Machine Learning User Presence Estimation for Power Mode Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information processing apparatuses struggle to accurately estimate user presence due to environmental factors, leading to incorrect power mode transitions and inefficient energy consumption, as they lack the ability to generate training data sets in real operating environments for optimizing estimation rules.
Innovation Solution
A machine learning system that includes a sensor, a machine learning model, and a user interface to generate training data sets during operation, associating time-series sensor data with user presence labels, enabling the model to learn and adapt to the specific operating environment for precise user presence estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional threshold-based estimation methods are used, then the apparatus can operate with simple estimation rules, but the estimation accuracy deteriorates in varying environmental conditions
Solution Approach 1:
The patent applies dynamics by transitioning from static threshold-based estimation rules to dynamic machine learning models that adapt to varying environmental conditions. The system continuously learns from operational data, adjusting its estimation behavior based on the specific installation environment, noise sources, and user patterns, thereby maintaining high accuracy across diverse conditions.
Solution Approach 2:
The patent changes the parameters of the estimation system by using machine learning models that can automatically adjust their internal parameters (weights, biases, decision boundaries) based on training data collected from the actual operating environment. This allows the system to optimize its estimation rules for each specific installation context rather than using fixed thresholds.
2Reliability
If the apparatus returns from power saving mode based on incorrect estimation, then the apparatus responds to apparent presence signals, but energy consumption increases due to unnecessary mode transitions
Solution Approach 1:
The patent implements feedback by using the actual user presence information (from user operations) to train and refine the machine learning model. The system continuously monitors whether its presence estimates were correct by comparing with actual user interactions, and uses this feedback to improve future estimation accuracy, thereby reducing unnecessary power mode transitions and energy waste.
Solution Approach 2:
The system performs self-service by automatically generating its own training data from its operational environment and using this data to improve its estimation capabilities. The apparatus learns from its own experiences and operational context, adapting to its specific installation environment without requiring manual configuration or external training datasets.
3Adaptability or versatility
If supervised learning is implemented with training data generated during operation, then the model can adapt to the specific operating environment, but the system requires continuous data collection and model training capability
Solution Approach 1:
The system applies self-service by automatically generating training data from its own operational context and performing self-training. The apparatus collects sensor data and user operation data during normal operation, uses this data to train its machine learning model, and continuously improves its estimation accuracy for its specific environment without requiring external intervention or complex manual configuration.
Solution Approach 2:
The patent applies preliminary action by collecting and storing training data during the apparatus operation in advance. The system accumulates sensor data and corresponding user presence labels over time, preparing the training dataset before formal model training occurs, which enables the model to learn from real operational patterns specific to each installation environment.
Data Source
AI summary
A machine learning system includes a sensor configured to sense an object which is present in front of an information processing apparatus, a machine learning model configured to input time-series sensed values output from the sensor and to estimate whether a user who uses the image processing apparatus is present, a user interface configured to receive an operation performed by a user, and a learning unit configured to cause the machine learning model to learn with use of training data including the time-series sensed values output from the sensor and labels that are based on presence and absence of an operation performed by a user and received by the user interface.