Implicit Neural Representation for Magnetotelluric Inversion: EGU 2026

At EGU General Assembly 2026 in Vienna, Central South University Ph.D. student Yanyi Wang presented a novel physics-driven deep learning approach for 2-D magnetotelluric (MT) inversion. This collaborative project with E. Attias (OCEEMlab) and Xavier Garcia (CSIC-ICM) leverages implicit neural representations (INRs) to advance geophysical inversion techniques.

Unlike traditional methods that rely on manually tuned penalty terms such as Tikhonov or Occam regularization to stabilize inversion results, the proposed approach exploits the inherent "frequency principle" of neural networks as an implicit regularization mechanism. This property improves inversion stability and reduces sensitivity to local minima. Synthetic and field-data experiments further demonstrate the method's ability to recover subsurface structures with high resolution and robustness.

Yanyi's participation in session EMRP 2.3 actively contributed to the discussion, providing feedback and insights that will be valuable for the continued development and optimization of the proposed method.

Well done, Yanyi! Excellent work.