Researchers from Tsinghua University and other Chinese institutes have developed a new AI framework called PhyE2E that can automatically derive symbolic physical representations from raw data. This innovation aims to help scientists uncover mathematical formulas that describe physical relationships, making it easier for them to test and explore.
PhyE2E was trained on both physical data and established mathematical equations, allowing it to recognize plausible physics-related formulas by synthesizing variations of well-known equations. The model uses a transformer to convert data into symbolic expressions and simplifies complex problems into manageable sub-problems.
The team tested PhyE2E on synthetic data produced by a large language model and real astrophysical data from NASA. The framework successfully derived formulas that explained physical relationships in five different space physics scenarios, matching or surpassing results from human physicists.
This new AI model has the potential to deconstruct complex physics problems into simpler components, drawing from established equations to generate new formulas. Researchers are optimistic about expanding its applicability to analyze experimental and astrophysical data across various fields, leading to discoveries of more accurate physical laws.
The team’s goal is to advance neuro-symbolic methodologies, ensuring that predictions made by deep neural networks are interpretable. By integrating explainability into the design, they hope to boost the AI system’s capacity to uncover scientifically accurate and reliable laws.
Source: https://news.ssbcrack.com/new-ai-framework-automatically-derives-equations-from-space-physics-data