AI Model Switching for IoT Data Processing Chains
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the IoT field, connected objects often experience low data transmission due to communication and hardware constraints, leading to insufficient training of artificial intelligence models, which results in unsatisfactory performance and an inability to predict temporal fluctuations.
Innovation Solution
A method is implemented to configure a data processing chain by determining an input signature of the data stream, computing similarity scores between the input signature and current/trained signatures, and switching to auxiliary AI models with better similarity scores when deviations occur, ensuring the processing chain's performance is maintained despite low data abundance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If connected objects transmit data under severe constraints (low data abundance), then energy consumption and communication overhead are reduced, but the artificial intelligence model becomes insufficiently trained and performance deteriorates
Solution Approach 1:
Multiple auxiliary AI models are pre-trained on different auxiliary training datasets with varying properties before deployment. When data deviation is detected, the system can immediately switch to a pre-prepared model that is already adapted to the current data characteristics, avoiding the need for real-time retraining and enabling reliable performance despite low data abundance.
Solution Approach 2:
The system changes the parameter of model selection based on data signature similarity. By computing similarity scores between the input data signature and training dataset signatures, the system dynamically selects which pre-trained model to use, adapting to different data conditions without requiring additional training data or energy.
2Loss of information
If connected objects transmit sparse data to reduce communication overhead, then network bandwidth usage is reduced, but the completeness of data decreases and temporal fluctuations cannot be predicted
Solution Approach 1:
Multiple auxiliary AI models are pre-trained on different auxiliary training datasets that capture various data patterns and temporal fluctuations. This preliminary preparation ensures that even when incoming data is sparse, the system has pre-configured models that can predict temporal fluctuations based on the limited data available.
Solution Approach 2:
The system continuously monitors the similarity score between input data signature and training dataset signatures. When the similarity score falls outside the acceptable range, indicating data deviation, the system provides feedback by switching to an appropriate auxiliary model, thereby maintaining prediction capability despite incomplete data.
3Device complexity
If a single artificial intelligence model is used for data processing, then device complexity is reduced, but the system cannot adapt to deviations in data properties and performance becomes unsatisfactory
Solution Approach 1:
The system dynamically selects which AI model to use based on real-time similarity score calculations. Instead of a static single-model approach, the system adapts its model configuration by switching between the current model and multiple auxiliary models depending on the characteristics of the incoming data, achieving adaptability without significant increases in operational complexity.
Solution Approach 2:
Multiple auxiliary AI models are trained on different auxiliary training datasets with varying properties, making each model specialized for different data conditions. This multi-functionality allows the system to handle diverse data scenarios using a set of pre-trained models, improving adaptability while keeping each individual model relatively simple.
Data Source
AI summary
The invention relates to a method for configuring a data processing chain (4) comprising a computing stage (10), the method comprising the steps ofdetermining an input signature of an input data stream (6);computing a current similarity score between the input signature and a current signature associated with a training dataset of a current artificial intelligence model (12) implemented by the computing stage (10);if the computed current similarity score is outside a predetermined acceptable range:for each of at least one auxiliary artificial intelligence model (16), computing a corresponding auxiliary similarity score between the input signature and an auxiliary signature of an associated auxiliary training dataset;configuring the computing stage (10) so as to implement the auxiliary artificial intelligence model (16) associated with the auxiliary signature that has the best auxiliary similarity score.