TSPINN: Thompson sampling-based adaptive training for physics-informed neural networks

Key Points

• Improved adaptive training enhances efficiency of physics-informed neural networks, yielding better outcomes.
• Thompson sampling shows significant performance gains, with up to 30% efficiency improvement observed in testing.
• Analysis focuses on the integration of adaptive learning and physics-informed methods in neural networks.
• This model implies greater adaptability in learning processes for complex physical systems, paving the way for innovative applications.