Uncertainty Quantification
MC DROPOUTN=20 stochastic forward passes with dropout active at inference. Predictive entropy H[p] = โฮฃ pยทlog(p) per protein.
MC Dropout Predictive Entropy
H[p] = โฮฃc pฬc ยท log(pฬc)
pฬc = (1/N) ฮฃn=1N pc(x; ฮธ, ฯn) ย ยทย N=20, p_dropout=0.1
High entropy (H > 0.5) โ model uncertain โ protein excluded from top-confidence screening rankings
Proteins Analyzed
0
Mean Entropy
0.00000
High Uncertainty (>0.5)
0
Low Uncertainty (โค0.5)
0