide material, and / or programmable control of the nonlinear evolution through an integrated active modulation mechanism.

[0041] Specifically, nonlinear optical effects or integrated active modulation mechanisms can include at least one of the following: Intrinsic Kerr nonlinearity includes the Kerr effect, self-phase modulation, cross-phase modulation, and four-wave mixing.

[0042] Active controllable nonlinearity: The nonlinear evolution process is dynamically controlled through the electro-optic effect or the thermo-optic effect (where the electro-optic effect itself is a second-order nonlinearity and can independently generate nonlinear transformations; the thermo-optic effect is mainly used to control nonlinear conditions).

[0043] Material-enhanced nonlinearity: Enhanced nonlinear response or saturable absorption characteristics are introduced through rare earth ion doping, two-dimensional material coating (such as graphene, transition metal sulfides) or saturable absorbers.

[0044] Traditional single-mode fiber is essentially a strict spatial mode filter. No matter how complex the input optical field is, its output will degenerate into a pure fundamental Gaussian distribution. This process is achieved by filtering out all higher-order modes. Although this purifies the beam, it also loses all spatial structure information in the input optical field, making it only suitable for transmission and unsuitable for preserving and computing complex optical latent representations. The mode-selective transmission module 102 in the scheme described in this disclosure employs a multi-channel waveguide array. Each waveguide channel can support multiple modes (operating in few-mode or multi-mode states), or multiple independent spatial degrees of freedom can be transmitted in parallel through array design. This allows each waveguide channel in the waveguide array to carry an independent "fundamental mode" component, just like a traditional single-mode fiber, and to distribute and preserve the complex information of the input optical field across multiple waveguide channels. Simultaneously, through an embedded nonlinear processing mechanism, the signal undergoes nonlinear interactions within and between waveguide channels (through evanescent field coupling or four-wave mixing, etc.), dynamically mixing and transforming this information to achieve true optical domain computation (activation and mixing), thus completely overturning the traditional waveguide's role as merely an "information conduit."

[0045] Correspondingly, the mode selection transmission module 102 not only physicalizes and preserves the high-dimensional optical latent space through the waveguide array, but also realizes dynamic, parallel, and simulated calculation of the latent space through an embedded nonlinear processing mechanism. Thus, it is directly equivalent to the nonlinear layer of deep learning at the physical level, such as realizing the "nonlinear activation" and cross-mode "feature fusion" functions that are crucial in deep learning.

[0046] The mode selection transmission module 102 can employ optical devices based on at least one of the following waveguide structures: few-mode or multimode fiber arrays, multi-core fibers (which can be doped or coated with nonlinear materials); planar waveguide arrays (such as silicon-based optical waveguides, silicon nitride waveguides); photonic crystal waveguides (utilizing their strong field confinement and tunable nonlinearity), and other specific optical waveguide structures. The output coupling interface 1023 can convert the nonlinearly processed signal output by the waveguide confinement and nonlinear processing unit 1022 into an output optical field through recombination, merging, or other methods.

[0047] It can be seen that the input coupling interface 1021 and the output coupling interface 1023 can realize the efficient transfer of the optical field between different media. The input coupling interface 1021, the waveguide constraint and nonlinear processing unit 1022 and the output coupling interface 1023 work together to form a complete optical field processing chain. The three can be designed as discrete modules or monolithic integration according to actual needs.

[0048] The decoder 103 can generate the task processing result corresponding to the input light field based on the output light field from the output coupling interface 1023, that is, convert the output light field into the final task output.

[0049] In some embodiments of this disclosure, decoder 103 may include: an optical decoder composed of a p-layer metasurface or diffractive optical elements, where p is a positive integer and its specific value can be determined according to actual needs. Alternatively, decoder 103 may include: an optical-digital hybrid decoder composed of a photosensitive element and a neural network model.

[0050] Accordingly, in conjunction with the foregoing introduction, Figure 3 This is a schematic diagram of the structural composition of the second embodiment 300 of the optical information processing system described in this disclosure. Figure 4 This is a schematic diagram of the structural composition of the third embodiment 400 of the optical information processing system described in this disclosure. Wherein, Figure 3 Decoder 103 in the code can be an optical decoder. Figure 4 The decoder 103 in the middle can be an optical-digital hybrid decoder, that is, it consists of a photosensitive element and a neural network model.

[0051] The decoder 103 outputs the task processing result, which can refer to outputting an image, outputting a reconstructed signal, or completing an inference task, similar to a decoder in deep learning.

[0052] In an optical-digital hybrid decoder, a photosensitive element converts optical signals into electrical signals, which are then used by a neural network model to generate task processing results. The neural network model can be a lightweight digital decoder, such as a single-layer fully connected network, because most of the computation is performed in the optical domain, thus minimizing overhead.

[0053] In summary, the optical information processing system described in this disclosure is a hierarchical all-optical physical neural network. It achieves intelligent processing of the input light field through a cascaded process of optical encoding, nonlinear latent space evolution, and optical / digital decoding. It transfers the core nonlinear computation process of deep learning from the digital domain to the latent space evolution stage of optics, and utilizes the intrinsic nonlinearity of materials to achieve low-power, ultra-high-speed physical computation.

[0054] Furthermore, the optical information processing system described in this disclosure places the key nonlinear calculation process directly within the optical latent space, rather than postponing it to the final electrical detection end. This achieves a true "optical feedforward neural network" at the hardware level for the first time. Moreover, the encoding and evolution processes are entirely based on passive optical elements or low-power active optical structures, providing a practical physical implementation path for next-generation intelligent optical computing with high energy efficiency, low latency, and high parallelism.

[0055] In relation to the aforementioned optical information processing system, this disclosure also discloses corresponding training methods and apparatus.

[0056] Figure 5 This is a flowchart illustrating an embodiment of the training method for the optical information processing system described in this disclosure. Figure 5 As shown, the specific implementation methods are as follows.

[0057] In step 501, training samples are obtained, which include the sample input light field and the real label.

[0058] In step 502, the sample input light field is input into the optical information processing system to obtain the output task processing result, which is then determined as the prediction label. The optical information processing system includes an encoder, a mode selection transmission module, and a decoder. The encoder is used to modulate the sample input light field to generate a continuous complex light field. The mode selection transmission module is used to discretize the continuous complex light field into a low-dimensional latent space and obtain the output light field through nonlinear evolution. The decoder is used to generate the prediction label based on the output light field.

[0059] In step 503, the loss value is determined based on the predicted label and the real label in the training samples. The real label is the actual task processing result corresponding to the input light field of the sample.

[0060] In step 504, the configuration parameters of the optical information processing system are updated based on the loss value.

[0061] The sample input light field can be an image light field carrying information, a modulated coherent beam, or encoded data from a previous system. The ground truth label is determined according to the target task type. For example, for classification tasks, the ground truth label is the category label vector; for generation tasks, the ground truth label is the target image; and for reconstruction tasks, the ground truth label is the input image itself.

[0062] After inputting the light field (such as the light field of an image) from the training samples into the optical information processing system, the output task processing result can be obtained and determined as the predicted label. Then, according to the specific task type and the corresponding loss value determination method, the difference between the predicted label and the true label in the training samples can be determined, i.e., the loss value can be determined. For example, for classification tasks, the cross-entropy loss method can be used to calculate the loss value; for reconstruction or generation tasks, the mean squared error loss or perceptual loss method can be used to calculate the loss value; for multi-task or customized tasks, a corresponding physical perception loss function can be designed, and then the loss value can be calculated according to the function.

[0063] The gradient of the loss value can be backpropagated to the configuration parameters of the optical information processing system, and these configuration parameters can be iteratively updated using gradient descent or its variants until a predetermined convergence condition is met.

[0064] Correspondingly, by constructing a complete training process that includes an encoder, a mode selection transmission module, and a decoder, closed-loop optimization of system configuration parameters can be achieved, and the optical field transformation process in the physical domain can be included in the trainable scope. This enables the system to autonomously learn the optimal optical coding and nonlinear evolution strategies based on a data-driven approach, thereby significantly improving the task performance and generalization ability of optical information processing.

[0065] In some embodiments of this disclosure, the mode selection transmission module may include: an input coupling interface, a waveguide constraint and nonlinear processing unit, and an output coupling interface. The input coupling interface can be used to couple the received continuous complex optical field to the waveguide constraint and nonlinear processing unit. The waveguide constraint and nonlinear processing unit can be used to discretize the continuous complex optical field and map it to multiple discrete modes in a low-dimensional latent space, and apply nonlinear evolution to the signal of each discrete mode, wherein each discrete mode corresponds to an independent dimension of the latent space. The output coupling interface can be used to synthesize the nonlinearly evolved signal into an output optical field.

[0066] In addition, in some embodiments of this disclosure, the configuration parameters may include: encoder configuration parameters, mode selection transmission module configuration parameters, and decoder configuration parameters.

[0067] Taking metasurfaces as an example, encoder configuration parameters determine how the input light field is mapped to a high-dimensional optical feature space, which may include: geometry and size, i.e., the height, diameter (or length and width), and rotation angle of the metasurface nanopillars; spatial arrangement, i.e., the periodicity, lattice arrangement (such as square, hexagonal, or non-periodic distribution) of the metasurface units in the plane; and material parameters (if tunable materials are used): which can be dynamically adjusted through phase change materials or electro-optic materials.

[0068] In some embodiments of this disclosure, discrete modes are implemented through at least one of the following physical forms: physically separated transmission channels, intrinsic modes of the optical field, and discrete response categories of optical devices.

[0069] The configuration parameters of the mode selection transmission module, depending on the physical implementation type of the discrete mode, include: 1) For physically separated transmission channels Waveguide structural parameters: waveguide length L (determines the cumulative strength of nonlinear interaction), channel spacing and coupling coefficient κ (controls the linear coupling strength between channels), effective mode area A_eff (together with the nonlinear coefficient, determines the strength of nonlinear effect), and number of channels N (determines the latent space dimension).

[0070] Nonlinear process control parameters: effective nonlinear coefficient γ (depends on the waveguide material and characterizes the core nonlinear intensity), injected optical power distribution (input power of each channel, which is the driving source of nonlinear effects).

[0071] External control parameters: voltage applied to the electrodes (electro-optic control) or operating temperature (thermo-optic control), used to dynamically adjust the coupling coefficient and propagation constant.

[0072] 2) Eigenmodes of the light field Mode selection parameters: mode excitation coefficient (amplitude) and relative phase, coupling coefficient of mode converter, and mode purity.

[0073] Topology parameter (OAM mode): Topology load number l.

[0074] Polarization parameters: polarization angle, ellipticity, Stokes parameters.

[0075] Wavelength / frequency parameters: center wavelength, wavelength detuning.

[0076] 3) Discrete response categories for optical devices Threshold parameters: saturation intensity, modulation depth, and recovery time.

[0077] Resonance parameters: resonance wavelength, Q value, and free spectral range.

[0078] Switch parameters: crystallization / amorphization threshold, phase transition rate.

[0079] In addition, the configuration parameters of the mode selection and transmission module also include coupling efficiency parameters, which may include geometric alignment parameters or mode field matching conditions of the input coupling interface and the output coupling interface, affecting energy transmission efficiency.

[0080] In addition, if the decoder is an optical decoder, the decoder configuration parameters can be similar to the encoder configuration parameters. If the decoder is an optical-digital hybrid decoder, the decoder configuration parameters can include the weight parameters, biases, etc. of the neural network model.

[0081] After training, an optical information processing system can be fabricated based on the determined configuration parameters, and the fabricated optical information processing system can be applied to actual task processing.

[0082] This disclosure also discloses a training device for an optical information processing system, and correspondingly, Figure 6 This is a schematic diagram of the structural composition of an embodiment 600 of the optical information processing system described in this disclosure. Figure 6 As shown, it includes: a sample acquisition module 601 and a system training module 602.

[0083] The sample acquisition module 601 is used to acquire training samples, which include the sample input light field and the real label.

[0084] The system training module 602 is used to input the sample input light field into the optical information processing system, obtain the output task processing result, and determine it as the prediction label. The optical information processing system includes an encoder, a mode selection transmission module, and a decoder. The encoder is used to modulate the sample input light field to generate a continuous complex light field. The mode selection transmission module is used to discretize the continuous complex light field into a low-dimensional latent space and obtain the output light field through nonlinear evolution. The decoder is used to generate prediction labels based on the output light field. The loss value is determined based on the prediction label and the real label in the training samples. The real label is the actual task processing result corresponding to the sample input light field. The configuration parameters of the optical information processing system are updated based on the loss value.

[0085] In some embodiments of this disclosure, the mode selection transmission module may include: an input coupling interface, a waveguide constraint and nonlinear processing unit, and an output coupling interface. The input coupling interface can be used to couple the received continuous complex optical field to the waveguide constraint and nonlinear processing unit. The waveguide constraint and nonlinear processing unit can be used to discretize the continuous complex optical field and map it to multiple discrete modes in a low-dimensional latent space, and apply nonlinear evolution to the signal of each discrete mode, wherein each discrete mode corresponds to an independent dimension of the latent space. The output coupling interface can be used to synthesize the nonlinearly evolved signal into an output optical field.

[0086] In some embodiments of this disclosure, the configuration parameters may include: encoder configuration parameters, mode selection transmission module configuration parameters, and decoder configuration parameters.

[0087] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0088] Figure 7 A schematic block diagram of an electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0089] like Figure 7As shown, the electronic device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. The RAM 703 may also store various programs and data required for the operation of the electronic device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0090] Multiple components in electronic device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of displays, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows electronic device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0091] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose AI computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as those described in this disclosure. For example, in some embodiments, the methods described in this disclosure can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the methods described in this disclosure can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the methods described herein by any other suitable means (e.g., by means of firmware).

[0092] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard parts (ASSPs), systems-on-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0093] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0094] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory (EPROM), flash memory, optical fiber, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0095] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0096] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0097] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0098] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0099] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An optical information processing system, characterized in that, include: Encoder, mode selection transmission module, and decoder; The encoder is used to modulate the input optical field to generate a continuous complex optical field; The mode selection transmission module includes an input coupling interface, a waveguide constraint and nonlinear processing unit, and an output coupling interface. The input coupling interface is used to couple the received continuous complex optical field to the waveguide constraint and nonlinear processing unit. The waveguide constraint and nonlinear processing unit is used to discretize the continuous complex optical field and map it to multiple discrete modes in a low-dimensional latent space, and apply nonlinear evolution to the signal of each discrete mode. Each discrete mode corresponds to an independent dimension of the latent space. The discrete modes are implemented through physically separated transmission channels. The output coupling interface is used to synthesize the nonlinearly evolved signal into an output optical field. The decoder is used to generate the task processing result corresponding to the input light field based on the output light field.

2. The system according to claim 1, characterized in that, The encoder includes an optical encoder composed of M layers of metasurfaces or diffractive optical elements, where M is a positive integer.

3. The system according to claim 1, characterized in that, The nonlinear evolution includes: nonlinear transformation and intermode coupling of the discrete mode signal using the inherent nonlinear optical effects of the waveguide material, and / or programmable control of the nonlinear evolution through an integrated active modulation mechanism.

4. The system according to claim 1, characterized in that, The decoder includes: an optical decoder composed of a P-layer metasurface or diffractive optical elements, where P is a positive integer; Alternatively, the decoder may include an optical-digital hybrid decoder consisting of a photosensitive element and a neural network model.

5. A training method for an optical information processing system, characterized in that, include: Acquire training samples, which include the sample input light field and the real label; The sample is input into the optical information processing system to obtain the output task processing result, which is then determined as the prediction label. The optical information processing system includes an encoder, a mode selection transmission module, and a decoder. The encoder modulates the sample input light field to generate a continuous complex light field. The mode selection transmission module includes an input coupling interface, a waveguide constraint and nonlinear processing unit, and an output coupling interface. The input coupling interface couples the received continuous complex light field to the waveguide constraint and nonlinear processing unit. The waveguide constraint and nonlinear processing unit discretizes the continuous complex light field and maps it to multiple discrete modes in a low-dimensional latent space, and applies nonlinear evolution to the signals of each discrete mode. Each discrete mode corresponds to an independent dimension of the latent space. The discrete modes are implemented through physically separated transmission channels. The output coupling interface synthesizes the nonlinearly evolved signals into an output light field. The decoder generates the predicted label based on the output light field. The loss value is determined based on the predicted label and the real label in the training sample, where the real label is the actual task processing result corresponding to the input light field of the sample. The configuration parameters of the optical information processing system are updated based on the loss value.

6. The method according to claim 5, characterized in that, The configuration parameters include: encoder configuration parameters, mode selection transmission module configuration parameters, and decoder configuration parameters.

7. A training device for an optical information processing system, characterized in that, include: Sample acquisition module and system training module; The sample acquisition module is used to acquire training samples, which include the sample input light field and the real label. The system training module is used to input the sample input light field into the optical information processing system, obtain the output task processing result, and determine it as the prediction label; The optical information processing system includes an encoder, a mode selection transmission module, and a decoder. The encoder modulates the sample input light field to generate a continuous complex light field. The mode selection transmission module includes an input coupling interface, a waveguide constraint and nonlinear processing unit, and an output coupling interface. The input coupling interface couples the received continuous complex light field to the waveguide constraint and nonlinear processing unit. The waveguide constraint and nonlinear processing unit discretizes the continuous complex light field and maps it to multiple discrete modes in a low-dimensional latent space, and applies nonlinear evolution to the signals of each discrete mode. Each discrete mode corresponds to an independent dimension of the latent space. The discrete modes are implemented through physically separated transmission channels. The output coupling interface synthesizes the nonlinearly evolved signals into an output light field. The decoder generates the predicted label based on the output light field. The loss value is determined based on the predicted label and the real label in the training sample, where the real label is the actual task processing result corresponding to the input light field of the sample; the configuration parameters of the optical information processing system are updated based on the loss value.

8. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 5-6.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 5-6.

Images