Optimal structure and parameter learning of Ising models.

Area of Science:

• Statistical physics
• Machine learning
• Computational biology
• Image processing

Background:

• Reconstructing Ising models is crucial across various scientific fields.
• Existing methods often require extensive data or are computationally intensive.
• There's a need for efficient algorithms with minimal data requirements.

Purpose of the Study:

• Introduce a novel method, interaction screening, for Ising model reconstruction.
• Achieve accurate parameter estimation and graph structure recovery.
• Develop a computationally efficient and data-minimal solution to the inverse Ising problem.

Main Methods:

• Utilizes local optimization problems for parameter estimation.
• Employs interaction screening for accurate model reconstruction.
• Analyzes performance on synthetic and real-world D-Wave quantum computer data.

Main Results:

• Interaction screening provably achieves perfect graph structure recovery.
• The method is information-theoretically optimal in sample complexity.
• Demonstrates efficacy in the difficult low-temperature regime.

Conclusions:

• Interaction screening is an exact, tractable, and optimal solution for the inverse Ising problem.
• The method offers a universal approach applicable to diverse Ising models.
• This technique significantly advances the field of model reconstruction.