Algorithm 2.

Active learning for predicting decision boundary.

Input: DiffusionModel, two d-dimensional points p1,p2

Output: evaluatedPoints, boundary Points

1:	procedure LearnDecisionBoundary
2:	Pick the minimum and maximum from the range of p1 and p2
3:	start = [p1Min,p2Min]
4:	end = [p1Max,p2Max]
5:	[bPt, bMean, bStdev] = BinarySearch(start, end)
6:	EvaluateNearbyPoints(bPt) Add these points to evaluatedPoints
7:	r=1
8:	while r <= K do  ▷ K is the budget on total number of active learning runs
9:	nPt = GetNextPointViaUncertaintySampling
10:	[nPt, nMean, nStdev] =RunDiffusionModel(nPt)
11:	if nStdev >= θsd then >= θsd is the standard deviation threshold
12:	Add nPt to boundaryPoints
13:	else
14:	Add [nPt, nMean, nLabel] to evaluatedPoints
15:	end if
16:	oppPt = A point with label opposite to nLabel and farthest (L2 norm) from nPt
17:	[bPt, bMean, bStdev] = BinarySearch(nPt, oppPt)
18:	EvaluateNearbyPoints(bPt) Add these points to evaluatedPoints
19:	r + +
20:	end while
21:	return evaluatedPoints, boundaryPoints
22:	end procedure