So, apparently "sharp" cos similarity can be as good as convolutions but with 10-100x less compute?
QOTO: Question Others to Teach Ourselves An inclusive, Academic Freedom, instance All cultures welcome. Hate speech and harassment strictly forbidden.