IoT Learning Framework for Activity Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques fail to adequately capture and provide meaningful predictive outputs from Internet of Things (IoT) data, particularly in inferring user activity patterns and distinguishing between IoT device malfunctions and legitimate user activities.
Innovation Solution
The implementation of an IoT learning framework that uses unsupervised learning algorithms to train an AI model, extracting representative user activity patterns from sequences of IoT device events and predicting user activities with high accuracy, even in scenarios with device malfunctions or transient failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are used to infer user activities from IoT device events, then predictive accuracy is improved, but the system becomes more complex and difficult to deploy in resource-constrained IoT environments
Solution Approach 1:
The system segments the machine learning workload by separating model training from model inference. Training is performed on centralized servers with sufficient computational resources, while inference is executed on edge devices with limited resources. This segmentation allows complex models to achieve high accuracy without burdening resource-constrained IoT devices.
Solution Approach 2:
The patent introduces an intermediary layer that collects and pre-processes IoT device events before feeding them to the machine learning model. This intermediary processing layer simplifies the input data structure and reduces computational complexity for the inference stage, enabling accurate predictions on resource-constrained devices.
2Measurement precision
If more IoT device events are collected and processed, then user activity prediction accuracy is improved, but data processing time and computational resources increase
Solution Approach 1:
The system performs preliminary actions by pre-training the machine learning model offline using extensive datasets. This pre-training phase captures complex patterns and relationships in advance, allowing the deployed model to make accurate predictions with minimal real-time processing. The model is prepared beforehand to handle inference queries efficiently.
Solution Approach 2:
The patent extracts and focuses on the most discriminative features and events from the IoT data stream that are most relevant for activity recognition. By selecting only the most informative events and features, the system maintains high prediction accuracy while reducing the volume of data that requires real-time processing.
3Adaptability or versatility
If the machine learning model is trained on diverse IoT device events, then adaptability to varying scenarios is improved, but the training data requirements and model complexity increase
Solution Approach 1:
The patent designs a universal machine learning model architecture that can handle multiple types of IoT devices and event formats through a unified interface. This universal model is trained on diverse datasets from various device types and scenarios, enabling it to generalize across different environments and device configurations without requiring device-specific models.
Solution Approach 2:
The system employs parameter-efficient adaptation techniques where the base model parameters are trained on extensive diverse data, and then fine-tuned or adapted to specific scenarios using smaller datasets. This approach allows the model to maintain high adaptability across varying scenarios while reducing the training data burden for each specific application.
Data Source
AI summary
An Internet-of-Things (IoT) learning framework (IoT learning framework) may train AI models to infer user activities from IoT connected device events using machine learning based algorithms. According to such an example, processing circuitry obtains a training dataset indicating IoT device events and extracts representative user activity patterns from the sequences of IoT device events. In such an example, processing circuitry trains the AI model to learn an optimal subset of the sequences of IoT device events corresponding to a smallest quantity of the sequences of IoT device events to predict user activities with accuracy that satisfies a threshold and outputs the AI model. According to such an example, processing circuitry may obtain new data indicating new sequences of IoT device events and generates output indicating one or more user activities predicted by the AI model to have occurred based on the new sequences of IoT device events.