Comment by nialse

Limited intuitive interpretability of phase likely restricts the broader use of discrete Fourier transforms in machine learning. Frequency, time, and amplitude are tangible and intuitive concepts, whereas phase often feels awkward and less accessible. Using a power spectrum is common practice, but it comes at the cost of losing precision.