Dyformer: A dynamic transformer-based architecture for multivariate time series classification

Keyphrases

• Multivariate Time Series Classification 100%
• Transformer Architecture 100%
• Dynamic Transformer 100%
• Transformer 66%
• Frequency Component 66%
• Distinctive Features 33%
• Training Model 33%
• State-of-the-art Techniques 33%
• Benchmark Dataset 33%
• Time Series Data 33%
• Component-based 33%
• Limited Training Data 33%
• Attention Mechanism 33%
• Subsequence 33%
• Feature Map 33%
• Temporal Dependence 33%
• Time Series Classification 33%
• Generic Approach 33%
• Dynamic Architecture 33%
• Hierarchical Pooling 33%
• Joint Loss Function 33%
• Architectural Constraints 33%
• Adaptive Learning System 33%

Engineering

• Frequency Component 100%
• Limitations 50%
• Tasks 50%
• Joints (Structural Components) 50%
• Multiscale 50%
• Engineering 50%
• State-of-the-Art Method 50%
• Data Series 50%
• Subsequence 50%
• Loss Function 50%
• Inherent Characteristic 50%

Computer Science

• Multivariate Time Series 100%
• Frequency Component 66%
• Training Sample 33%
• Time Series Data 33%
• Baseline Approach 33%
• Training Model 33%
• Adaptive Learning 33%
• Attention (Machine Learning) 33%