Sparse Matrix Compression Core With Row-Pointer Fast Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine intelligence systems are computationally and energy intensive due to large data structures and immense discrete computations, posing challenges in storage, memory management, and parallelization, particularly with traditional artificial neural networks (ANNs) that are difficult to optimize for sparse data.
Innovation Solution
The development of methods and systems for rapid and efficient compression and decompression of sparse data using specialized data structures and clocked logic to store and retrieve sparse data values, non-sparse data value run lengths, and row pointers, facilitating efficient storage and computation in machine intelligence systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional ANN data structures are used to store sparse data, then storage space is wasted, but storage efficiency deteriorates
Solution Approach 1:
The patent segments sparse matrix data into three separate compressed structures: non-zero values, column indices, and row pointers. This segmentation allows efficient storage by only storing non-zero elements while maintaining fast access capabilities through the pointer structures.
Solution Approach 2:
The patent transforms the storage format by changing parameters from dense matrix representation to sparse representation using compression techniques. This involves converting the data structure to store only essential information (non-zero values and their positions) rather than all elements, significantly reducing storage requirements.
2Quantity of substance
If sparse data is compressed to save storage space, then storage efficiency improves, but computation speed deteriorates due to decompression overhead
Solution Approach 1:
The patent performs preliminary organization of sparse data into compressed formats with pre-computed row pointers and column indices during data loading or training phases. This preliminary structuring enables direct access to non-zero elements during inference without requiring decompression, thus maintaining computation speed while achieving storage compression.
Solution Approach 2:
The patent introduces intermediary data structures (row pointers, column indices) that act as mediators between the compressed storage format and the computational processes. These intermediaries enable efficient navigation and access to sparse data elements without full decompression, bridging the gap between compressed storage and fast computation.
3Ease of operation
If dense matrix format is used for storage, then access simplicity is maintained, but storage space increases significantly
Solution Approach 1:
The patent extracts only the essential non-zero elements from the dense matrix and stores them in a compressed format. By taking out and storing only the necessary data (non-zero values with their positions), the system achieves significant storage space reduction while maintaining the ability to access required elements efficiently.
Data Source
AI summary
Methods and systems regarding the rapid and efficient compression and decompression of sparse data are disclosed. One method for compressing a set of data from a sparse matrix includes, evaluating a sequence of data entries from the set of data, extracting a sequence of sparse data values from the sequence, extracting a sequence of non-sparse data value run lengths from the sequence, formulating a set of row pointers from the sequence, storing the sequence of sparse data values in a first set of memory addresses, and storing the sequence of non-sparse data value run lengths in a second set of memory addresses. The set of row pointers identify a set of rows of the sparse matrix in both the first and second sets of memory addresses. Rapid decompression can be conducted using the row pointers.