System, method and related method for implementing symmetric tensor network for quantum machine learning

Abstract

To provide a computer calculation framework and a method for implementing a symmetric tensor network for quantum machine learning using symmetric deep layer learning.SOLUTION: A framework and a method include executing a technique for developing a computer calculation framework, creating a mathematical structure for machine learning, and utilizing deep layer learning for faster convergence and training. A system includes a hardware component for mounting a model, an optimization technique for adjusting elements, and an inference module for predicting a new data point. The used mathematical structure is a symmetric tensor network constructed by using high-level symmetry and exchanges a weight matrix in a deep layer learning module with the symmetric tensor network.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to the implementation of symmetric tensor networks in quantum machine learning, particularly symmetric deep learning. It includes systems and methods that utilize the high symmetry of datasets for investigation.

Background Art

[0002] Quantum machine learning is a rapidly evolving field that combines quantum physics and machine learning techniques to solve complex computer calculation problems. One of the important challenges in this field is the efficient handling and processing of large datasets. Conventional machine learning methods often struggle with the high dimensionality and complexity of these datasets. Moreover, traditional deep learning models, although powerful, often require significant computer calculation resources and time for training, which can be a limiting factor in many applications. Additionally, these models do not always take into account the inherent high symmetry present in many datasets, which can lead to inefficiencies and inaccuracies in the learning process. Therefore, there is a need to obtain more efficient and accurate methods for implementing machine learning algorithms in the quantum realm.

Summary of the Invention

Means for Solving the Problems

[0003] According to an embodiment, a computer computing framework is provided for implementing a symmetric tensor network for quantum machine learning using symmetric deep learning. Techniques or procedures for deploying the computer computing framework are executed, and mathematical structures are created for the utilization of types of machine learning. Types of deep learning are utilized for faster convergence and training and better accuracy. Hardware components are utilized to implement the model, and data is extracted from a set of data. A computer computing unit is arranged within a layer, and a numerical array within a deep learning module is replaced by a mathematical structure. A specific type of mathematical structure is used, and optimization techniques are applied to adjust adjustable elements. A performance measure for a subset of the dataset is reduced, and an output of the model for unseen data is generated.

[0004] According to another embodiment, a method is provided for implementing a symmetric tensor network for quantum machine learning using symmetric deep learning. The method includes executing techniques or procedures for deploying a computer computing framework, creating mathematical structures for the utilization of types of machine learning, utilizing types of deep learning for faster convergence and training and better accuracy, utilizing hardware components to implement the model, extracting data from a set of data, arranging a computer computing unit within a layer, replacing a numerical array within a deep learning module with a mathematical structure, using a specific type of mathematical structure, applying optimization techniques to adjust adjustable elements, reducing a performance measure for a subset of the dataset, and generating an output of the model for unseen data.

Brief Description of the Drawings

[0005]

Figure 1

Figure 1A

Figure 1B

Figure 1C

Figure 1D

Figure 1E

Figure 1F

Figure 1G

Figure 1H

Figure 1I

Figure 1J

Figure 2

Figure 2A

Figure 2B

DETAILED DESCRIPTION OF THE INVENTION

[0006] Step 100 relates to the process of setting up a computer computing environment that supports the use of symmetric tensor networks for machine learning. This involves establishing a system or structure that can accommodate the implementation of these networks.

[0007] A system or structure called a computer computing framework is the comprehensive environment in which the network is implemented. This can take the form of various types of computing systems. The methods or algorithms used to establish and configure the computer computing framework are called techniques or procedures. These can include initialization procedures, setting up the environment, and deployment strategies.

[0008] A symmetric tensor network is a mathematical structure with symmetry that remains invariant under certain specific transformations. In the context of machine learning, these networks are used to represent states and manipulate them in an efficient manner in terms of computer computing.

[0009] The process of deploying a computer computing framework includes configuring a system for using symmetric tensor networks, setting up the necessary resources, and initializing a network with appropriate parameters. This enables the efficient execution of machine learning algorithms and, in some cases, leads to faster convergence and training, as well as improved accuracy.

[0010] The deployment of a computer computing framework sets the stage for all subsequent steps and operations and affects the overall performance and effectiveness of the system.

[0011] Step 102 relates to the development and implementation of mathematical structures for use in machine learning, particularly symmetric tensor networks. These networks are mathematical constructs designed to represent states and manipulate them in an efficient manner.

[0012] Symmetric tensor networks are characterized by their symmetry, which means that they remain invariant under certain transformations. This property is exploited in machine learning, where the efficient representation and manipulation of data can improve the performance of algorithms.

[0013] The development of these mathematical structures involves defining the network architecture, initializing the network parameters, and setting up the network environment. These operations are performed using various techniques and procedures selected based on the specific requirements of the machine learning task.

[0014] The development of symmetric tensor networks is part of the process of setting up the computer calculation framework described in step 100. It provides the basis for subsequent operations, including the placement of computer calculation units within layers (step 110), the replacement of numerical arrays in deep learning modules with mathematical structures (step 112), and the application of optimization techniques to adjust the elements (step 116).

[0015] In summary, step 102 involves the development of mathematical structures for use in machine learning, particularly symmetric tensor networks. This step is part of the process of setting up the computer calculation framework and affects subsequent steps and operations.

[0016] Step 104 relates to the application of a specific type of deep learning known as symmetric deep learning. This type of deep learning uses symmetric tensor networks, which are mathematical structures designed to efficiently represent and manipulate states.

[0017] Symmetric tensor networks have symmetry such that they remain invariant under certain transformations. This property is exploited in machine learning, where the efficient representation and manipulation of data can improve the performance of the algorithm.

[0018] The application of symmetric deep learning involves the use of machine learning algorithms designed to cooperate with symmetric tensor networks. These algorithms are implemented within the computer computing framework established in step 100 and use the mathematical structures developed in step 102.

[0019] The goal of this step is to improve the speed of convergence and the accuracy of the learning process. Convergence relates to the process of the learning algorithm reaching its optimal state, while accuracy relates to the certainty of the results produced by the learning process.

[0020] In summary, step 104 involves the application of symmetric deep learning in the computer computing framework. This step is part of the learning process and affects the efficiency and performance of the system.

[0021] Step 106 relates to the use of specific hardware such as a deep learning chip corresponding to the computer computing framework and the execution of the symmetric deep learning process.

[0022] A deep learning chip is hardware designed to accelerate computationally intensive tasks involved in deep learning. These tasks include operations such as matrix multiplication and convolution, which underlie many deep learning algorithms.

[0023] The use of a deep learning chip involves integrating these hardware components into the computer computing framework. This integration may require not only the physical assembly of the chip into the computing system but also software configuration to ensure that the deep learning algorithm can effectively utilize the computing power of the chip.

[0024] By using deep learning chips, the speed of the deep learning process can be improved, resulting in faster convergence and training time. This is because these chips are designed to execute common computations in deep learning more quickly and efficiently than general-purpose CPUs.

[0025] In summary, step 106 involves the use of deep learning chips in a computer computing framework. This step provides the necessary hardware resources to handle intensive tasks in the computer computations involved in deep learning.

[0026] Step 108 relates to the process of reading and processing data from a dataset. This data, known as symmetric classical data, functions as input for a symmetric deep learning process.

[0027] The process of extracting data includes reading data from the dataset and converting it into a form that can be used by a deep learning algorithm. This includes various data preprocessing steps such as normalization, transformation, and encoding, depending on the requirements of the deep learning algorithm and the nature of the data.

[0028] Symmetric classical data is data that has certain symmetries and remains invariant under certain specific transformations. These properties can be exploited by symmetric tensor networks to improve the efficiency and performance of the deep learning process.

[0029] The extraction of data provides the input data that the deep learning algorithm uses to learn and make predictions. The characteristics of the extracted data can affect the performance of the deep learning process.

[0030] In summary, step 108 includes the extraction of symmetric classical data from a dataset. This step provides input data for a symmetric deep learning process.

[0031] Step 110 relates to the process of organizing computer computing units known as neurons within a layered structure in a symmetric tensor deep learning module.

[0032] A neuron is a computer computing unit that processes a set of inputs, applies weights to these inputs, and passes the result through an activation function to produce an output. The weights of a neuron are adjusted during the learning process to minimize the difference between the predicted output and the actual output.

[0033] The organization of neurons into layers is a structure used in neural networks. Each layer of neurons processes the output of the previous layer at its input and sends its output to the next layer. The first layer of neurons processes the input data, and the last layer produces the final output of the network.

[0034] The organization of neurons within a layer enables the network to learn patterns in the data. Each layer of neurons can recognize different features of the data, and the layers can be built on top of each other to recognize complex patterns.

[0035] In summary, step 110 involves the organization of neurons within a layer in a symmetric tensor deep learning module. This step affects the structure and function of the neural network and its ability to learn and make predictions.

[0036] Step 112 relates to the process of exchanging numerical arrays in a deep learning module with mathematical structures. In particular, this involves exchanging the weight matrix, which is a numerical array used within a neuron, with a symmetric tensor network.

[0037] The weight matrix is a numerical array that stores the weights of the connections between neurons in a neural network. These weights are adjusted during the learning process to minimize the difference between the predicted output and the actual output.

[0038] Symmetric tensor networks are mathematical structures designed to efficiently represent and skillfully manipulate states. They have the property of being able to remain invariant under certain specific transformations, and this property can be utilized to improve the efficiency of the learning process.

[0039] Exchanging the weight matrix with a symmetric tensor network involves initializing a symmetric tensor network with appropriate parameters and configuring the neurons to use these networks instead of the weight matrix.

[0040] Since symmetric tensor networks can represent and manipulate data more efficiently than conventional weight matrices, the efficiency of the learning process can be improved by exchanging the weight matrix with a symmetric tensor network.

[0041] In summary, step 112 involves exchanging the weight matrix in the deep learning module with a symmetric tensor network. This step affects the structure and function of the neural network and its ability to learn and make predictions.

[0042] Step 114 relates to the application of a specific mathematical structure known as a symmetric matrix generation operator in the symmetric tensor deep learning module.

[0043] Symmetric matrix generation operators are mathematical structures used to efficiently represent and skillfully manipulate states in quantum machine learning. They are a type of symmetric tensor network, which is a mathematical structure that remains invariant under certain specific transformations.

[0044] The application of symmetric matrix generation operators involves integrating these mathematical structures into the symmetric tensor deep learning module. This may include initializing a symmetric matrix generation operator with appropriate parameters and configuring the module to use these operators instead of other mathematical structures.

[0045] By applying a symmetric matrix generation operator, the efficiency of the deep learning process can be improved. Due to its symmetry, the symmetric matrix generation operator can represent data more efficiently and manipulate data skillfully compared to other mathematical structures.

[0046] In summary, step 114 involves the application of a symmetric matrix generation operator in a symmetric tensor deep learning module. This step affects the structure and function of the neural network and its ability to learn and make predictions.

[0047] Step 116 relates to the use of optimization techniques to adjust elements within the network. In particular, this involves using classical optimization algorithms such as gradient descent to adjust the parameters of the symmetric tensor deep learning network.

[0048] Parameters are elements within the neural network that are updated during the learning process to minimize the difference between the predicted output and the actual output. In the context of symmetric tensor deep learning, parameters include the weights of the symmetric tensor network and other variables that affect the behavior of the network.

[0049] Classical optimization algorithms are mathematical methods that repeatedly adjust the parameters of the network to find the values that minimize the cost function. The cost function measures the difference between the predicted output and the actual output of the network, and the learning process aims to find the parameters that minimize this cost.

[0050] The use of classical optimization algorithms involves calculating the gradient of the cost function with respect to the parameters, updating the parameters in the direction of the negative gradient, and repeating these steps until the cost function reaches its minimum value.

[0051] In summary, step 116 involves using a classical optimization algorithm to adjust the parameters of the symmetric tensor deep learning network. This step affects how the network adjusts its parameters to learn from the data.

[0052] Step 118 relates to the process of minimizing a performance metric for a subset of the dataset. Specifically, this involves minimizing a cost function for the training set, which is a subset of the dataset used to train the symmetric tensor deep learning network.

[0053] The cost function is a metric that quantifies the difference between the predicted output and the actual output of the network. The learning process aims to find the network's parameters that minimize this cost.

[0054] The training set is a subset of the dataset used to train the network. The network adjusts its parameters to minimize the cost function based on the patterns in the training data.

[0055] Minimizing the cost function involves calculating the cost for the current parameters, adjusting the parameters in the direction of reducing the cost, and repeating these steps until the cost reaches its minimum value.

[0056] Minimizing the cost function induces parameter adjustment and affects the performance of the network. By minimizing the cost function, the network can learn to make accurate predictions on the training data and then use it to make predictions on new data.

[0057] In summary, step 118 involves minimizing the cost function for the training set. This step affects parameter adjustment and the performance of the network.

[0058] Step 120 relates to a process of making predictions for a new set of data points that are data not encountered by the model during the training process.

[0059] The generation of the output involves applying the trained symmetric tensor deep learning network to the new set of data points. The network processes these data points through that layer of neurons and produces an output for each data point. The output is a prediction of the target variable for each data point, based on the patterns that the network learned from the training data.

[0060] The new set of data points is a collection of data separate from the training set. These data points are used to evaluate the performance of the model on data that the model has not previously encountered.

[0061] By generating the output, the model becomes able to apply the patterns learned from the training data to make predictions on new data. This can provide insights and predictions for various applications.

[0062] In summary, step 120 includes the generation of the model's output for out-of-sample data. This step enables the model to apply the learned patterns to make predictions on new data.

[0063] The symmetric tensor network quantum machine learning system numbered 200 is designed to implement a symmetric tensor network for quantum machine learning using symmetric deep learning. This system includes a quantum learning core numbered 202 that executes techniques or procedures for deploying a computer computing framework. The core creates a mathematical structure for quantum machine learning, particularly a symmetric tensor network, and utilizes a type of deep learning for faster convergence and training.

[0064] Within the quantum learning core, the data processing unit numbered 202-a extracts data from a set of data. This extraction process provides the inputs necessary for the system to function. The neural network architecture numbered 202-b smooths the learning process by arranging computer calculation units within the layers, enabling the system to learn complex patterns and relationships in the data. The symmetric tensor network builder numbered 202-c enables efficient representation and skillful manipulation of high-dimensional data by replacing numerical arrays in the deep learning module with mathematical structures.

[0065] The system works by deploying a computer calculation framework, extracting and processing data, arranging computer calculation units within the layers, and replacing numerical arrays with mathematical structures. Each of these steps contributes to the functioning of the system and results in its overall performance. The system is further designed to utilize hardware components to implement the model and ensure that it can handle the computer calculation requirements of the tasks it is designed to execute.

[0066] The quantum learning core numbered 202 is part of a symmetric tensor network quantum machine learning system. It executes techniques or procedures for deploying a computer calculation framework. The core creates mathematical structures for quantum machine learning, particularly symmetric tensor networks. It further utilizes symmetric deep learning, a type of deep learning.

[0067] The quantum learning core includes a data processing unit numbered 202-a. This unit extracts data from a set of data. The neural network architecture numbered 202-b arranges computer calculation units within the layers. The symmetric tensor network builder numbered 202-c replaces numerical arrays in the deep learning module with mathematical structures, particularly symmetric tensor networks.

[0068] The quantum learning core works by deploying a computer computing framework, extracting and processing data, placing computer computing units within layers, and replacing numerical arrays with mathematical structures. Each of these steps contributes to the core's operation.

[0069] The training and optimization engine numbered 204 is part of a symmetric tensor network quantum machine learning system. This engine applies optimization techniques to adjust elements and reduce a performance measure of a subset of the dataset. The optimization technique used is a classical optimization algorithm. This algorithm fine-tunes the parameters of the symmetric tensor deep learning network. The performance measure to be reduced is the cost function for the training set.

[0070] The training and optimization engine works by applying optimization techniques to adjust elements and reduce a performance measure of a subset of the dataset. The optimization technique used is a classical optimization algorithm. This algorithm adjusts the parameters of the symmetric tensor deep learning network to minimize the cost function for the training set. The cost function is a measure of the system's performance, and minimizing it results in improved performance of the system. The training set is a subset of the dataset used to train the system. The optimization process is executed under certain conditions, such as when the system's performance does not meet the desired criteria or when there is a need to improve the system's performance. The method includes repeatedly adjusting the parameters of the symmetric tensor deep learning network until the cost function is minimized.

[0071] The prediction generator numbered 206 is part of a symmetric tensor network quantum machine learning system. This component generates the output of the model for unseen data. The generated output is the prediction made by the system for a new set of data points.

[0072] The prediction generator works by generating the output of the model for the first - sight data. The generated output is the prediction made by the system for a new set of data points. This process is executed after the system has been trained and is ready to make predictions. The method includes using a model trained to make predictions regarding a new set of data points. Thus, the predictions are used for various purposes such as making decisions or providing insights. This process contributes to the operation of the prediction generator and the system as a whole.

[0073] The hardware - integrated module numbered 208 is part of a symmetric tensor network quantum machine learning system. This component utilizes hardware components to implement the model. The hardware components used may include deep - learning chips.

[0074] The hardware - integrated module works by utilizing hardware components to implement the model. The hardware components used may include deep - learning chips. This process is executed to ensure that the system can reliably handle the computer - computing requirements of the tasks it is designed to execute. The method includes integrating the model with the hardware components. This process contributes to the operation of the hardware - integrated module and the system as a whole.

Claims

1. Implementing techniques or procedures for deploying a computational framework; Creating mathematical structures for machine learning type applications; Leveraging deep learning types for faster convergence and training, and better accuracy; The use of hardware components to implement the model; Extracting data from a data set; Arranging a computing unit within the layer; Replacing numerical arrays with mathematical structures in deep learning modules; the use of certain types of mathematical structures; applying optimization techniques to adjust the tunable elements; A reduction of the performance metric for a subset of the data set; and generating an output of said model for first-sight data.

2. 2. The computing framework of claim 1, wherein the mathematical structure is a symmetric tensor network.

3. 3. The computing framework of claim 2, wherein the symmetric tensor network is constructed using a high degree of symmetry of the data set.

4. 4. The computing framework of claim 3, wherein a weight matrix in the deep learning module is exchanged into the symmetric tensor network.

5. 5. The computing framework of claim 4, wherein the system includes a classical optimization algorithm for fine-tuning parameters of a symmetric tensor deep learning network to minimize a cost function for a training set.

6. 6. The computing framework of claim 5, wherein the system includes an inference module that makes predictions on new sets of data points.

7. 7. The computing framework of claim 6, wherein the model can be implemented on a deep learning chip.

8. 1. A method for implementing a symmetric tensor network for quantum machine learning using symmetric deep learning, comprising: Implementing techniques or procedures for deploying a computational framework; Creating mathematical structures for machine learning type applications; Leveraging deep learning types for faster convergence and training, and better accuracy; The use of hardware components to implement the model; Extracting data from a data set; Arranging a computing unit within the layer; Replacing numerical arrays with mathematical structures in deep learning modules; the use of certain types of mathematical structures; applying optimization techniques to adjust the tunable elements; A reduction of the performance metric for a subset of the data set; and generating an output of said model for first-sight data.

9. 9. The method of claim 8, wherein the mathematical structure is a symmetric tensor network.

10. 10. The method of claim 9, wherein the symmetric tensor network is constructed using a high degree of symmetry of the data set.

11. 11. The method of claim 10, wherein a weight matrix in the deep learning module is exchanged into the symmetric tensor network.

12. 12. The method of claim 11, wherein the system includes a classical optimization algorithm that fine-tunes parameters of a symmetric tensor deep learning network to minimize a cost function for a training set.

13. 13. The method of claim 12, wherein the system includes an inference module that makes predictions for new sets of data points.

14. 14. The method of claim 13, wherein the model can be implemented on a deep learning chip.

15. 15. The method of claim 14, wherein the symmetric tensor network is a symmetric matrix generating operator.

16. 16. The method of claim 15, wherein the classical optimization algorithm uses backpropagation or the like.

17. 17. The method of claim 16, wherein the parameters are fine-tuned to minimize a cost function for a training set.

18. 20. The method of claim 17, wherein the inference module makes predictions for new sets of data points.

19. 20. The method of claim 18, wherein the prediction is performed on a new set of data points.

20. 20. The method of claim 19, wherein the new set of data points is first-sight data.

Images