Organ-like device, and method and device for realizing reserve pool calculation based on organ-like device

Abstract

The invention discloses a method for realizing reserve pool calculation based on organoid, which comprises the following steps: acquiring an audio signal, and converting the acquired audio signal into a photoelectric stimulation pulse signal; acquiring a fluorescence signal responding to the photoelectric stimulation pulse signal through the organ-like device; according to the obtained fluorescence signals, high-dimensional feature data are obtained through calculation by using a storage pool; and decoding the high-dimensional feature data to complete identification processing of the audio signal. According to the method, a biological organoid is adopted as a core of a reserve pool, energy consumption is extremely low when the organoid executes perception and cognition tasks depending on the natural storage and calculation fusion characteristic of a biological neural network, organic fusion of data storage and processing is achieved through neural synapses, a large amount of energy consumption of CPU and memory data transmission in a traditional von Neumann architecture is avoided, and the reliability of the von Neumann architecture is improved. And the development requirement of low-power AI hardware is better met, and a thought is provided for developing an efficient and low-cost neuromorphic chip.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence algorithm technology. Specifically, it relates to an organoid device, a method and apparatus for realizing reservoir calculation based on organoids, a computer device, and a computer-readable storage medium. Background Technology

[0002] In today's era of booming artificial intelligence (AI), silicon-based computer chips serve as the core driving force, supporting the operation of artificial neural networks (ANNs), which are a crucial foundation of AI. However, training ANNs on current AI computing hardware is both energy-intensive and time-consuming, significantly impacting the speed of AI technology iteration. This traditional computing system is based on the "von Neumann" architecture, a storage-compute separation system where data and data processing units are physically separated. Furthermore, the system consumes a significant amount of energy due to data transfer between the central processing unit (CPU) and memory, hindering the development of the information age. In addition, as Moore's Law gradually approaches its physical limits, the growth rate of integrated circuit transistor density is slowing down, further restricting the continuous improvement of silicon-based chip performance. Therefore, it is necessary to explore other methods for AI hardware development.

[0003] Neuromorphic computing, a novel brain-inspired computing technology, breaks through the bottlenecks of traditional computing through innovative architecture and models. It employs a memory-computing fusion architecture, integrating storage and processing units to effectively reduce data transmission latency; its event-driven model based on neural impulses responds only to critical information, reducing unnecessary computation; and it simulates parallel computing mechanisms, utilizing the characteristics of continuous signals to efficiently process complex data. This technology holds the promise of solving the current bottlenecks in computing power and energy consumption in AI hardware.

[0004] Organoids, as novel three-dimensional in vitro models, are formed by the self-assembly of stem cells under specific in vitro conditions. They are three-dimensional cell aggregates that are highly similar in structure and function to corresponding human tissues or organs. They are diverse in type; for example, brain organoids can simulate neural structures and signal transmission, vascular organoids can demonstrate vascular structures and substance transport functions, kidney organoids can simulate filtration functions, and tumor organoids retain tumor heterogeneity and drug response characteristics.

[0005] Reservoir Computing (RC) is an algorithmic model based on recurrent neural networks (RNNs) and an important implementation method in neuromorphic computing technology. Its architecture includes an input layer, a reservoir, and an output layer. The reservoir consists of a large number of randomly interconnected nonlinear nodes, constructing a high-dimensional state space to process dynamic information. During training, the input and internal weights of RC are randomly initialized and fixed, with only the output layer weights being optimized. This mechanism simplifies the training process, making it perform exceptionally well in tasks such as time series prediction and speech recognition, and it is widely used in various fields.

[0006] Neuromorphic chips, also known as neuromorphic chips, can be systematically divided into three major technical systems based on different physical implementations: digital CMOS, mixed-signal CMOS, and novel principle devices. Digital CMOS neuromorphic chips simulate the behavioral characteristics of biological neurons and synapses through the construction of logic gate arrays and digital signal processing, achieving programmable and reconfigurable computational functions. Mixed-signal CMOS neuromorphic chips employ design concepts such as subthreshold analog circuits to simulate the characteristics of biological neural units. Novel principle device neuromorphic chips, taking memristors as an example, utilize ion migration to modulate resistive states, simulating the plasticity of biological synapses; their ion dynamics are similar to the electrochemical processes of biological neurons, and they have received widespread attention from industry and academia in recent years.

[0007] However, current neuromorphic chips can only partially simulate brain functions. While they possess basic functions such as simulating neuronal firing, their information processing paradigms differ significantly from the brain. Faced with complex scenarios, chips encounter bottlenecks in achieving advanced cognitive functions such as multimodal fusion and causal reasoning. Furthermore, their energy efficiency is 1-2 orders of magnitude lower than that of a biological brain, and they lack fault tolerance and self-repair capabilities. The biological brain, as a complex information processing system, possesses powerful computational performance due to its three-dimensional network of neurons and synapses. And through approximately 10... 14 The massive neural synapses enable the fusion of data storage and processing, and the organoids consume extremely low energy when performing sensory and cognitive tasks.

[0008] In diverse neural network training algorithms, reservoir computing maps time-series signals from a low-dimensional feature space to a high-dimensional feature space, enabling efficient extraction of input signal features through matrix operations on the device array within the reservoir. Compared to other neural network algorithms, reservoir computing only requires training the connection weights between the reservoir and the output layer. This architectural design significantly reduces the complexity and time consumption of model training, providing technical feasibility for ultra-high-speed, ultra-low-power information processing scenarios.

[0009] Various novel devices, such as memristors, achieve computation by simulating the morphology and operation of neurons. However, due to inherent power consumption such as leakage current, they struggle to achieve the low energy consumption levels of biological organoids. Therefore, leveraging the inherent high energy efficiency and low power consumption of neurons, and using biological organoids as a reservoir model, the development of novel neuromorphic chips based on new principles holds promise for solving the current bottlenecks in computing power and energy consumption in AI hardware, and exploring new approaches to AI hardware development. Summary of the Invention

[0010] To achieve the above objectives, embodiments of the present invention provide an organoid device, a method and apparatus for calculating a reservoir based on organoids, a computer device, and a computer-readable storage medium.

[0011] The method for calculating a reservoir based on organoids according to a first aspect of an embodiment of the present invention includes: acquiring an audio signal and converting the acquired audio signal into a photoelectric stimulation pulse signal; acquiring a fluorescence signal that responds to the photoelectric stimulation pulse signal via an organoid device; calculating high-dimensional feature data based on the acquired fluorescence signal and using a reservoir; and decoding the high-dimensional feature data to complete the recognition processing of the audio signal.

[0012] In one example of the method provided in the first aspect above, the step of calculating high-dimensional feature data based on the acquired fluorescence signal and using a reservoir includes:

[0013] Based on the acquired fluorescence signal, the relative fluorescence intensity change is calculated using the formula ΔF / F0=(Fi-F0) / F0×100%; where F0 is the fluorescence intensity at the initial time 0, Fi is the fluorescence intensity at time i, ΔF represents the fluorescence intensity change at time i, and ΔF / F0 represents the relative fluorescence intensity change at time i.

[0014] The change in relative fluorescence intensity over time is obtained and converted into high-dimensional feature data.

[0015] In one example of the method provided in the first aspect above, decoding the high-dimensional feature data to complete the recognition processing of the audio signal includes:

[0016] The high-dimensional feature data is processed to obtain decoded input data of consistent length;

[0017] The following formula is used to perform a convolution operation on the decoded input data to obtain the convolutional output result;

[0018]

[0019] Where X is the decoding input data, X∈R N×Cin×L Y is the output of the convolution operation, Y∈R N×Cout×M N is the batch size, M is the length of the output after the convolution operation, K is the kernel size, and C is the length of the output. in It is the number of input channels, C out is the number of output channels, W is the kernel weight tensor, and b is the corresponding bias function;

[0020] The output of the convolution is activated using an activation function to obtain the activated output.

[0021] A fully connected layer is used to linearly transform the activated output from high dimension to low dimension;

[0022] The output of the fully connected layer after linear transformation is classified to achieve the recognition and classification of audio signals.

[0023] An apparatus for calculating a reservoir based on organoids, provided in a second aspect of an embodiment of the present invention, comprises: an input module for acquiring an audio signal and converting the acquired audio timing signal into a photoelectric stimulation pulse signal; a reservoir module for acquiring a fluorescence signal that responds to the photoelectric stimulation pulse signal via an organoid device, and calculating high-dimensional feature data based on the acquired fluorescence signal and using the reservoir; and an output module for decoding the high-dimensional feature data to complete the recognition processing of the audio signal.

[0024] In one example of the apparatus provided in the second aspect above, the reservoir module includes: a fluorescence change calculation unit, used to calculate the relative fluorescence intensity change based on the acquired fluorescence signal and using the formula ΔF / F0=(Fi-F0) / F0×100%; where F0 is the fluorescence intensity at initial time 0, Fi is the fluorescence intensity at time i, ΔF represents the fluorescence intensity change at time i, and ΔF / F0 represents the relative fluorescence intensity change at time i; and a high-dimensional feature data acquisition unit, used to acquire the change of the relative fluorescence intensity change over time, thereby converting it into high-dimensional feature data.

[0025] In one example of the apparatus provided in the second aspect above, the output module includes:

[0026] The data preprocessing unit is used to process the high-dimensional feature data to obtain decoded input data of consistent length.

[0027] A convolutional unit is used to perform a convolution operation on the decoded input data using the following formula to obtain the convolutional output result;

[0028]

[0029] Where X is the decoding input data, X∈R N×Cin×L Y is the output of the convolution operation, Y∈R N×Cout×M N is the batch size, M is the length of the output after the convolution operation, K is the kernel size, and C is the length of the output. in It is the number of input channels, C out is the number of output channels, W is the kernel weight tensor, and b is the corresponding bias function;

[0030] The activation unit is used to activate the output of the convolution using an activation function to obtain the activated output.

[0031] Fully connected units are used to linearly transform the activated output from high dimension to low dimension using fully connected layers;

[0032] The output unit is used to classify the output of the fully connected layer after linear transformation, thereby realizing the recognition and classification processing of audio signals.

[0033] An organoid device provided according to a third aspect of an embodiment of the present invention is used in the above-described apparatus, wherein the organoid device is used to receive a photoelectric stimulation pulse signal and generate a fluorescence signal in response to the received photoelectric stimulation pulse signal.

[0034] In one example of the organoid device provided in the third aspect above, the method for fabricating the organoid device includes: dissolving 0.2 g of dopamine hydrochloride and 0.12 g of tris(hydroxymethyl)aminomethane in 100 mL of deionized water, stirring at room temperature for a predetermined time, centrifuging at 15,000 rpm for 60 minutes, washing the resulting precipitate twice with deionized water and once with ethanol, and then drying it in an oven at 60°C for a predetermined time to obtain polydopamine particles; dispersing the polydopamine particles in dimethyl sulfoxide, sonicating them, adding 10 w / v% of polyvinylidene fluoride-trifluoroethylene copolymer, stirring for a predetermined time, pouring the mixture onto the surface of a silicon template, drying it in an oven at 80°C under normal pressure for a predetermined time, annealing it in a vacuum oven at 130°C for a predetermined time, peeling the film off the silicon template, and subjecting it to corona polarization at 22 kV for a predetermined time; seeding pheochromocytoma cells on the surface of the peeled film and culturing them for a predetermined time to form the organoid device.

[0035] A computer device provided according to a fourth aspect of an embodiment of the present invention includes a processor and a memory, the memory storing program data, and the processor executing the program data to implement the method described above.

[0036] A computer-readable storage medium provided according to a fifth aspect of an embodiment of the present invention is used to store program data, which, when executed by a processor, is used to implement the method described above.

[0037] Beneficial effects: This invention uses biological organoids as the core of the storage pool. Relying on the natural storage-computation fusion characteristics of biological neural networks, the organoids consume very little energy when performing perception and cognition tasks. The organic integration of data storage and processing is achieved through neural synapses, avoiding the large energy consumption of CPU and memory data transmission in the traditional von Neumann architecture. This is more in line with the development needs of low-power AI hardware and provides ideas for developing efficient and low-cost neuromorphic chips.

[0038] Furthermore, the cultured organoids can achieve "unsupervised learning," autonomously adapting to changes in input signals through neural plasticity (such as the growth and connection of neural neurites triggered by changes in calcium ion concentration), without the need for complex weight initialization and adjustment. Compared to traditional reservoir calculations that require optimization of output layer weights, this invention can complete training using only a lightweight one-dimensional convolutional neural network as the readout function, significantly reducing the complexity and time consumption of model training and enabling ultra-high-speed information processing. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0040] Figure 1 This is a flowchart of a method for calculating a reservoir based on organoids according to an embodiment of the present invention;

[0041] Figure 2 This is a structural diagram of a reservoir network and a one-dimensional convolutional neural network according to an embodiment of the present invention;

[0042] Figure 3 This is a module unit diagram of an apparatus for calculating a reservoir based on organoids according to an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention;

[0044] Figure 5 This is a schematic diagram of the structure of a computer-readable storage medium according to an embodiment of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0046] The method and apparatus for reservoir calculation based on organoids according to embodiments of the present invention can be applied to the recognition of audio signals. Here, taking the recognition of audio segments of four Arabic numerals (0, 1, 2, 3) as an example, the method for reservoir calculation based on organoids according to embodiments of the present invention will be described in detail. Next, the fabrication of the organoid device will be described in detail.

[0047] Here, the photoelectric material film used for cell culture and photoelectrostimulation is composed of polyvinylidene fluoride-trifluoroethylene copolymer (P(VDF-TrFE)) and polydopamine. Methods for fabricating organoid devices include:

[0048] First, 0.2 g of dopamine hydrochloride and 0.12 g of tris(hydroxymethyl)aminomethane were dissolved in 100 mL of deionized water and stirred overnight at room temperature, followed by centrifugation at 15,000 rpm for 60 minutes. The resulting precipitate was washed twice with deionized water and once with ethanol, and then dried overnight in an oven at 60 °C to obtain polydopamine particles.

[0049] Next, polydopamine particles were dispersed in dimethyl sulfoxide, and after ultrasonic treatment, 10 w / v% (mass percentage concentration) of P(VDF-TrFE) was added. After stirring for 12 hours, the mixture was poured onto the surface of a 6-inch silicon template, dried overnight at atmospheric pressure in an oven at 80°C, and annealed in a vacuum oven at 130°C for 6 hours. The film peeled from the silicon template was then subjected to corona polarization at 22 kV for 25 minutes. Through the above steps, a photoelectric material film composed of P(VDF-TrFE) and polydopamine was prepared.

[0050] Finally, adrenal pheochromocytoma cell line (PC12) cells were seeded on the surface of a photoelectric material film composed of P(VDF-TrFE) and polydopamine and cultured for 4 days to form organoid devices.

[0051] Figure 1 This is a flowchart of a method for calculating a reservoir based on organoids according to an embodiment of the present invention. Figure 2 This is a structural diagram of a reservoir network and a one-dimensional convolutional neural network according to an embodiment of the present invention.

[0052] Refer to together Figure 1 and Figure 2 In step S110, an audio signal is acquired and converted into a photoelectric stimulation pulse signal.

[0053] Specifically, an audio signal is acquired, and at time t, the instantaneous audio signal input to the organoid reservoir constitutes the input state x(t). The photoelectric stimulation pulse signal u(t) of the organoid changes with x(t), therefore, the audio signal x(t) is encoded and converted into the photoelectric stimulation pulse signal u(t).

[0054] Therefore, the four Arabic numerals (i.e. the input audio signal) are encoded into four photoelectric stimulation pulse sequences with different light intensities and photoelectric stimulation frequencies, and each photoelectric stimulation pulse sequence is labeled with a corresponding data label, which serves as the data label required for subsequent training of the convolutional neural network model.

[0055] In step S120, a fluorescence signal that responds to a photoelectric stimulation pulse signal via an organoid device is acquired.

[0056] Specifically, to evaluate the electrophysiological response of organoids to photo-electrostimulation, we performed cellular fluorescence imaging using the intracellular calcium ion fluorescent probe Fluo4-AM. First, PC12 cells cultured for 4 days were incubated with Fluo4-AM in DMEM medium (Dulbecco's Modified Eagle Medium) without fetal bovine serum for 30 minutes, followed by incubation in warm artificial cerebrospinal fluid for another 30 minutes. Using laser scanning confocal microscopy, considering the decay time of calcium fluorescence intensity in the organoids used in this example was several tens of seconds, a 1 Hz photo-stimulation pulse sequence was used, with different duty cycles (10% or 50%) and power densities (1 or 2 W·cm²). -2 This is used to correspond to different audio signals. For example, a photoelectric pulse with a light intensity of 1W and a 50% duty cycle represents the audio signal of the number 0, and so on for other numbers, while the spot size remains unchanged. 1 cm).

[0057] The light source was turned off from 0 to 20 seconds, and the baseline calcium fluorescence intensity was measured. From 20 to 80 seconds, the light source was turned on, and a photoelectric stimulation pulse signal was input to the organoid device, recording its calcium fluorescence intensity response. From 80 to 100 seconds, the light source was turned off, and time-series fluorescence images of the incubated organoids were recorded before (0 to 20 seconds), during (20 to 80 seconds), and after (80 to 100 seconds) near-infrared light irradiation. The acquired time-series fluorescence images were used, for example, to extract the fluorescence signal using software such as ZEN 2 (blue version, Zeiss).

[0058] In step S130, high-dimensional feature data is obtained based on the acquired fluorescence signal and calculated using the reservoir.

[0059] Specifically, after acquiring the fluorescence signal, the relative fluorescence change value (ΔF / F0) for each cell was calculated according to the following formula, and its change over time was statistically analyzed.

[0060] ΔF / F0=(Fi-F0) / F0×100%

[0061] Where F0 is the fluorescence intensity at the initial time 0, Fi is the fluorescence intensity at time i, ΔF represents the change in fluorescence intensity at time i, and ΔF / F0 represents the relative change in fluorescence intensity at time i.

[0062] When the light source is turned on for 20–80 seconds and photoelectric stimulation is applied to the organoid device, the calcium ion concentration changes due to the activation of ion channels on the organoid cell membrane. This means the organoid device exhibits a nonlinear coupled photoresponse to the photoelectric stimulation pulse signal; different photoelectric stimulation pulse signals cause different changes in calcium ion concentration in the organoid. The temporal information transformed during this period is transported through the input layer to the high-dimensional space in the reservoir. The connections and activities between neurons represent their dynamic interactive responses. The activity of the output neurons is acquired using a laser scanning confocal microscope as a new high-dimensional feature space y(t), where y(t) is derived from u(t) at time t. j The data set consists of the values ​​taken at each time point. j is the number of segments in the entire input time of the photoelectric stimulus, and t is the time interval. j The time is the end time of each time interval, where j = 80, t j =80 seconds. The light source will be turned off for the next 80-100 seconds.

[0063] As described above, the activity state of output neurons is recorded by laser scanning confocal microscopy. The calcium fluorescence intensity response signal generated by neurons is acquired by laser scanning confocal microscopy. This signal characterizes the activity state response of output neurons representing the reserve pool in brain organoids. The fluorescence signal is extracted using image acquisition software, and the relative fluorescence change value (ΔF / F0) of each cell is calculated according to the formula and its change over time is statistically analyzed, thereby converting it into numerical data that can be further analyzed and processed.

[0064] In step S140, the high-dimensional feature data is decoded to complete the recognition processing of the audio signal.

[0065] Specifically, the lengths of the multiple sets of relative fluorescence change data (high-dimensional feature data) output after the reservoir calculation may be inconsistent. This is because the length of the fluorescence change data can be determined by different sampling intervals (sampling precision). To ensure that the sequence lengths of the multiple sets of high-dimensional feature data input into the convolutional neural network are consistent, the data is sampled at random intervals to unify the length. For example, the data length of 500 is cropped into a time series dataset with a data length of 90, which is used as the input data of the one-dimensional convolutional neural network.

[0066] The basic structure of the constructed one-dimensional convolutional neural network is as follows: Figure 2 As shown, it includes an input layer, convolutional layers, activation layers, fully connected layers, and an output layer. The input layer has already performed random interval sampling on the temporal data. The convolutional layers use one-dimensional convolutional kernels to perform convolution operations, extracting features from the temporal data. Then, the convolution results of different neurons are summed and a bias is added. The convolution operation can be represented by the following formula:

[0067]

[0068] Where X is the decoding input data, X∈R N×Cin×L Y is the output of the convolution operation, Y∈R N×Cout×M N is the batch size, M is the length of the output after the convolution operation, K is the kernel size, and C is the length of the output. in It is the number of input channels, C out is the number of output channels, W is the kernel weight tensor, and b is the corresponding bias function.

[0069] Here, the number of input channels C in The output channel number is 1, C. out The number of convolution kernels is 8, and the kernel size K is 5. For each kernel, the output Y after the convolution operation is... Cout .

[0070] The activation layer uses the ReLU activation function to simulate the nonlinear mapping between the input and output, which can be expressed by the following formula: Z c [m] = max(0, Y) c [m]), where Z c [m] is the output value after activation by the ReLU activation function.

[0071] Finally, through a fully connected layer, the softmax function is used to output the 4-dimensional classification, which corresponds to 4 data labels, thus obtaining the final classification result.

[0072] Furthermore, the processed dataset was divided into a 75% training set and a 25% test set. After 200 training cycles, the method based on organoid-based reservoir calculation according to an embodiment of the present invention achieved a recognition rate of up to 98.4% for audio signal classification. Moreover, the model's accuracy increased with the increase in the number of test samples.

[0073] Figure 3 This is a module unit diagram of an apparatus for calculating a reservoir based on organoids according to an embodiment of the present invention.

[0074] Reference Figure 3 According to an embodiment of the present invention, an apparatus for calculating a reservoir based on organoids includes: an input module 310 for performing step S110, which acquires an audio signal and converts the acquired audio timing signal into a photoelectric stimulation pulse signal; a reservoir module 320 for performing steps S120 and S130, which acquires a fluorescence signal that responds to the photoelectric stimulation pulse signal via an organoid device, and calculates high-dimensional feature data based on the acquired fluorescence signal and using a reservoir; and an output module 330 for performing step S140, which decodes the high-dimensional feature data to complete the recognition processing of the audio signal.

[0075] Specifically, the reservoir module 320 includes: a fluorescence change calculation unit 321, used to calculate the relative fluorescence intensity change based on the acquired fluorescence signal and using the formula ΔF / F0=(Fi-F0) / F0×100%; where F0 is the fluorescence intensity at the initial time 0, Fi is the fluorescence intensity at time i, ΔF represents the fluorescence intensity change at time i, and ΔF / F0 represents the relative fluorescence intensity change at time i; and a high-dimensional feature data acquisition unit 322, used to acquire the change of the relative fluorescence intensity change over time, thereby converting it into high-dimensional feature data.

[0076] The output module 330 includes:

[0077] The data preprocessing unit 331 is used to process the high-dimensional feature data to obtain decoded input data of consistent length.

[0078] Convolutional unit 332 is used to perform a convolution operation on the decoded input data using the following formula to obtain the convolutional output result;

[0079]

[0080] Where X is the decoding input data, X∈R N×Cin×L Y is the output of the convolution operation, Y∈R N×Cout×M N is the batch size, M is the length of the output after the convolution operation, K is the kernel size, and C is the length of the output. in It is the number of input channels, C out is the number of output channels, W is the kernel weight tensor, and b is the corresponding bias function;

[0081] Activation unit 333 is used to activate the output result after convolution using an activation function to obtain the activated output result;

[0082] Fully connected unit 334 is used to linearly transform the activated output from high dimension to low dimension using a fully connected layer;

[0083] Output unit 335 is used to classify the output of the fully connected layer after linear transformation, thereby realizing the recognition and classification processing of audio signals.

[0084] In summary, the embodiments of the present invention use biological organoids as the core of the storage pool. Relying on the natural storage-computation fusion characteristics of biological neural networks, the organoids consume very little energy when performing perception and cognitive tasks. The organic integration of data storage and processing is achieved through neural synapses, avoiding the large energy consumption of CPU and memory data transmission in the traditional von Neumann architecture. This is more in line with the development needs of low-power AI hardware and provides ideas for developing efficient and low-cost neuromorphic chips.

[0085] Furthermore, the cultured organoids can achieve "unsupervised learning," autonomously adapting to changes in input signals through neural plasticity (such as the growth and connection of neural neurites triggered by changes in calcium ion concentration), without the need for complex weight initialization and adjustment. Compared to traditional reservoir calculations that require optimization of output layer weights, embodiments of the present invention can complete training using only a lightweight one-dimensional convolutional neural network as the readout function, significantly reducing the complexity and time consumption of model training and enabling ultra-high-speed information processing.

[0086] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0087] To implement the organoid-based reservoir calculation method of the above embodiments, this application also proposes a computer device, which can be found in detail below. Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention.

[0088] The computer device 400 of this application embodiment includes a memory 41 and a processor 42, wherein the memory 41 and the processor 42 are coupled together.

[0089] The memory 41 is used to store program data, and the processor 42 is used to execute the program data to implement the method for calculating the reserve pool based on organoids as described in the above embodiments.

[0090] In this embodiment, processor 42 can also be referred to as a CPU (Central Processing Unit). Processor 42 may be an integrated circuit chip with signal processing capabilities. Processor 42 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor, or processor 42 can be any conventional processor.

[0091] To implement the organoid-based reservoir calculation method of the above embodiments, this application also provides a computer-readable storage medium, such as... Figure 5As shown, the computer-readable storage medium 500 is used to store program data 51, which, when executed by the processor, is used to implement the method for calculating the reservoir based on organoids as described in the above embodiments.

[0092] This application also provides a computer program product, wherein the computer program product includes a computer program operable to cause a computer to execute the method for calculating a reservoir based on organoids as described in the embodiments of this application. The computer program product can be a software installation package.

[0093] The method for calculating the reserve pool based on organoids described in the above embodiments of this application, when implemented as a software functional unit and sold or used as an independent product, can be stored in a device, such as a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0094] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for calculating a reservoir based on organoids, characterized in that, The method includes: Acquire audio signals and convert them into photoelectric stimulation pulse signals; Acquire fluorescence signals that respond to photoelectric stimulation pulse signals via organoid devices; Based on the acquired fluorescence signals, high-dimensional feature data are calculated using the reservoir. The high-dimensional feature data is decoded to complete the recognition and processing of the audio signal.

2. The method according to claim 1, characterized in that, The step of calculating high-dimensional feature data based on the acquired fluorescence signal and using a storage pool includes: Based on the acquired fluorescence signal, the relative fluorescence intensity change is calculated using the formula ΔF / F0=(Fi-F0) / F0×100%; where F0 is the fluorescence intensity at the initial time 0, Fi is the fluorescence intensity at time i, ΔF represents the fluorescence intensity change at time i, and ΔF / F0 represents the relative fluorescence intensity change at time i. The change in relative fluorescence intensity over time is obtained and converted into high-dimensional feature data.

3. The method according to claim 1 or 2, characterized in that, Decoding the high-dimensional feature data to complete the recognition processing of the audio signal includes: The high-dimensional feature data is processed to obtain decoded input data of consistent length; The following formula is used to perform a convolution operation on the decoded input data to obtain the convolutional output result; Where X is the decoding input data, X∈R N×Cin×L Y is the output of the convolution operation, Y∈R N×Cout×M N is the batch size, M is the length of the output after the convolution operation, K is the kernel size, and C is the length of the output. in It is the number of input channels, C out is the number of output channels, W is the kernel weight tensor, and b is the corresponding bias function; The output of the convolution is activated using an activation function to obtain the activated output. A fully connected layer is used to linearly transform the activated output from high dimension to low dimension; The output of the fully connected layer after linear transformation is classified to achieve the recognition and classification of audio signals.

4. A device for calculating a reservoir based on organoids, characterized in that, The device includes: The input module is used to acquire audio signals and convert the acquired audio timing signals into photoelectric stimulation pulse signals; The reservoir module is used to acquire fluorescence signals that respond to photoelectric stimulation pulse signals via organoid devices, and to calculate high-dimensional feature data based on the acquired fluorescence signals and the reservoir. The output module is used to decode the high-dimensional feature data to complete the recognition and processing of the audio signal.

5. The apparatus according to claim 4, characterized in that, The reserve pool module includes: The fluorescence change calculation unit is used to calculate the relative fluorescence intensity change based on the acquired fluorescence signal and using the formula ΔF / F0=(Fi-F0) / F0×100%; where F0 is the fluorescence intensity at the initial time 0, Fi is the fluorescence intensity at time i, ΔF represents the fluorescence intensity change at time i, and ΔF / F0 represents the relative fluorescence intensity change at time i. The high-dimensional feature data acquisition unit is used to acquire the change in relative fluorescence intensity over time, thereby converting it into high-dimensional feature data.

6. The apparatus according to claim 4 or 5, characterized in that, The output module includes: The data preprocessing unit is used to process the high-dimensional feature data to obtain decoded input data of consistent length. A convolutional unit is used to perform a convolution operation on the decoded input data using the following formula to obtain the convolutional output result; Where X is the decoding input data, X∈R N×Cin×L Y is the output of the convolution operation, Y∈R N×Cout×M N is the batch size, M is the length of the output after the convolution operation, K is the kernel size, and C is the length of the output. in It is the number of input channels, C out is the number of output channels, W is the kernel weight tensor, and b is the corresponding bias function; The activation unit is used to activate the output of the convolution using an activation function to obtain the activated output. Fully connected units are used to linearly transform the activated output from high dimension to low dimension using fully connected layers; The output unit is used to classify the output of the fully connected layer after linear transformation, thereby realizing the recognition and classification processing of audio signals.

7. An organoid device for use with the apparatus of any one of claims 4 to 6, wherein, The organoid device is used to receive photoelectric stimulation pulse signals and generate fluorescence signals in response to the received photoelectric stimulation pulse signals.

8. The organoid device according to claim 7, characterized in that, The method for fabricating the organoid device includes: 0.2 g of dopamine hydrochloride and 0.12 g of tris(hydroxymethyl)aminomethane were dissolved in 100 mL of deionized water and stirred at room temperature for a predetermined time. The mixture was then centrifuged at 15,000 rpm for 60 minutes. The resulting precipitate was washed twice with deionized water and once with ethanol. The precipitate was then dried in an oven at 60 °C for a predetermined time to obtain polydopamine particles. Polydopamine particles were dispersed in dimethyl sulfoxide, ultrasonically treated, and then 10 w / v% of polyvinylidene fluoride-trifluoroethylene copolymer was added. After stirring for a predetermined time, the mixture was poured onto the surface of a silicon template and dried at atmospheric pressure in an oven at 80°C for a predetermined time. Then, it was annealed in a vacuum oven at 130°C for a predetermined time. After peeling the film off the silicon template, it was subjected to corona polarization at 22 kV for a predetermined time. Adrenal pheochromocytoma cells were seeded onto the surface of a peeled membrane and cultured for a predetermined time to form the organoid device.

9. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing program data, and the processor executing the program data to implement the method as described in any one of claims 1-3.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program data, which, when executed by a processor, is used to implement the method according to any one of claims 1-3.

Images