PARTITIONING SIGNAL CLASSES USING TRANSPORT TRANSFORMS FOR DATA ANALYSIS AND MACHINE LEARNING.

Area of Science:

• Applied Mathematics
• Signal Processing
• Data Science

Background:

• Transport-based transforms, including Continuous Wavelet Transform (CDT), Rotational CDT (R-CDT), and Line Integral Transform (LOT), are emerging tools in image and data processing.
• These transforms have demonstrated significant potential in applications like signal estimation, classification, and medical diagnostics (e.g., cancer detection).

Purpose of the Study:

• To investigate the mathematical properties underlying the success of transport-based transforms.
• To identify conditions under which algebraic generative models result in convex sets after transformation.
• To analyze the capabilities and limitations of these transforms in convexifying signal classes within an algebraic generative modeling framework.

Main Methods:

• Mathematical analysis of transport-based transforms (CDT, R-CDT, LOT).
• Investigation of signal classes generated by algebraic models.
• Study of the convexification properties of these transforms.

Main Results:

• Established conditions for transport transforms to map algebraically generated signal classes into convex sets.
• Demonstrated that this convexification simplifies classification and data analysis in the transform domain.
• Quantified the extent and limitations of the convexification ability of these transforms.

Conclusions:

• Transport-based transforms offer a powerful mechanism for simplifying complex data structures.
• The convexification property is key to their effectiveness in various data analysis and signal processing applications.
• Further theoretical and algorithmic research is encouraged to fully leverage these transforms.