Eh, I work in audio ml, where most of the interesting and useful work is in conditioned generative models. Compression, TTS, source separation. The bias towards

Eh, I work in audio ml, where most of the interesting and useful work is in conditioned generative models. Compression, TTS, source separation. The bias towards classification is a side effect of people starting at imagenet too long; I really think it's holding the field back.