A Novel THz Differential Spectral Clustering Recognition Method Based on t-SNE

Algorithm 1

The pseudocode of t-SNE algorithm.

	Algorithm: t-SNE
	Input data: the sample terahertz spectral data set X = {x1, x2,…, xn}
	Cost function:
	Output the result: low-dimensional spatial data representation Y (t) = {y1, y2,…, yn}
	Optimize training process
	begin
	Set iteration times T, learning rate η, and momentum α (t)
	Calculate the perplexity Perp and the conditional probability according to
	Randomly initialize Y (0) = {y1, y2,…, yn} with a normal N (0, 10–4, I) distribution
	For t = 1 to T, do
	Calculate qij in lower dimensions with formula (5)
	Compute gradient according to formula (8)
	Update
	end
	end
	return Y