Homomorphic encryption (HE)-based deep neural network (DNN) inference protects data and model privacy but suffers from significant computation overhead. We observe transforming the DNN weights into circulant matrices converts general matrix-vector multiplications into HE-friendly 1-dimensional convolutions, drastically reducing the HE computation cost. Hence, in this paper, we propose PrivCirNet, a protocol/network co-optimization framework based on block circulant transformation. At the protocol level, PrivCirNet customizes the HE encoding algorithm that is fully compatible with the block circulant transformation and reduces the computation latency in proportion to the block size. At the network level, we propose a latency-aware formulation to search for the layer-wise block size assignment based on second-order information.

Reports surfacing today indicate that General Motors will build and test several thousands of self-driving Chevrolet Bolts starting at the beginning of next year, according to Reuters. The American automaker will partner with Lyft Inc--the ride-sharing company in which GM has invested $500 million--to deploy the autonomous Bolts in Lyft's ride-sharing fleets. Even though GM has no immediate plans to sell self-driving Bolts to consumers quite yet, Lyft will reportedly be testing these specially-prepped Bolts across several different states. Over the past year, GM execs have mentioned the American automaker would manufacture and deploy autonomous vehicles in ride-sharing fleets, but never released when this would actually happen and with whom as a partner, even though it was assumed it would be Lyft, given GM's investment. In response to today's news from Reuters, though, GM has released a statement that hints at at some sort of timeline.