Optical neural network reconstruction-inhibition neuron model and architecture

Abstract

The invention relates to the technical field of optical computing, in particular to an optical neural network reconstruction-inhibition neuron model and architecture. The model comprises an optical reconstruction module used for receiving an optical input signal and carrying out reconfigurable modulation on the optical input signal to obtain a modulated optical signal, and a modulation parameter corresponding to the reconfigurable modulation is determined by an optical calculation task corresponding to the optical input signal; and the inhibition neuron module is used for constructing a filtering channel corresponding to the optical calculation task, filtering the modulated optical signal according to the filtering channel, and obtaining and outputting an optical output signal. By the adoption of the scheme, high-performance optical calculation can be achieved.

Description

Technical Field

[0001] The present disclosure relates to the field of optical computing technology, and in particular to an optical neural network reconstruction-inhibition neuron model and architecture. Background Art

[0002] With the rapid development of artificial intelligence and scientific computing, the complexity and scale of computing needs are also increasing. However, the existing electronic computing technology is limited by Moore's Law, and its performance is gradually approaching saturation, making it difficult to effectively cope with the increasingly stringent requirements of large-scale complex algorithms on computing power and power consumption. Light has natural advantages such as high throughput and low latency in the propagation process. Optical computing technology that uses photons instead of electrons as computing carriers is seen as the key to breaking the existing computing bottleneck. Summary of the invention

[0003] The present disclosure aims to solve one of the technical problems in the related art at least to some extent.

[0004] To this end, the first objective of the present disclosure is to propose an optical neural network reconstruction-inhibition neuron model to achieve high-performance optical computing.

[0005] The second objective of the present disclosure is to propose an optical neural network reconstruction-inhibition neuron architecture.

[0006] To achieve the above-mentioned purpose, the first embodiment of the present disclosure proposes an optical neural network reconstruction-inhibition neuron model, including: An optical reconstruction module, used for receiving an optical input signal and performing reconfigurable modulation on the optical input signal to obtain a modulated optical signal, wherein a modulation parameter corresponding to the reconfigurable modulation is determined by an optical computing task corresponding to the optical input signal; The inhibitory neuron module is used to construct a filter channel corresponding to the optical computing task, and filter the modulated optical signal according to the filter channel to obtain and output a light output signal.

[0007] Optionally, before the optical reconstruction module is used to perform reconfigurable modulation on the optical input signal, it is further used to: Loading the pre-trained weights corresponding to the optical computing task, and determining the phase modulation parameters corresponding to the pre-trained weights; The optical input signal is phase modulated according to the phase modulation parameter.

[0008] Optionally, when the optical reconstruction module is used to perform phase modulation on the optical input signal according to the phase modulation parameter, it is specifically used to: Determining a thermal adjustment voltage corresponding to the phase modulation parameter; The thermal modulation voltage is applied to the thermal phase modulator to control the thermal phase modulator to perform phase modulation on the optical input signal.

[0009] Optionally, when the inhibitory neuron module is used to construct a filter channel corresponding to the optical computing task, it is specifically used to: Inputting broadband light in a wavelength range corresponding to the optical computing task into an optical cavity of a tunable periodic arrangement structure to obtain a first coupled total energy corresponding to the wavelength range and a second coupled total energy corresponding to outside the wavelength range; With the goal of minimizing the difference between the first coupling total energy and the second coupling total energy, the optical cavity is tuned until a target optical cavity that meets iteration requirements is obtained, so as to obtain a filtering channel corresponding to the optical computing task.

[0010] Optionally, when the inhibitory neuron module is used to filter the modulated optical signal according to the filtering channel, it is specifically used to: The modulated optical signal is input into the target optical cavity to control the target optical cavity to filter the modulated optical signal.

[0011] Optionally, the model further includes: A feedback control module is used to detect the optical output signal and optimize the modulation parameter and the filter channel according to the detection result.

[0012] Optionally, the feedback control module is used to detect the optical output signal and optimize the modulation parameter and the filter channel according to the detection result, specifically for: Obtaining a small signal of distortion caused by the optical output signal to the environment; Performing parameter search on reconstruction parameters based on the distorted small signal to obtain target reconstruction parameters; The modulation parameters and the filter channel are optimized according to the target reconstruction parameters.

[0013] Optionally, the model further includes: The mode adjustment module is used to sense the execution information of the optical computing task and adjust the working modes of the optical reconstruction module and the neuron inhibition module according to the execution information.

[0014] To achieve the above-mentioned purpose, the second aspect of the present disclosure proposes an optical neural network reconstruction-inhibition neuron architecture, including: at least one optical neural network reconstruction-inhibition neuron model shown in any one of the aforementioned first aspects.

[0015] In summary, the optical neural network reconstruction-inhibition neuron model and architecture provided in the present invention, by deeply integrating optical computing with brain-like neuron methods, innovatively combines and realizes "reconfigurable modulation" and "reconfigurable filtering", which not only improves the computing density and efficiency, but also enhances the model's adaptability to complex tasks, thereby achieving high-performance optical computing.

[0016] Additional aspects and advantages of the present disclosure will be given in part in the following description and in part will be obvious from the following description or learned through practice of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present disclosure will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A schematic diagram of the structure of an optical neural network reconstruction-inhibition neuron model provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] Embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, and should not be construed as limiting the present disclosure.

[0019] The brain's efficient information processing capabilities have always been an important source of inspiration for artificial intelligence and neural network research. In particular, in complex tasks such as perception, memory, and reasoning, the brain exhibits the characteristics of low power consumption, high speed, and high robustness. In recent years, neuroscience research has found that reconstruction-inhibition neurons in the brain play an important role in information processing and energy optimization. These neurons regulate the activity of specific neurons, ensuring the effective expression of information and inhibiting the spread of redundant information, thereby improving the overall efficiency of the neural network.

[0020] Although traditional electronic neural networks have achieved great success in many applications, their energy consumption and speed limitations remain urgent issues. The power consumption problem of electronic neural networks is particularly prominent in scenarios that require real-time processing of massive amounts of data, such as autonomous driving, drone control, and biomedical signal processing. In addition, as the complexity of deep learning models continues to increase, the performance bottleneck of hardware systems is becoming increasingly apparent.

[0021] As an emerging computing framework, optical neural networks (ONNs) have shown great potential. Thanks to the high-speed propagation and high-bandwidth characteristics of light, optical neural networks can achieve extremely low-power parallel computing and are much faster than traditional electronic neural networks. However, existing optical neural network architectures often lack in-depth simulation of biological neural network mechanisms, especially in terms of information selectivity and redundant suppression. In order to further improve the performance of optical neural networks, it is an extremely innovative and challenging direction to introduce the mechanism of reconstruction-inhibition of neurons in the brain into optical neural networks and design an optical computing architecture with adaptive information regulation capabilities.

[0022] The present disclosure is described in detail below with reference to specific embodiments.

[0023] It should be noted that the human brain can be reconfigured to cope with a variety of tasks by adjusting attention, separating and integrating networks, etc. Brain-inspired neural networks are artificial neural networks designed with inspiration from the structure and function of the human brain. They simulate certain characteristics of biological nervous systems in order to process information in a more natural and efficient way. Networks usually try to imitate the brain's learning mechanism, neuronal connections, and information processing methods to improve computing power and adaptability. Intelligent optical computing networks can also achieve brain-like structures through hardware selection and architecture design.

[0024] Figure 1 This is a schematic diagram of the structure of an optical neural network reconstruction-inhibition neuron model provided by an embodiment of the present disclosure. Figure 1 As shown, the optical neural network reconstruction-inhibition neuron model includes: An optical reconstruction module, used for receiving an optical input signal and performing reconfigurable modulation on the optical input signal to obtain a modulated optical signal; The inhibitory neuron module is used to construct a filter channel corresponding to the optical computing task, and filter the modulated optical signal according to the filter channel to obtain and output the optical output signal.

[0025] According to some embodiments, the modulation parameters corresponding to the reconfigurable modulation are determined by the optical computing task corresponding to the optical input signal, thereby reflecting the reconfigurability of neurons.

[0026] In some embodiments, the modulated optical signal is filtered according to the filtering channel corresponding to the optical computing task. Therefore, irrelevant information or redundant components in the modulated optical signal can be accurately screened out and effectively removed from the modulated optical signal, thereby improving the computational accuracy and robustness of the model. This mechanism corresponds to the role of "reconstruction-inhibition neurons" in the brain, which can dynamically adjust the information flow and ensure that the final output retains only the most relevant information.

[0027] It is easy to understand that this model deeply integrates optical computing with brain-like neuron methods, innovatively combines and implements "reconfigurable modulation" and "reconfigurable filtering" to extract the key features corresponding to the optical computing task from the input signal. This key feature can then be input as a new input signal to the next optical neural network reconstruction-inhibition neuron model. This process not only ensures the integrity of the signal and improves the computing density and efficiency, but also enhances the model's adaptability to complex tasks and achieves high-performance optical computing.

[0028] Optionally, before the optical reconstruction module is used to perform reconfigurable modulation on the optical input signal, it is also used to: Loading the pre-trained weights corresponding to the optical computing task, and determining the phase modulation parameters corresponding to the pre-trained weights; The optical input signal is phase modulated according to the phase modulation parameter.

[0029] According to some embodiments, the pre-trained weights are obtained by training a preset model according to an optical computing task.

[0030] In some embodiments, the phase modulation of the optical input signal can be achieved by designing a thermal phase modulator on the optical waveguide path. For example, a modulated voltage can be applied to the thermal phase modulator through the lead-out electrode. , thereby heating the area where the thermal phase modulator is located. The temperature changes the refractive index of the local area, thus affecting the optical path of the light on the waveguide. , the phase can be modulated by thermally adjustable voltage, realizing the phase reconfigurability of neurons. Secondly, since the responses of light with different coupling modes and light with different wavelengths after passing through the thermal phase adjuster are different, the input light mode and thermally adjustable voltage of the optical input signal can be jointly optimized to achieve richer neuron reconfigurability. Among them, Q The heat generated when heating the thermal phase modulator, U is the thermal regulation voltage, For the light path.

[0031] That is, when the optical reconstruction module is used to phase modulate the optical input signal according to the phase modulation parameter, it can be used to: Determine a thermal adjustment voltage corresponding to a phase modulation parameter; A thermal modulation voltage is applied to the thermal phase modulator to control the thermal phase modulator to perform phase modulation on the optical input signal.

[0032] Optionally, when the inhibitory neuron module is used to construct a filter channel corresponding to an optical computing task, it is specifically used to: Inputting broadband light in a wavelength range corresponding to the optical computing task into an optical cavity of a tunable periodic arrangement structure to obtain a first coupling total energy corresponding to the wavelength range and a second coupling total energy corresponding to outside the wavelength range; With the goal of minimizing the difference between the first coupling total energy and the second coupling total energy, the optical cavity is tuned until a target optical cavity that meets the iteration requirements is obtained, so as to obtain a filter channel corresponding to the optical computing task.

[0033] Taking a scenario as an example, when the wavelength range corresponding to the optical computing task is 1.5 microns to 1.6 microns, the difference between the first coupled total energy and the second coupled total energy is E It can be determined according to the following formula:

[0034] According to some embodiments, during the process of tuning the optical cavity, the particle swarm genetic algorithm may be used to perform optimization iterations and solve the minimum value of the difference to search for the optimal design of the optical cavity, thereby obtaining the target optical cavity.

[0035] In some embodiments, the inhibitory neuron module may input the modulated optical signal into the target optical cavity to control the target optical cavity to filter the modulated optical signal.

[0036] It should be noted that the initial optical input signal can be pre-processed and compressed into a one-dimensional signal, and then coupled through the optical cavity of the tunable periodic arrangement structure of the neuron to selectively adjust the on-chip light intensity absorption curve, thereby achieving preliminary encoding of optical information and obtaining the optical input signal.

[0037] It is easy to understand that the inhibitory neuron module can achieve richer neuron reconfigurability by adopting a functional hardware design based on photonic crystals to realize reconfigurable filtering.

[0038] Optionally, the optical neural network reconstruction-inhibition neuron model also includes: The feedback control module is used to detect the optical output signal and optimize the modulation parameters and the filter channel according to the detection result.

[0039] It should be noted that by introducing an adaptive adjustment mechanism based on feedback control, the characteristic changes of the optical input signal are perceived in real time based on the detection of the optical output signal, and the parameter configurations of the optical reconstruction module and the inhibitory neuron module are automatically adjusted according to the characteristic changes, thereby realizing dynamic adjustment of the optical neural network reconstruction-inhibition neuron model, so that the model can flexibly respond to different task requirements, maintain excellent performance, and ensure that the entire model can operate efficiently in a changing task environment.

[0040] According to some embodiments, the feedback control module is used to detect the optical output signal and optimize the modulation parameters and the filter channel according to the detection result, specifically for: Obtain the small signal of distortion caused by the optical output signal to the environment; Perform parameter search on reconstruction parameters based on the distorted small signal to obtain target reconstruction parameters; The modulation parameters and filter channels are optimized according to the target reconstruction parameters.

[0041] In some embodiments, a genetic algorithm-based parameter search may be performed on the reconstruction parameters based on a preset maximum search step length. The preset maximum search step length does not specifically refer to a fixed step length. For example, the preset maximum search step length may be 10.

[0042] In some embodiments, the modulation parameters and the filtering channels may be optimized by optimizing the weight distribution of the optical modulation and filtering processes according to the target reconstruction parameters.

[0043] Optionally, the optical neural network reconstruction-inhibition neuron model also includes: The mode adjustment module is used to sense the execution information of the optical computing task and adjust the working modes of the optical reconstruction module and the neuron inhibition module according to the execution information.

[0044] According to some embodiments, the execution information includes but is not limited to information such as calculation accuracy and running time.

[0045] Taking a scenario as an example, in image processing tasks, the mode adjustment module can dynamically optimize the interaction between the optical reconstruction module and the inhibition neuron module according to different scenarios. For example, if it is sensed that the optical scene coupled into the network does not change much, the maximum search step of the search algorithm can be reduced, thereby obtaining the best processing speed and the lowest power consumption.

[0046] It should be noted that through the "reinforcement learning" method, the execution status of the optical computing task is perceived and the entire optical input signal processing path is intelligently adjusted, that is, the working mode of the optical reconstruction module and the inhibitory neuron module is automatically adjusted. Therefore, the robustness and adaptability of the model can be enhanced.

[0047] In summary, the model provided in this embodiment is inspired by the fully reconfigurable mechanism of the human brain, and realizes "reconfigurable modulation" and "reconfigurable filtering" of input information, thereby significantly improving the density and accuracy of calculations while maintaining high-speed calculations, and can also achieve adaptive reconfigurable adjustments, improve the robustness and efficiency of network reasoning, and effectively reduce system energy consumption. This innovative optical architecture has broad application prospects, such as edge computing, real-time signal processing, and artificial intelligence applications in energy-constrained devices.

[0048] In order to implement the above embodiments, the present disclosure also proposes an optical neural network reconstruction-inhibition neuron architecture, including: an optical neural network reconstruction-inhibition neuron model provided by at least one of the above embodiments.

[0049] In summary, the architecture provided in this embodiment not only has efficient parallel computing capabilities, but also can maintain stable performance in complex tasks and changing environments, and can lay a solid foundation for the development of the next generation of artificial intelligence hardware.

[0050] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this disclosure shall comply with the relevant laws and regulations and shall not violate public order and good morals.

[0051] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.

[0052] The present disclosure anticipates providing implementation schemes for users to selectively block the use or access of personal information data. That is, the present disclosure anticipates providing hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, risks can be minimized by limiting data collection and deleting the data. In addition, when applicable, such personal information is de-identified to protect the privacy of the user.

[0053] The acquisition, transmission, storage, use, and processing of data in the technical solution disclosed in this disclosure are in compliance with the relevant provisions of national laws and regulations.

[0054] It should be noted that in the embodiments of the present disclosure, certain software, components, models and other existing solutions in the industry may be mentioned, which should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0055] In the description of the aforementioned embodiments, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they contradict each other.

[0056] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, "plurality" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0057] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present disclosure includes additional implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present disclosure belong.

[0058] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or otherwise processing in a suitable manner if necessary, and then stored in a computer memory.

[0059] It should be understood that the various parts of the present disclosure can be implemented in hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0060] A person skilled in the art may understand that all or part of the steps in the above-mentioned embodiment method may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0061] In addition, each functional unit in each embodiment of the present disclosure may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0062] The storage medium mentioned above may be a read-only memory, a disk or an optical disk, etc. Although the embodiments of the present disclosure have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present disclosure. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present disclosure.

Claims

1. An optical neural network reconstruction-inhibition neuron model, characterized in that: include: An optical reconstruction module, used for receiving an optical input signal and performing reconfigurable modulation on the optical input signal to obtain a modulated optical signal, wherein a modulation parameter corresponding to the reconfigurable modulation is determined by an optical computing task corresponding to the optical input signal; The inhibitory neuron module is used to construct a filter channel corresponding to the optical computing task, and filter the modulated optical signal according to the filter channel to obtain and output a light output signal.

2. The model according to claim 1, characterized in that Before the optical reconstruction module is used to perform reconfigurable modulation on the optical input signal, the optical reconstruction module is also used to: Loading the pre-trained weights corresponding to the optical computing task, and determining the phase modulation parameters corresponding to the pre-trained weights; The optical input signal is phase modulated according to the phase modulation parameter.

3. The model according to claim 2, characterized in that When the optical reconstruction module is used to perform phase modulation on the optical input signal according to the phase modulation parameter, it is specifically used to: Determining a thermal adjustment voltage corresponding to the phase modulation parameter; The thermal modulation voltage is applied to the thermal phase modulator to control the thermal phase modulator to perform phase modulation on the optical input signal.

4. The model according to claim 1, characterized in that When the inhibitory neuron module is used to construct the filter channel corresponding to the optical computing task, it is specifically used to: Inputting broadband light in a wavelength range corresponding to the optical computing task into an optical cavity of a tunable periodic arrangement structure to obtain a first coupled total energy corresponding to the wavelength range and a second coupled total energy corresponding to outside the wavelength range; With the goal of minimizing the difference between the first coupling total energy and the second coupling total energy, the optical cavity is tuned until a target optical cavity that meets iteration requirements is obtained, so as to obtain a filtering channel corresponding to the optical computing task.

5. The model according to claim 4, characterized in that When the inhibitory neuron module is used to filter the modulated optical signal according to the filtering channel, it is specifically used to: The modulated optical signal is input into the target optical cavity to control the target optical cavity to filter the modulated optical signal.

6. The model according to claim 1, characterized in that The model also includes: A feedback control module is used to detect the optical output signal and optimize the modulation parameter and the filter channel according to the detection result.

7. The model according to claim 6, characterized in that The feedback control module is used to detect the optical output signal and optimize the modulation parameter and the filter channel according to the detection result, specifically for: Obtaining a small signal of distortion caused by the optical output signal to the environment; Performing parameter search on reconstruction parameters based on the distorted small signal to obtain target reconstruction parameters; The modulation parameters and the filter channel are optimized according to the target reconstruction parameters.

8. The model according to claim 1, characterized in that The model also includes: The mode adjustment module is used to sense the execution information of the optical computing task and adjust the working modes of the optical reconstruction module and the neuron inhibition module according to the execution information.

9. An optical neural network reconstruction-inhibition neuron architecture, characterized in that: include: At least one optical neural network reconstruction-inhibitory neuron model as described in any one of claims 1 to 8.

Images