Adaptive Configuration-Data Inference for Resource-Limited Edge Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for inference processing, such as those using deep learning, do not effectively improve throughput, particularly in edge environments with limited computational resources.
Innovation Solution
An information processing device that generates configuration data using multiple pieces of data, performs inference on this data, and determines success or failure of inference for each piece, allowing it to adaptively perform inference based on successful or failed results using configuration and data-specific models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If inference processing is performed on individual data pieces, then processing accuracy can be maintained, but throughput is low due to insufficient parallelism
Solution Approach 1:
The patent merges multiple individual data pieces into a single configuration data structure that contains multiple data pieces. This allows the inference processing unit to process multiple data pieces simultaneously in parallel, significantly improving throughput while maintaining manageable processing complexity through unified data organization.
Solution Approach 2:
The patent segments the inference processing into two distinct pathways: a configuration data inference pathway for successful inferences and a data-specific inference pathway for failed inferences. This segmentation allows optimized processing strategies for different scenarios, improving overall throughput while managing complexity through specialized processing routes.
2Productivity
If configuration data is generated using multiple data pieces, then parallelism and throughput are improved, but resource consumption increases
Solution Approach 1:
The patent applies local quality by differentiating processing strategies based on inference outcomes. Successful inferences from configuration data are processed using efficient parallel processing, while failed inferences are handled through alternative data-specific processing. This localized optimization ensures resources are allocated efficiently based on actual processing needs rather than uniformly across all data.
Solution Approach 2:
The patent implements feedback mechanisms where the result of configuration data inference is fed back to determine whether additional data-specific inference is needed. This feedback loop allows the system to adjust resource consumption dynamically, avoiding unnecessary processing for successful inferences while ensuring thorough processing for failed ones.
3Reliability
If inference processing uses a single unified approach, then device complexity is low, but reliability of inference results decreases due to inability to handle failures
Solution Approach 1:
The patent introduces dynamic processing complexity based on inference outcomes. The system automatically adjusts the processing pathway dynamically: using simplified configuration data inference for successful cases and switching to more robust data-specific inference for failed cases. This dynamic adaptation improves reliability without permanently increasing device complexity.
Solution Approach 2:
The patent prepares multiple inference pathways in advance, with configuration data inference serving as a first line of defense and data-specific inference as a backup. This beforehand cushioning ensures that if the primary inference approach fails, alternative methods are ready to maintain reliability without requiring complex real-time decision-making.
4Reliability
If multiple inference models are employed to handle failures, then inference reliability improves, but device complexity and model management burden increase
Solution Approach 1:
The patent extracts the failure handling functionality from the main inference process, creating a separate data-specific inference pathway. This extraction isolates model management complexity to a dedicated subsystem, allowing the main configuration data inference to remain simple while ensuring reliable fallback through extracted specialized processing.
Data Source
AI summary
An information processing device according to the present invention includes: a memory configured to store instructions; and one or more processors configured to execute the instructions to: generate configuration data comprising a plurality of sets of data including an inference target; infer a target included in the configuration data; infer a target; and determine, for each set of data constituting the configuration data, whether the inference on the target included in the configuration data is successful or unsuccessful, infer a target included in a set of data for which the inference on the target included in the configuration data is successful and the result of the inference on the target, and infer a target included in a set of data for which the inference on the target is unsuccessful, based on the result of performing inference on a target included in data.