Matches in SemOpenAlex for { <https://semopenalex.org/work/W4283696039> ?p ?o ?g. }
Showing items 1 to 75 of
75
with 100 items per page.
- W4283696039 abstract "In this work, we provide an in-depth characterization study of the performance overhead for running Transformer models with secure multi-party computation (MPC). MPC is a cryptographic framework for protecting both the model and input data privacy in the presence of untrusted compute nodes. Our characterization study shows that Transformers introduce several performance challenges for MPC-based private machine learning inference. First, Transformers rely extensively on “softmax” functions. While softmax functions are relatively cheap in a non-private execution, softmax dominates the MPC inference runtime, consuming up to 50% of the total inference runtime. Further investigation shows that computing the maximum, needed for providing numerical stability to softmax, is a key culprit for the increase in latency. Second, MPC relies on approximating non-linear functions that are part of the softmax computations, and the narrow dynamic ranges make optimizing softmax while maintaining accuracy quite difficult. Finally, unlike CNNs, Transformer-based NLP models use large embedding tables to convert input words into embedding vectors. Accesses to these embedding tables can disclose inputs; hence, additional obfuscation for embedding access patterns is required for guaranteeing the input privacy. One approach to hide address accesses is to convert an embedding table lookup into a matrix multiplication. However, this naive approach increases MPC inference runtime significantly. We then apply tensor-train (TT) decomposition, a lossy compression technique for representing embedding tables, and evaluate its performance on embedding lookups. We show the trade-off between performance improvements and the corresponding impact on model accuracy using detailed experiments." @default.
- W4283696039 created "2022-06-30" @default.
- W4283696039 creator A5000913628 @default.
- W4283696039 creator A5011904691 @default.
- W4283696039 creator A5018033573 @default.
- W4283696039 creator A5024329178 @default.
- W4283696039 creator A5035740553 @default.
- W4283696039 creator A5056882178 @default.
- W4283696039 creator A5061848053 @default.
- W4283696039 date "2022-05-01" @default.
- W4283696039 modified "2023-10-01" @default.
- W4283696039 title "Characterization of MPC-based Private Inference for Transformer-based Models" @default.
- W4283696039 cites W2010243335 @default.
- W4283696039 cites W2108255910 @default.
- W4283696039 cites W2791827614 @default.
- W4283696039 cites W3016063723 @default.
- W4283696039 cites W3035390927 @default.
- W4283696039 cites W3106501308 @default.
- W4283696039 cites W3106542468 @default.
- W4283696039 cites W3155184874 @default.
- W4283696039 doi "https://doi.org/10.1109/ispass55109.2022.00025" @default.
- W4283696039 hasPublicationYear "2022" @default.
- W4283696039 type Work @default.
- W4283696039 citedByCount "1" @default.
- W4283696039 countsByYear W42836960392023 @default.
- W4283696039 crossrefType "proceedings-article" @default.
- W4283696039 hasAuthorship W4283696039A5000913628 @default.
- W4283696039 hasAuthorship W4283696039A5011904691 @default.
- W4283696039 hasAuthorship W4283696039A5018033573 @default.
- W4283696039 hasAuthorship W4283696039A5024329178 @default.
- W4283696039 hasAuthorship W4283696039A5035740553 @default.
- W4283696039 hasAuthorship W4283696039A5056882178 @default.
- W4283696039 hasAuthorship W4283696039A5061848053 @default.
- W4283696039 hasConcept C113775141 @default.
- W4283696039 hasConcept C11413529 @default.
- W4283696039 hasConcept C121332964 @default.
- W4283696039 hasConcept C154945302 @default.
- W4283696039 hasConcept C165801399 @default.
- W4283696039 hasConcept C188441871 @default.
- W4283696039 hasConcept C2776214188 @default.
- W4283696039 hasConcept C41008148 @default.
- W4283696039 hasConcept C41608201 @default.
- W4283696039 hasConcept C50644808 @default.
- W4283696039 hasConcept C62520636 @default.
- W4283696039 hasConcept C66322947 @default.
- W4283696039 hasConcept C80444323 @default.
- W4283696039 hasConceptScore W4283696039C113775141 @default.
- W4283696039 hasConceptScore W4283696039C11413529 @default.
- W4283696039 hasConceptScore W4283696039C121332964 @default.
- W4283696039 hasConceptScore W4283696039C154945302 @default.
- W4283696039 hasConceptScore W4283696039C165801399 @default.
- W4283696039 hasConceptScore W4283696039C188441871 @default.
- W4283696039 hasConceptScore W4283696039C2776214188 @default.
- W4283696039 hasConceptScore W4283696039C41008148 @default.
- W4283696039 hasConceptScore W4283696039C41608201 @default.
- W4283696039 hasConceptScore W4283696039C50644808 @default.
- W4283696039 hasConceptScore W4283696039C62520636 @default.
- W4283696039 hasConceptScore W4283696039C66322947 @default.
- W4283696039 hasConceptScore W4283696039C80444323 @default.
- W4283696039 hasLocation W42836960391 @default.
- W4283696039 hasOpenAccess W4283696039 @default.
- W4283696039 hasPrimaryLocation W42836960391 @default.
- W4283696039 hasRelatedWork W2612900438 @default.
- W4283696039 hasRelatedWork W2964140943 @default.
- W4283696039 hasRelatedWork W3137022369 @default.
- W4283696039 hasRelatedWork W3146091044 @default.
- W4283696039 hasRelatedWork W3197644873 @default.
- W4283696039 hasRelatedWork W3203166921 @default.
- W4283696039 hasRelatedWork W3205431227 @default.
- W4283696039 hasRelatedWork W3207862512 @default.
- W4283696039 hasRelatedWork W4287241953 @default.
- W4283696039 hasRelatedWork W4307933444 @default.
- W4283696039 isParatext "false" @default.
- W4283696039 isRetracted "false" @default.
- W4283696039 workType "article" @default.