10^5 speed up will get them to 10x slower than plaintext processing of deep neural networks. See page 15 of DARPA's BAA: https://beta.sam.gov/api/prod/opps/v3/

10^5 speed up will get them to 10x slower than plaintext processing of deep neural networks. See page 15 of DARPA's BAA: https://beta.sam.gov/api/prod/opps/v3/opportunities/resources/files/50030b64c04a4fe3ab289ad67c61f21e/download?api_key=null&status=archived&token= https://beta.sam.gov/api/prod/opps/v3/opportunities/resource... For simple computing like logistic regression, they aim to be as fast as plaintext.