CN-121981184-A - Organoid device, method and device for realizing reservoir calculation based on organoid

CN121981184ACN 121981184 ACN121981184 ACN 121981184ACN-121981184-A

Abstract

The invention discloses a method for realizing reservoir calculation based on organoids, which comprises the steps of obtaining an audio signal, converting the obtained audio signal into a photoelectric stimulation pulse signal, obtaining a fluorescence signal responding to the photoelectric stimulation pulse signal through organoids, obtaining high-dimensional characteristic data according to the obtained fluorescence signal and by utilizing the reservoir calculation, and decoding the high-dimensional characteristic data to finish the identification processing of the audio signal. According to the invention, biological organoids are used as a core of a reserve pool, the natural computational fusion characteristic of a biological neural network is utilized, the energy consumption is extremely low when the organoids execute sensing and cognition tasks, and the organic fusion of data storage and processing is realized through nerve synapses, so that a large amount of energy consumption of CPU and memory data transmission in the traditional von Neumann architecture is avoided, the development requirement of low-power AI hardware is met, and a thought is provided for developing an efficient and low-cost neuromorphic chip.

Inventors

HUANG MINGQIANG
Zhan Zihai
HUANG XINGYU
ZHAO QILONG
Zhu Xule
LIANG DONG

Assignees

中国科学院深圳先进技术研究院

Dates

Publication Date: 20260505
Application Date: 20251014

Claims (10)

1. A method for performing reservoir calculations based on organoids, the method comprising: acquiring an audio signal, and converting the acquired audio signal into a photoelectric stimulation pulse signal; Acquiring a fluorescence signal responsive to the photoexcitation pulse signal via the organoid device; according to the obtained fluorescence signals, calculating to obtain high-dimensional characteristic data by utilizing a reserve pool; and decoding the high-dimensional characteristic data to finish the identification processing of the audio signal.
2. The method of claim 1, wherein the calculating the high-dimensional feature data from the acquired fluorescence signal using the reservoir comprises: Calculating the relative fluorescence intensity change amount according to the acquired fluorescence signal by using the formula delta F/F0= (Fi-F0)/F0 multiplied by 100%, wherein F0 is the fluorescence intensity at the initial 0 moment, fi is the fluorescence intensity at the i moment, delta F represents the fluorescence intensity change amount at the i moment, and delta F/F0 represents the relative fluorescence intensity change amount at the i moment; And acquiring the change of the relative fluorescence intensity change quantity along with time, so as to convert the relative fluorescence intensity change quantity into high-dimensional characteristic data.
3. The method according to claim 1 or 2, wherein decoding the high-dimensional feature data to complete the identification process of the audio signal comprises: performing data processing on the high-dimensional characteristic data to obtain decoding input data with consistent length; performing convolution operation on the decoded input data by using the following formula to obtain a convolved output result; Wherein X is the decoded input data, X epsilon R N×Cin×L , Y is the output result after convolution operation, Y epsilon R N×Cout×M , N is the batch size, M is the length of the output result after convolution operation, K is the convolution kernel size, C in is the number of input channels, C out is the number of output channels, W is the convolution kernel weight tensor, and b is the corresponding bias function; activating the convolved output result by using an activating function to obtain an activated output result; utilizing a full connection layer to linearly transform the activated output result from high dimension to low dimension; and classifying the output of the full-connection layer after the linear transformation, thereby realizing the identification and classification processing of the audio signals.
4. An apparatus for performing reservoir calculations based on organoids, said apparatus comprising: The input module is used for acquiring an audio signal and converting the acquired audio time sequence signal into a photoelectric stimulation pulse signal; The storage pool module is used for acquiring a fluorescence signal responding to the photoelectric stimulation pulse signal through the organoid device, calculating high-dimensional characteristic data according to the acquired fluorescence signal by utilizing the storage pool; And the output module is used for decoding the high-dimensional characteristic data to finish the identification processing of the audio signal.
5. The apparatus of claim 4, wherein the reservoir module comprises: A fluorescence change amount calculation unit configured to calculate a relative fluorescence intensity change amount from the acquired fluorescence signal by using the equation Δf/f0= (Fi-F0)/f0×100%, wherein F0 is a fluorescence intensity at an initial 0 time, fi is a fluorescence intensity at an i time, Δf represents a fluorescence intensity change amount at the i time, and Δf/F0 represents a relative fluorescence intensity change amount at the i time; and the high-dimensional characteristic data acquisition unit is used for acquiring the change of the relative fluorescence intensity variation along with time so as to convert the relative fluorescence intensity variation into high-dimensional characteristic data.
6. The apparatus of claim 4 or 5, wherein the output module comprises: the data preprocessing unit is used for carrying out data processing on the high-dimensional characteristic data to obtain decoding input data with consistent length; The convolution unit is used for carrying out convolution operation on the decoded input data by using the following formula so as to obtain a convolved output result; Wherein X is the decoded input data, X epsilon R N×Cin×L , Y is the output result after convolution operation, Y epsilon R N×Cout×M , N is the batch size, M is the length of the output result after convolution operation, K is the convolution kernel size, C in is the number of input channels, C out is the number of output channels, W is the convolution kernel weight tensor, and b is the corresponding bias function; The activating unit is used for activating the convolved output result by using an activating function so as to obtain an activated output result; The full-connection unit is used for linearly transforming the activated output result from high dimension to low dimension by utilizing the full-connection layer; And the output unit is used for classifying the output of the full-connection layer after the linear transformation, thereby realizing the identification and classification processing of the audio signals.
7. A organoid device for use in the apparatus of any of claims 4 to 6, wherein said organoid device is adapted to receive a photoexcitation pulse signal and to generate a fluorescence signal in response to the received photoexcitation pulse signal.
8. The organoid device of claim 7, wherein the method of making the organoid device comprises: 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 15000 rpm for 60 minutes, washing the obtained precipitate twice with deionized water and once with ethanol, and then drying in an oven at 60 ℃ for a predetermined time to obtain polydopamine particles; Dispersing polydopamine particles in dimethyl sulfoxide, adding 10w/v% of polyvinylidene fluoride-trifluoroethylene copolymer after ultrasonic treatment, stirring for a preset time, pouring the mixed liquid on the surface of a silicon template, drying for a preset time in an oven at 80 ℃ under normal pressure, annealing for a preset time in a vacuum oven at 130 ℃, stripping a film from the silicon template, and carrying out corona polarization for a preset time at 22 kilovolts; The adrenal gland eosinophil tumor cells are inoculated on the surface of the peeled film and cultured for a predetermined time to form the organoid device.
9. A computer device comprising a processor and a memory, the memory having stored therein program data, the processor being configured to execute the program data to implement the method of any of claims 1-3.
10. A computer readable storage medium for storing program data, which when executed by a processor is adapted to carry out the method of any one of claims 1-3.

Description

Organoid device, method and device for realizing reservoir calculation based on organoid Technical Field The invention belongs to the technical field of artificial intelligence algorithms, and particularly relates to an organoid device, a method and a device for realizing reservoir calculation based on organoids, computer equipment and a computer readable storage medium. Background In the era of the explosive development of Artificial Intelligence (AI), silicon-based computer chips are used as a core driving force to support the operation of artificial neural networks (ans), which are an important basis for artificial intelligence. However, training ANNs on current AI computing hardware is both energy-consuming and time-consuming, greatly affecting the speed of artificial intelligence technology iterations. The traditional computing system is a computational separation system based on a von Neumann architecture, physical separation exists between data and a data processing unit, and the system generates a large amount of energy consumption due to data transmission between a Central Processing Unit (CPU) and a memory, which is unfavorable for the development of the information age. In addition, as moore's law gradually approaches physical limits, the integrated circuit transistor density increases slowly, further restricting the continuous increase in silicon-based chip performance, and therefore, other methods for AI hardware development need to be explored. The brain-like computing technology is used as a novel brain inspiring computing technology, and breaks through the traditional computing bottleneck through innovative architecture and mode. The system adopts a memory computation fusion architecture, integrates a memory and processing unit, effectively reduces data transmission delay, responds to only key information based on an event driving mode of nerve pulse, reduces invalid operation, simulates a parallel computing mechanism and efficiently processes complex data by utilizing continuous signal characteristics. The technology is expected to solve the bottleneck problem of the current AI hardware in terms of calculation power and energy consumption. The organoid is used as a novel three-dimensional in-vitro model, is formed by self-assembly of stem cells under specific in-vitro conditions, and is a three-dimensional cell aggregate which has high similarity in structure and function with corresponding tissues or organs of a human body. The type is rich, for example, brain organoids can simulate nerve structures and signal transmission, blood vessel organoids can display blood vessel structures and substance transport functions, kidney organoids can simulate filtration functions, and tumor organoids can retain tumor heterogeneity, drug response characteristics and the like. Reservoir computation (Reservoir Computing, RC) is an improved algorithmic model based on Recurrent Neural Networks (RNNs), and is also an important implementation in brain-like computing technology. The architecture includes an input layer, a pool, and an output layer. The reserve pool is composed of a large number of nonlinear nodes which are randomly interconnected, and high-dimensional state space processing dynamic information is constructed. During training, the input of RC and the internal weight of the reserve pool are randomly initialized and fixed, and only the weight of an output layer is optimized. The mechanism simplifies the training process, makes the training process excellent in tasks such as time sequence prediction, voice recognition and the like, and is widely applied to various fields. Brain-like chips, also known as neuromorphic chips. Based on different physical realization bases, the brain-like chip can be systematically divided into a digital CMOS type technology system, a digital-analog hybrid CMOS type technology system and a new principle device type technology system. The digital CMOS brain-like chip simulates the behavior characteristics of biological neurons and synapses in a digital signal processing mode by constructing a logic gate circuit array, so that the programmable and reconfigurable computing function is realized. The digital-analog hybrid CMOS brain-like chip adopts a characteristic design concept of simulating a biological nerve unit by adopting a subthreshold analog circuit. The novel principle device type brain-like chip takes a memristor as an example, utilizes ion migration to modulate a resistance state to simulate biological synaptic plasticity, and has ion dynamics similar to an electrochemical process of biological neurons, so that the novel principle device type brain-like chip has been widely paid attention to in industry and academia in recent years. But current brain-like chips only partially mimic brain function. Although it has basic functions such as simulating neuron discharge, it is significantly different from the brain in the information processing paradigm. In the face o