Can dictionary-based computational models outperform the best linear ones?

Area of Science:

• Computational Mathematics
• Approximation Theory
• Numerical Analysis

Background:

• Computational models are essential for approximating complex functions.
• Understanding the efficiency of different model types is crucial for selecting appropriate methods.

Purpose of the Study:

• To compare the approximation capabilities of dictionary-based models and linear models.
• To analyze and contrast their approximation rates (speed of error decrease) as the number of basis functions increases.

Main Methods:

• Investigated upper bounds on approximation rates for dictionary-based models.
• Compared these bounds with those of linear models using a fixed set of basis functions.
• Analyzed approximation errors for individual functions and for specific sets of functions.

Main Results:

• For individual functions, dictionary-based models do not offer superior approximation rates compared to some linear models.
• However, for specific sets of functions, dictionary-based models demonstrate the potential for significantly faster approximation rates.
• Geometric upper bounds on approximation errors were established for these specific function sets using dictionary-based models.

Conclusions:

• Dictionary-based models can outperform linear models in approximation speed for certain function classes.
• The choice of model depends on the specific approximation task and the properties of the function set being approximated.