Distributed Inference Model Adaptation via Local Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing devices face limitations in accurately identifying features in unlabeled data due to limited labeled data training and changing conditions, leading to reduced accuracy over time, especially in scenarios with low computational resources like autonomous vehicles.
Innovation Solution
A system that periodically retracts inference models using a combination of labeled and unlabeled data, where the information handling system performs calculations on unlabeled data and sends only the results to the information handling system manager, maintaining privacy and reducing communication resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If inference models are trained only with limited labeled data, then device complexity is reduced, but measurement precision deteriorates over time due to changing conditions
Solution Approach 1:
The patent segments the model adaptation process into two parts: (1) initial training with limited labeled data to create a base model, and (2) distributed continuous adaptation where multiple information handling systems contribute unlabeled data locally without centralizing it. This segmentation allows the system to maintain low complexity while improving precision through distributed learning from segmented data sources across multiple systems.
Solution Approach 2:
The patent introduces an intermediary mechanism where information handling systems perform local calculations on unlabeled data and exchange only calculation results (not raw data) with the model provider. This intermediary approach enables precision improvement through additional data while maintaining privacy and reducing communication overhead, resolving the contradiction between model accuracy and system complexity.
2Measurement precision
If unlabeled data is collected and processed locally, then measurement precision is improved, but loss of information increases due to data privacy requirements
Solution Approach 1:
The patent extracts only the essential computational results from local unlabeled data processing while leaving the raw sensitive data localized. Information handling systems perform calculations on their local unlabeled data and extract only the necessary statistical or feature information to send to the model provider, thereby improving measurement precision without exposing or transmitting sensitive unlabeled data that would cause information loss due to privacy concerns.
3Measurement precision
If full model retraining is performed centrally, then measurement precision is improved, but use of energy increases due to data transmission requirements
Solution Approach 1:
The patent implements partial action by having information handling systems perform only the necessary local calculations on unlabeled data and transmit only the essential results rather than complete datasets. This partial approach (performing only critical computations locally and transmitting minimal information) achieves model accuracy improvement while significantly reducing communication energy consumption compared to full data transmission for centralized retraining.
Data Source
AI summary
An information handling system includes storage and a processor. The processor identifies an occurrence of an inference model update event; in response to identifying the inference model update event: generates an inference model update package; provides the inference model update package to an entity that generated an inference model used by the information handling system; obtains, from the entity, a hybrid data adapted inference model that is based on the inference model, the inference model update package, and labeled data used to train the inference model; and obtains an inference, using the hybrid data adapted inference model, that indicates a feature is present in collected data.