Sequential safe static and dynamic screening rule for accelerating support tensor machine.

Area of Science:

• Machine Learning
• Data Science
• Computational Statistics

Background:

• Support Tensor Machine (STM) is effective for tensorial data classification, preserving structure and mitigating dimensionality.
• Traditional STM methods rely on time-consuming alternating projection iterative techniques.

Purpose of the Study:

• To propose an efficient sequential safe static and dynamic screening rule (SS-SDSR) for accelerating STM.
• To reduce computational cost by identifying and removing redundant variables during training without accuracy loss.

Main Methods:

• Developed static and dynamic screening rules based on variational inequality and duality gap.
• Implemented a sequential screening process using static rules with varying parameters and dynamic rules with consistent parameters.

Main Results:

• Experiments on artificial datasets show SS-SDSR is effective across data forms with appropriate parameter intervals and screening frequencies.
• Numerical experiments on 11 vector and 6 tensor datasets demonstrate SS-SDSR's feasibility and validity compared to five other algorithms.

Conclusions:

• The proposed SS-SDSR effectively accelerates STM training.
• The method is efficient, safe, and demonstrates strong performance on diverse datasets.