A single-dimension vector operation method, electronic device, computer readable storage medium and integrated circuit

Abstract

This invention provides a method for operating on single-dimensional vectors, an electronic device, a computer-readable storage medium, and an integrated circuit, relating to the fields of data processing and artificial intelligence. The method includes: reading a first single-dimensional vector data to be operated on, wherein the data structure of the vector data includes at least two types of fields: a dimension identifier and a magnitude value. The dimension identifier is used to uniquely identify the dimension of the vector, and the magnitude value is used to represent the data or numerical value of the vector. Based on the dimension identifier of the first single-dimensional vector data to be operated on, one or more single-dimensional vector data with the same dimension identifier value are searched in memory, and mathematical or logical operations are performed on the magnitude values ​​of these single-dimensional vector data to obtain the result. Because the dimension identifier is a globally unique identifier and does not depend on the position of the vector data in the storage sequence, dimensions can be independently identified and referenced across dimensions, vectors, models, and systems, breaking the constraint of rigidly locking the dimension space in traditional vector schemes and realizing independent operation and interoperability at the dimension level.

Description

Technical Field

[0001] This invention relates to the fields of data processing and artificial intelligence technology, and in particular to a method for operating on a single-dimensional vector, an electronic device, a computer-readable storage medium, and an integrated circuit. Background Technology

[0002] In the fields of artificial intelligence and machine learning, vectors are the basic structure for representing data. Traditional methods encode data as vectors, which are N-dimensional floating-point arrays. In this approach, all dimensions of a vector are bound to the same array, and the identity of each dimension is determined solely by its index within the array.

[0003] Traditional vector algorithms suffer from the following problems: the vector space is fixed after training, and the number of dimensions cannot be dynamically increased or decreased according to the scale of knowledge or task requirements. Sparse and dense vectors only differ in computational scale, without changing the dependence on fixed dimensions and local indices. Furthermore, a large number of near-zero values ​​forming "silent slots" continue to consume storage and computational resources. The dimensional spaces of vector data with different dimensions are not interconnected; even vector data with the same number of dimensions but different construction methods are also mutually exclusive. Dimensional identity exists only in the local array index of each vector data; even if vector data with the same index are not interconnected, they do not necessarily belong to the same dimension. Therefore, traditional vector algorithms lack the ability to recognize cross-vector data. Knowledge coupling exists only in a single matrix parameter space; during incremental learning, new knowledge overwrites old knowledge, and there is a chance of catastrophic forgetting due to changes in relative positions. Furthermore, single-vector compression is prone to losing details and relationships; a unified distance metric is difficult to adapt to different tasks; distance concentration and hub effects are prone to occur in high-dimensional spaces; it is difficult to accurately express structured relationships such as causality, negation, temporal sequence, and quantity; model upgrades usually require full recoding and index reconstruction; in large-scale retrieval, it is difficult to balance accuracy, speed, and memory; retrieval results lack interpretability and are highly sensitive to the data partitioning method. Summary of the Invention

[0004] This invention provides a method for operating on single-dimensional vectors, an electronic device, a computer-readable storage medium, and an integrated circuit to solve problems in traditional vector schemes, such as rigid locking of the dimension space, forced participation of all dimensions in the calculation, dimension identity determined only by array subscripts leading to non-interoperability between dimensions, and catastrophic forgetting caused by knowledge coupling.

[0005] This invention provides a method for operating on a single-dimensional vector, including: Read the first single-dimensional vector data to be processed. The data structure of the vector data must include at least two types of fields: A dimension identifier is used to uniquely identify the dimension of the vector, enabling it to be independently identified and referenced. A quantity or value used to represent the data or numerical value of the vector; Based on the dimension identifier of the first single-dimensional vector data to be processed, search in the memory for one or more single-dimensional vector data with the same dimension identifier value, and perform mathematical or logical operations on the values ​​of these single-dimensional vector data to obtain the result.

[0006] The single-dimensional vector data structure with identifiable dimensions used in the single-dimensional vector operation method provided by the present invention has two types of fields: dimension identifier and value. At the same time, the dimension identifier is a globally unique identifier and does not depend on the position of the vector data in its storage sequence or the position of other vector data.

[0007] The present invention also provides an electronic device, including a processor and a memory, wherein the memory stores at least one type of data stored in the single-dimensional vector data structure of the single-dimensional vector operation method.

[0008] According to an electronic device provided by the present invention, the processor is configured to perform the operation on the single-dimensional vector data based on the single-dimensional vector operation method described above.

[0009] The present invention also provides a computer-readable storage medium, wherein the memory stores at least one type of data stored in the single-dimensional vector data structure of the single-dimensional vector operation method.

[0010] The present invention also provides an integrated circuit, including a processing circuit and a storage circuit, wherein the storage circuit is configured with data stored in the single-dimensional vector data structure in the single-dimensional vector operation method.

[0011] According to an integrated circuit provided by the present invention, the processing circuit is configured to perform the operation on the single-dimensional vector data based on the single-dimensional vector operation method.

[0012] The present invention provides a method, electronic device, computer-readable storage medium, and integrated circuit for operating on single-dimensional vectors. This method reads the single-dimensional vector data to be operated on, searches for vector data with the same dimension identifier value based on the data's dimension identifier, and performs mathematical or logical operations on the values ​​of these vector data to obtain the result. Since the dimension identifier is a globally unique identifier, it does not depend on the data's position in the storage sequence. Dimensions can be independently identified and referenced across models and systems, breaking the constraint of rigidly locking the dimension space in traditional vector schemes and achieving independent operation and interoperability at the dimension level. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0014] In traditional vector schemes, vectors are represented by fixed-length arrays, and the identity of a dimension is determined solely by its array index. The index is a local number within the array, and dimensions at the same index in different vectors have no semantic connection. Furthermore, all dimensions are bound to the same array, forcing them to participate in all calculations regardless of whether their values ​​have any practical significance. This rigid structure results in a dimension space that cannot be dynamically adjusted, dimensions that cannot interoperate across arrays, and knowledge that cannot be incrementally updated due to its coupling to a single parameter space.

[0015] This invention proposes a single-dimensional vector operation method, the core of which is to assign a globally unique dimension identifier to each dimension of the vector data, thereby decoupling the dimension identity from the array subscript, so that each dimension can be stored, identified and operated independently.

[0016] The following describes the single-dimensional vector operation method of the present invention in detail with reference to an embodiment.

[0017] Step 1: Based on the dimension identifier in the first single-dimensional vector data to be processed, retrieve and find the second single-dimensional vector data to be processed with the same dimension identifier in the memory.

[0018] The data structure of the single-dimensional vector data includes at least two types of fields: dimension identifier and magnitude. The dimension identifier uniquely identifies the dimension of the vector, enabling it to be independently identified and referenced. The magnitude represents the data or numerical value of the vector. The dimension identifier is a globally unique identifier and does not depend on the position of the vector data within its storage sequence compared to other vector data.

[0019] Step 2: Extract the magnitude data from the first and second single-dimensional vector data to be processed.

[0020] Step 3: Perform mathematical or logical operations on the two extracted data values ​​and obtain the results.

[0021] Step 4: Combine the consistent dimension identifiers in the two single-dimensional vector data to be processed with the result of the mathematical or logical operation as a new value to form a new single-dimensional vector data.

[0022] Step 5: Store the new single-dimensional vector data in memory, overwriting the original second single-dimensional vector data to be processed.

[0023] The calculation is complete.

[0024] In the above steps, the structure of a single single-dimensional vector data represents only one dimension. When it is necessary to represent the multi-dimensional features of an entity, multiple single-dimensional vector data are combined into a vector set. The number of different vector data in the vector set can change dynamically, and there is no preset upper limit.

[0025] Because of the global uniqueness of dimension identifiers, the same dimension identifier can be referenced simultaneously by multiple vector sets of different entities. When the value of the single-dimensional vector data corresponding to that dimension identifier is updated, all vector sets referencing that data receive the updated value because they use the same dimension identifier. This allows the same dimension to be shared and synchronized across models and systems.

[0026] The present invention also provides an electronic device, including a processor and a memory, wherein the memory stores at least one type of data stored in a structure corresponding to the vector data in the single-dimensional vector operation method. The processor is configured to perform the operation on the vector data based on the single-dimensional vector operation method.

[0027] The present invention also provides a computer-readable storage medium, wherein the memory stores at least one type of data stored in the structure of the vector data in the single-dimensional vector operation method.

[0028] The present invention also provides an integrated circuit, including a processing circuit and a storage circuit, wherein the storage circuit is configured to store data in the structure of the vector data in the single-dimensional vector operation method. The processing circuit is configured to perform the operation on the vector data based on the single-dimensional vector operation method.

[0029] The embodiments described above are merely illustrative, and those skilled in the art can understand and implement them without any creative effort.

[0030] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for operating on a single-dimensional vector, characterized in that, include: Read the first single-dimensional vector data to be processed. The data structure of the vector data must include at least two types of fields: A dimension identifier is used to uniquely identify the dimension of the vector, enabling it to be independently identified and referenced. A quantity or value used to represent the data or numerical value of the vector; Based on the dimension identifier of the first single-dimensional vector data to be processed, search in the memory for one or more single-dimensional vector data with the same dimension identifier value, and perform mathematical or logical operations on the values ​​of these single-dimensional vector data to obtain the result.

2. The structure of the vector data in the single-dimensional vector operation method according to claim 1, characterized in that, It possesses both dimension identifier and value fields. Furthermore, the dimension identifier is a globally unique identifier and does not depend on the position of the vector data in its storage sequence or the position of other vector data.

3. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores at least one type of data stored in the structure of the vector data in the single-dimensional vector operation method as described in any one of claims 1 to 2.

4. The electronic device according to claim 3, characterized in that, The processor is configured to perform the operation on the vector data based on the single-dimensional vector operation method according to claim 1.

5. A computer-readable storage medium, wherein the memory stores at least one type of data stored in the structure of the vector data in the single-dimensional vector operation method as described in any one of claims 1 to 2.

6. An integrated circuit, characterized in that, The data includes data stored in a structure configured in the processing circuit and the storage circuit as described in any one of claims 1 to 2, using the vector data in the single-dimensional vector operation method.

7. The integrated circuit according to claim 6, characterized in that, The processing circuit is configured to perform the operation on the vector data based on the single-dimensional vector operation method according to claim 1.