Information processing apparatus, information processing method and program
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods for classifying time-series data, such as those using convolutional neural networks (CNN) and recurrent neural networks (RNN), require large amounts of teaching data and are inefficient in processing long sequences, leading to unstable classification results.
Innovation Solution
A learning network architecture utilizing two inception neural networks with different filter sizes processes teaching and query sets to efficiently classify time-series data by updating network parameters to minimize the distance between similar samples and maximize the distance between dissimilar samples, enabling accurate classification with a small amount of teaching data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional classification models (ARIMA, statistical models) are used for time-series data classification, then classification can be performed based on feature extraction, but great effort is required for manual model selection and parameter setting, and accuracy is unstable
Solution Approach 1:
The system automatically learns features from time-series data using machine learning models without requiring manual model selection or parameter setting. The feature learning process is self-service, where the system adapts to the data characteristics automatically, eliminating the need for expert knowledge in model selection while maintaining stable and accurate classification results
2Ease of operation
If machine learning is used to automatically learn features of time-series data, then classification can be performed based on feature similarity, but a large amount of teaching data is needed which makes it practically difficult to realize classification in many cases
Solution Approach 1:
The system uses a limited amount of teaching data (smaller than traditional machine learning requirements) by focusing on learning only the essential features needed for classification. Instead of requiring comprehensive teaching data covering all possible scenarios, the system learns partial features that are sufficient for accurate classification, thereby reducing the data quantity requirement while maintaining automatic feature learning capabilities
3Extent of automation
If convolutional neural networks (CNN) are used for time-series classification, then classification can be performed using convolution layers, but filter sizes are typically fixed and cannot be dynamically changed, and classification results significantly differ depending on design of convolution layers
Solution Approach 1:
The system dynamically adjusts filter sizes in convolution layers based on the characteristics of the input time-series data. Instead of using fixed filter sizes, the convolution layers can adapt their filter sizes during processing, allowing the system to capture features at multiple scales and improving adaptability to different data patterns while maintaining automatic classification processing
4Measurement precision
If recurrent neural networks (RNN) are used for text classification with recursive combination structure, then classification can be performed by focusing on context, but it is difficult to parallelize the entire processing and accurate recognition of context features is difficult when sentences are too long
Solution Approach 1:
The system segments the time-series data into multiple sections or windows and processes each segment independently using parallel computation. This segmentation approach allows the system to maintain context awareness within each segment while enabling parallel processing across segments, thereby improving productivity without sacrificing the ability to recognize context features accurately
Data Source
AI summary
A learning network includes: a first extraction unit that extracts each feature vector of different pieces of teaching data designated in a support set using a first inception neural network that extracts a feature vector of time-series data by convolution layers with different filter sizes; a second extraction unit that extracts a feature vector of teaching data designated in a query set using a second inception neural network that extracts a feature vector of time-series data by a plurality of convolution layers with different filter sizes; and a learning unit that updates a network parameter of the learning network such that a relative distance between feature vectors of teaching data of the query set and of a positive sample of the support set is small and a relative distance between feature vectors of teaching data of the query set and of a negative sample of the support set is large.