Proposal of a diffusion and vision transformer that reconstructs low-temperature Majorana features from high-temperature nanowire conductance alone

1/ AI 활용 방법: Shifted Window U-Net Transformer 기반 확산형 모델과 Video Vision Transformer
2/ 적용 과제: 고온 전도도 데이터로 저온 전도도 복원 및 topological visibility 예측
3/ 보고된 결과: 저온 전도도 복원에서 local R² 약 0.95, nonlocal R² 약 0.91, topological visibility 예측에서 R² 약 0.80을 보고
4/ It was reported that a machine learning method uses nanowire conductance data obtained at high temperature as input to reconstruct low-temperature conductance and directly predict low-temperature topological visibility.
5/ The paper reconstructed low-temperature local and nonlocal conductance with a diffusion-inspired model based on a Shifted Window U-Net Transformer, and predicted topological visibility with a Video Vision Transformer.
6/ The reported performance is local R² 약 0.95 and nonlocal R² 약 0.91 for low-temperature conductance reconstruction, and R² 약 0.80 for topological visibility prediction.
7/ The practical significance for materials research lies in the possibility of screening devices using only high-temperature data before low-temperature validation.
8/ However, these results are reported for a specific nanowire conductance dataset and problem setting, and they cannot be assumed to generalize directly to other material systems or measurement protocols.
9/ In real applications, input data quality, the temperature range, and the model’s extrapolation capability for samples outside the learned distribution must be checked separately. [1][2]