Feature Map Decomposition for Streaming Data Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models face latency and inefficiencies when processing streaming data due to the need for a fixed amount of data before processing can begin, leading to delayed inference and redundant processing of overlapping data sets.
Innovation Solution
The use of feature map decomposition and operator decomposition allows for the processing of streaming data in smaller portions, reducing latency and computational complexity by generating partial feature maps as data is received and aggregating results from smaller operations, enabling earlier data processing and minimizing redundancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models wait for a fixed amount of data before processing, then processing accuracy is improved, but latency increases
Solution Approach 1:
The patent divides the fixed data set into multiple smaller sliding windows that process data incrementally as it arrives. Instead of waiting for the complete fixed data set, the system segments processing into overlapping time windows that can be handled independently and combined, thereby reducing latency while maintaining processing accuracy through the aggregation of partial results.
2Measurement precision
If machine learning models process complete data sets, then processing accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the computational workload by dividing the complete data set into smaller sliding windows that can be processed independently. Each window generates partial feature maps that are subsequently combined, reducing the computational complexity of individual processing steps while maintaining overall processing accuracy through the aggregation of results from multiple windows.
Solution Approach 2:
The patent applies partial action by processing only the portion of data contained within each sliding window at any given time, rather than processing the entire data set simultaneously. This allows the system to perform incomplete processing on individual windows that, when combined, achieve the same result as complete processing of the entire data set, thereby reducing computational complexity.
3Loss of information
If machine learning models process overlapping data sets, then data completeness is improved, but redundancy increases
Solution Approach 1:
The patent merges the processing results from multiple overlapping sliding windows by combining their partial feature maps. This merging approach ensures that all data points are processed at least once across the overlapping windows, maintaining data completeness while allowing the system to efficiently aggregate results rather than treating each window as completely independent, thereby reducing computational redundancy.
Data Source
AI summary
Certain aspects of the present disclosure provide techniques for processing streaming data using machine learning models. An example method generally includes generating a first feature map for a first set of streaming data using a machine learning model. To generate the first feature map, results of one or more operations performed on each respective item in the first set of streaming data are combined into the first feature map, and the results of the one or more operations performed for each respective item in the first set of streaming data are combined into the first feature map. A second feature map is generated for a second set of streaming data using the machine learning model. A result of processing the total set of data through the machine learning model is generated based at least on a combination of the first feature map and the second feature map.