Characterizing and optimizing end to end systems for private inference asplos ’23, march 25–29, 2023, vancouver, bc, canada this paper makes the following contributions: (1) the first end to end characterization of private inference using arrival rates. our analysis reveals the key sources of inefficiency with respect to storage. Asplos'23: the 28th international conference on architectural support for programming languages and operating systemssession 7b: software security and priva.
Characterizing and optimizing end to end systems for private inference authors : karthik garimella , zahra ghodsi , nandan kumar jha , siddharth garg , brandon reagen authors info & claims asplos 2023: proceedings of the 28th acm international conference on architectural support for programming languages and operating systems, volume 3. Characterizing and optimizing end to end systems for private inference. karthik garimella, zahra ghodsi, nandan kumar jha, siddharth garg, brandon reagen. in two party machine learning prediction services, the client's goal is to query a remote server's trained machine learning model to perform neural network inference in some application domain. 1. the first end to end characterization of private infer ence using arrival rates. our analysis reveals the key sources of inefficiency with respect to storage, com munication, and computation. 2. a protocol optimization (client garbler) to reduce the significant storage pressure on the client by 5 . 3. identifying a new form of parallelism. Asplos'23: the 28th international conference on architectural support for programming languages and operating systemssession 7b: software security and priva.
1. the first end to end characterization of private infer ence using arrival rates. our analysis reveals the key sources of inefficiency with respect to storage, com munication, and computation. 2. a protocol optimization (client garbler) to reduce the significant storage pressure on the client by 5 . 3. identifying a new form of parallelism. Asplos'23: the 28th international conference on architectural support for programming languages and operating systemssession 7b: software security and priva. Compared to the state of the art pi protocol, these optimizations provide a total pi speedup of 1.8 × with the ability to sustain inference requests up to a 2.24 × greater rate. looking ahead, we conclude our paper with an analysis of future research innovations and their effects and improvements on pi latency. original language. Add to calendar 2023 12 13 16:00:00 2023 12 13 17:00:00 america new york characterizing and optimizing end to end systems for private inference in two party machine learning prediction services, the client's goal is to query a remote server's trained machine learning model to perform neural network inference in some application domain. however.
Comments are closed.