Algorithm 2 GPR-VLP under Data-Efficient Training with Quadrant-Based Active Learning (Q-AL-GPR)

Input/output and all steps are the same as Algorithm 1 except for the initialization step in Active Learning. * Active Learning:   Initialization: $\mathrm{Divide}\mathcal{D}$$\mathrm{into}\mathrm{an}\mathrm{initial}\mathrm{effective}\mathrm{training}\mathrm{dataset}{\mathcal{D}}_e^0(\left|{\mathcal{D}}_e^0\right|=m)$$\mathrm{and}\mathrm{the}\mathrm{corresponding}\mathrm{complement}\mathrm{dataset}{\mathcal{D}}_{\ast }^0\left(\left|{\mathcal{D}}_e^0\right|=\left|\mathcal{D}\right|-m\right).$$\mathrm{For}\mathrm{each}\mathrm{of}\mathrm{the}\mathrm{four}\mathrm{quadrants}\mathrm{of}\mathrm{the}\mathrm{test}\mathrm{space},⌊m/4⌋$$\mathrm{data}\mathrm{are}\mathrm{randomly}\mathrm{drawn}\mathrm{according}\mathrm{to}\mathrm{their}\mathrm{labels}\mathrm{to}\mathrm{build}{\mathcal{D}}_e^0$.   Other steps: The same as Algorithm 1. Positioning Stage: The same as Algorithm 1.