Cryptographic protocols for secure multi-
party computation in distributed
systems
Introduction
Distributed systems spanning multiple organizations handle increasingly
sensitive data that needs joint analysis and insights while maintaining
privacy. Secure multi-party computation (MPC) enables this by allowing
distributed parties to compute functions over their private inputs without
revealing anything beyond the outputs. MPC finds applications in finance,
healthcare, supply chains and other sectors involving collaborative data
analytics across borders. This paper discusses cryptographic protocols
underlying MPC and reviews representative techniques employed to ensure
privacy and security in distributed systems.
Privacy Requirements for MPC
Key privacy requirements for MPC protocols include:
- Input privacy: No party should learn anything about others’ private
inputs beyond what can be inferred from their own inputs and the
function output.
- Output privacy: For functions like comparisons, the output too should
reveal minimum information.
- Semi-honest and malicious security: Protocols should be secure under
passive attacks by semi-honest adversaries following the protocol, and
active attacks by malicious adversaries.
- Collusion resistance: No set of colluding parties should gain privacy
breach beyond individual capabilities.
- Circular security: Even if intermediate values are revealed in multi-
round protocols, the final inputs should remain private.
Protocols aim to achieve these through secure computation using
cryptographic constructions that simulate a “trusted third party”
functionality in a decentralized model.
Yao’s Garbled Circuits
Yao’s protocol is one of the earliest and most practical MPC techniques. It
works by garbling or encrypting a boolean circuit representation of the
function using free-XOR technique and oblivious transfer.
- Circuit encryption uses random keys so no information about
inputs/outputs leak beyond what can be computed from output.
- Inputs are mapped to keys which participant obtains through 1-2 OT to
keep inputs private.
- Evaluation reveals only keys and not the original inputs. Output
consists of keys decrypting to result.
Optimizations include fixed-key garbling, row reduction, halfgates etc. It
provides security against passive attacks and has been deployed for
applications like private set intersection.
GMW and BGW Protocols
The GMW protocol by Goldreich-Micali-Wigderson was one of the first general
MPC protocols and introduced the idea of secret sharing. Inputs are secret
shared among parties who locally perform operations and reconstruct output
shares.
Ben-Or, Goldwasser and Wigderson (BGW) protocol concurrently introduced
verifiable secret sharing addressing active attacks, by requiring parties to
prove they followed protocol using zero-knowledge proofs over secret shares.
- Inputs are shared additively/multiplicatively among parties who
perform local computations.
- Multiplication needs multiplication triples generated during pre-
processing.
- Reconstruction reveals output but not individual inputs under
appropriate threshold.
Variants like SPDZ provide improved efficiency and fault tolerance for
arithmetic circuits. They work for both honest-majority and dishonest-
majority settings but have higher overhead than garbled circuits.
Oblivious Transfer Extensions
Oblivious transfer (OT) forms a core building block in most MPC protocols to
transfer inputs privately. The cost of cryptographic OT scales poorly, so OT
extensions are used as optimizations.
- Parties engage in series of OT calls using short authentication strings
which are correlated to subsequent transactions, rather than fresh keys
each time.
- Only the first interaction uses public-key based OT primitives like Rabin
OT or RSA OT while others exploit the correlations.
OT-extensions including the NUOT and RandOT protocols significantly
improve OT costs and enable practical secure computation for large
domains. Network coding variants like OT-combine further optimize
bandwidth usage.
Secure Computation for Machine Learning
MPC enables distributed training of machine learning models like neural
networks, SVM over privacy-sensitive datasets across organizations:
- Secret shares of inputs are used with techniques like Yao’s garbled
circuits and SPDZ-like multiplication triples to evaluate activations
layer-by-layer privately.
- Gradients are computed locally on shares and aggregated. Updates are
applied using MPC primitives.
- FlexSC and ABY frameworks enable training commonly used ML models
with low latency compared to general MPC techniques.
Differentially private approaches incorporate calibrated noise to enable
model analytics while provably preventing membership inference attacks.
Blockchain techniques may also be integrated for immutable audit logs
during distributed ML computations.
Applications and Real-world Deployments
Some real-world uses of MPC protocols for data analytics across industries
and organizations:
- Banks conduct credit scores and loan approvals on joint customer
database without data sharing in SecureDec.
- Hospitals collaborate on genomic studies and clinical trials via private
record matching in PSI.
- Government agencies run analytics on covid spread and contact
tracing preserving citizen privacy.
- Enterprises jointly optimize supply chains minimizing customer/product
privacy risks using MPC supply chain management platforms.
- Partners jointly optimize Internet ad auction by running second price
auction on private bids with techniques from P4P.
MPC deployments address privacy mandates while unlocking business value
from distributed analytics at global scales. Standardization efforts further
drive adoption in privacy-critical applications.
Conclusion
Cryptographic protocols like Yao’s garbled circuits, GMW, BGW and their
variants provide secure foundations for MPC enabling distributed parties to
jointly compute over private inputs. Optimizations including OT extensions
and protocols tailored for machine learning scale these techniques to real-
world deployments handling massive volumes of sensitive data across
organizational boundaries. MPC addresses privacy challenges through its
provable security definitions, heralding a new paradigm of collaborative
analytics respecting information ownership.
Distributed systems spanning multiple organizations handle increasingly
sensitive data that needs joint analysis and insights while maintaining
privacy. Secure multi-party computation (MPC) enables this by allowing
distributed parties to compute functions over their private inputs without
revealing anything beyond the outputs. MPC finds applications in finance,
healthcare, supply chains and other sectors involving collaborative data
analytics across borders. This paper discusses cryptographic protocols
underlying MPC and reviews representative techniques employed to ensure
privacy and security in distributed systems.
Privacy Requirements for MPC
Key privacy requirements for MPC protocols include:
- Input privacy: No party should learn anything about others’ private
inputs beyond what can be inferred from their own inputs and the
function output.
- Output privacy: For functions like comparisons, the output too should
reveal minimum information.
- Semi-honest and malicious security: Protocols should be secure under
passive attacks by semi-honest adversaries following the protocol, and
active attacks by malicious adversaries.
- Collusion resistance: No set of colluding parties should gain privacy
breach beyond individual capabilities.
- Circular security: Even if intermediate values are revealed in multi-
round protocols, the final inputs should remain private.
Protocols aim to achieve these through secure computation using
cryptographic constructions that simulate a “trusted third party”
functionality in a decentralized model.
Yao’s Garbled Circuits
Yao’s protocol is one of the earliest and most practical MPC techniques. It
works by garbling or encrypting a boolean circuit representation of the
function using free-XOR technique and oblivious transfer.
- Circuit encryption uses random keys so no information about
inputs/outputs leak beyond what can be computed from output.
- Inputs are mapped to keys which participant obtains through 1-2 OT to
keep inputs private.
- Evaluation reveals only keys and not the original inputs. Output
consists of keys decrypting to result.
Optimizations include fixed-key garbling, row reduction, halfgates etc. It
provides security against passive attacks and has been deployed for
applications like private set intersection.
GMW and BGW Protocols
The GMW protocol by Goldreich-Micali-Wigderson was one of the first general
MPC protocols and introduced the idea of secret sharing. Inputs are secret
shared among parties who locally perform operations and reconstruct output
shares.
Ben-Or, Goldwasser and Wigderson (BGW) protocol concurrently introduced
verifiable secret sharing addressing active attacks, by requiring parties to
prove they followed protocol using zero-knowledge proofs over secret shares.
- Inputs are shared additively/multiplicatively among parties who
perform local computations.
- Multiplication needs multiplication triples generated during pre-
processing.
- Reconstruction reveals output but not individual inputs under
appropriate threshold.
Variants like SPDZ provide improved efficiency and fault tolerance for
arithmetic circuits. They work for both honest-majority and dishonest-
majority settings but have higher overhead than garbled circuits.
Oblivious Transfer Extensions
Oblivious transfer (OT) forms a core building block in most MPC protocols to
transfer inputs privately. The cost of cryptographic OT scales poorly, so OT
extensions are used as optimizations.
- Parties engage in series of OT calls using short authentication strings
which are correlated to subsequent transactions, rather than fresh keys
each time.
- Only the first interaction uses public-key based OT primitives like Rabin
OT or RSA OT while others exploit the correlations.
OT-extensions including the NUOT and RandOT protocols significantly
improve OT costs and enable practical secure computation for large
domains. Network coding variants like OT-combine further optimize
bandwidth usage.
Secure Computation for Machine Learning
MPC enables distributed training of machine learning models like neural
networks, SVM over privacy-sensitive datasets across organizations:
- Secret shares of inputs are used with techniques like Yao’s garbled
circuits and SPDZ-like multiplication triples to evaluate activations
layer-by-layer privately.
- Gradients are computed locally on shares and aggregated. Updates are
applied using MPC primitives.
- FlexSC and ABY frameworks enable training commonly used ML models
with low latency compared to general MPC techniques.
Differentially private approaches incorporate calibrated noise to enable
model analytics while provably preventing membership inference attacks.
Blockchain techniques may also be integrated for immutable audit logs
during distributed ML computations.
Applications and Real-world Deployments
Some real-world uses of MPC protocols for data analytics across industries
and organizations:
- Banks conduct credit scores and loan approvals on joint customer
database without data sharing in SecureDec.
- Hospitals collaborate on genomic studies and clinical trials via private
record matching in PSI.
- Government agencies run analytics on covid spread and contact
tracing preserving citizen privacy.
- Enterprises jointly optimize supply chains minimizing customer/product
privacy risks using MPC supply chain management platforms.
- Partners jointly optimize Internet ad auction by running second price
auction on private bids with techniques from P4P.
MPC deployments address privacy mandates while unlocking business value
from distributed analytics at global scales. Standardization efforts further
drive adoption in privacy-critical applications.
Conclusion
Cryptographic protocols like Yao’s garbled circuits, GMW, BGW and their
variants provide secure foundations for MPC enabling distributed parties to
jointly compute over private inputs. Optimizations including OT extensions
and protocols tailored for machine learning scale these techniques to real-
world deployments handling massive volumes of sensitive data across
organizational boundaries. MPC addresses privacy challenges through its
provable security definitions, heralding a new paradigm of collaborative
analytics respecting information ownership.
Distributed systems spanning multiple organizations handle increasingly
sensitive data that needs joint analysis and insights while maintaining
privacy. Secure multi-party computation (MPC) enables this by allowing
distributed parties to compute functions over their private inputs without
revealing anything beyond the outputs. MPC finds applications in finance,
healthcare, supply chains and other sectors involving collaborative data
analytics across borders. This paper discusses cryptographic protocols
underlying MPC and reviews representative techniques employed to ensure
privacy and security in distributed systems.
Privacy Requirements for MPC
Key privacy requirements for MPC protocols include:
- Input privacy: No party should learn anything about others’ private
inputs beyond what can be inferred from their own inputs and the
function output.
- Output privacy: For functions like comparisons, the output too should
reveal minimum information.
- Semi-honest and malicious security: Protocols should be secure under
passive attacks by semi-honest adversaries following the protocol, and
active attacks by malicious adversaries.
- Collusion resistance: No set of colluding parties should gain privacy
breach beyond individual capabilities.
- Circular security: Even if intermediate values are revealed in multi-
round protocols, the final inputs should remain private.
Protocols aim to achieve these through secure computation using
cryptographic constructions that simulate a “trusted third party”
functionality in a decentralized model.
Yao’s Garbled Circuits
Yao’s protocol is one of the earliest and most practical MPC techniques. It
works by garbling or encrypting a boolean circuit representation of the
function using free-XOR technique and oblivious transfer.
- Circuit encryption uses random keys so no information about
inputs/outputs leak beyond what can be computed from output.
- Inputs are mapped to keys which participant obtains through 1-2 OT to
keep inputs private.
- Evaluation reveals only keys and not the original inputs. Output
consists of keys decrypting to result.
Optimizations include fixed-key garbling, row reduction, halfgates etc. It
provides security against passive attacks and has been deployed for
applications like private set intersection.
GMW and BGW Protocols
The GMW protocol by Goldreich-Micali-Wigderson was one of the first general
MPC protocols and introduced the idea of secret sharing. Inputs are secret
shared among parties who locally perform operations and reconstruct output
shares.
Ben-Or, Goldwasser and Wigderson (BGW) protocol concurrently introduced
verifiable secret sharing addressing active attacks, by requiring parties to
prove they followed protocol using zero-knowledge proofs over secret shares.
- Inputs are shared additively/multiplicatively among parties who
perform local computations.
- Multiplication needs multiplication triples generated during pre-
processing.
- Reconstruction reveals output but not individual inputs under
appropriate threshold.
Variants like SPDZ provide improved efficiency and fault tolerance for
arithmetic circuits. They work for both honest-majority and dishonest-
majority settings but have higher overhead than garbled circuits.
Oblivious Transfer Extensions
Oblivious transfer (OT) forms a core building block in most MPC protocols to
transfer inputs privately. The cost of cryptographic OT scales poorly, so OT
extensions are used as optimizations.
- Parties engage in series of OT calls using short authentication strings
which are correlated to subsequent transactions, rather than fresh keys
each time.
- Only the first interaction uses public-key based OT primitives like Rabin
OT or RSA OT while others exploit the correlations.
OT-extensions including the NUOT and RandOT protocols significantly
improve OT costs and enable practical secure computation for large
domains. Network coding variants like OT-combine further optimize
bandwidth usage.
Secure Computation for Machine Learning
MPC enables distributed training of machine learning models like neural
networks, SVM over privacy-sensitive datasets across organizations:
- Secret shares of inputs are used with techniques like Yao’s garbled
circuits and SPDZ-like multiplication triples to evaluate activations
layer-by-layer privately.
- Gradients are computed locally on shares and aggregated. Updates are
applied using MPC primitives.
- FlexSC and ABY frameworks enable training commonly used ML models
with low latency compared to general MPC techniques.
Differentially private approaches incorporate calibrated noise to enable
model analytics while provably preventing membership inference attacks.
Blockchain techniques may also be integrated for immutable audit logs
during distributed ML computations.
Applications and Real-world Deployments
Some real-world uses of MPC protocols for data analytics across industries
and organizations:
- Banks conduct credit scores and loan approvals on joint customer
database without data sharing in SecureDec.
- Hospitals collaborate on genomic studies and clinical trials via private
record matching in PSI.
- Government agencies run analytics on covid spread and contact
tracing preserving citizen privacy.
- Enterprises jointly optimize supply chains minimizing customer/product
privacy risks using MPC supply chain management platforms.
- Partners jointly optimize Internet ad auction by running second price
auction on private bids with techniques from P4P.
MPC deployments address privacy mandates while unlocking business value
from distributed analytics at global scales. Standardization efforts further
drive adoption in privacy-critical applications.
Conclusion
Cryptographic protocols like Yao’s garbled circuits, GMW, BGW and their
variants provide secure foundations for MPC enabling distributed parties to
jointly compute over private inputs. Optimizations including OT extensions
and protocols tailored for machine learning scale these techniques to real-
world deployments handling massive volumes of sensitive data across
organizational boundaries. MPC addresses privacy challenges through its
provable security definitions, heralding a new paradigm of collaborative
analytics respecting information ownership.
Distributed systems spanning multiple organizations handle increasingly
sensitive data that needs joint analysis and insights while maintaining
privacy. Secure multi-party computation (MPC) enables this by allowing
distributed parties to compute functions over their private inputs without
revealing anything beyond the outputs. MPC finds applications in finance,
healthcare, supply chains and other sectors involving collaborative data
analytics across borders. This paper discusses cryptographic protocols
underlying MPC and reviews representative techniques employed to ensure
privacy and security in distributed systems.
Privacy Requirements for MPC
Key privacy requirements for MPC protocols include:
- Input privacy: No party should learn anything about others’ private
inputs beyond what can be inferred from their own inputs and the
function output.
- Output privacy: For functions like comparisons, the output too should
reveal minimum information.
- Semi-honest and malicious security: Protocols should be secure under
passive attacks by semi-honest adversaries following the protocol, and
active attacks by malicious adversaries.
- Collusion resistance: No set of colluding parties should gain privacy
breach beyond individual capabilities.
- Circular security: Even if intermediate values are revealed in multi-
round protocols, the final inputs should remain private.
Protocols aim to achieve these through secure computation using
cryptographic constructions that simulate a “trusted third party”
functionality in a decentralized model.
Yao’s Garbled Circuits
Yao’s protocol is one of the earliest and most practical MPC techniques. It
works by garbling or encrypting a boolean circuit representation of the
function using free-XOR technique and oblivious transfer.
- Circuit encryption uses random keys so no information about
inputs/outputs leak beyond what can be computed from output.
- Inputs are mapped to keys which participant obtains through 1-2 OT to
keep inputs private.
- Evaluation reveals only keys and not the original inputs. Output
consists of keys decrypting to result.
Optimizations include fixed-key garbling, row reduction, halfgates etc. It
provides security against passive attacks and has been deployed for
applications like private set intersection.
GMW and BGW Protocols
The GMW protocol by Goldreich-Micali-Wigderson was one of the first general
MPC protocols and introduced the idea of secret sharing. Inputs are secret
shared among parties who locally perform operations and reconstruct output
shares.
Ben-Or, Goldwasser and Wigderson (BGW) protocol concurrently introduced
verifiable secret sharing addressing active attacks, by requiring parties to
prove they followed protocol using zero-knowledge proofs over secret shares.
- Inputs are shared additively/multiplicatively among parties who
perform local computations.
- Multiplication needs multiplication triples generated during pre-
processing.
- Reconstruction reveals output but not individual inputs under
appropriate threshold.
Variants like SPDZ provide improved efficiency and fault tolerance for
arithmetic circuits. They work for both honest-majority and dishonest-
majority settings but have higher overhead than garbled circuits.
Oblivious Transfer Extensions
Oblivious transfer (OT) forms a core building block in most MPC protocols to
transfer inputs privately. The cost of cryptographic OT scales poorly, so OT
extensions are used as optimizations.
- Parties engage in series of OT calls using short authentication strings
which are correlated to subsequent transactions, rather than fresh keys
each time.
- Only the first interaction uses public-key based OT primitives like Rabin
OT or RSA OT while others exploit the correlations.
OT-extensions including the NUOT and RandOT protocols significantly
improve OT costs and enable practical secure computation for large
domains. Network coding variants like OT-combine further optimize
bandwidth usage.
Secure Computation for Machine Learning
MPC enables distributed training of machine learning models like neural
networks, SVM over privacy-sensitive datasets across organizations:
- Secret shares of inputs are used with techniques like Yao’s garbled
circuits and SPDZ-like multiplication triples to evaluate activations
layer-by-layer privately.
- Gradients are computed locally on shares and aggregated. Updates are
applied using MPC primitives.
- FlexSC and ABY frameworks enable training commonly used ML models
with low latency compared to general MPC techniques.
Differentially private approaches incorporate calibrated noise to enable
model analytics while provably preventing membership inference attacks.
Blockchain techniques may also be integrated for immutable audit logs
during distributed ML computations.
Applications and Real-world Deployments
Some real-world uses of MPC protocols for data analytics across industries
and organizations:
- Banks conduct credit scores and loan approvals on joint customer
database without data sharing in SecureDec.
- Hospitals collaborate on genomic studies and clinical trials via private
record matching in PSI.
- Government agencies run analytics on covid spread and contact
tracing preserving citizen privacy.
- Enterprises jointly optimize supply chains minimizing customer/product
privacy risks using MPC supply chain management platforms.
- Partners jointly optimize Internet ad auction by running second price
auction on private bids with techniques from P4P.
MPC deployments address privacy mandates while unlocking business value
from distributed analytics at global scales. Standardization efforts further
drive adoption in privacy-critical applications.
Conclusion
Cryptographic protocols like Yao’s garbled circuits, GMW, BGW and their
variants provide secure foundations for MPC enabling distributed parties to
jointly compute over private inputs. Optimizations including OT extensions
and protocols tailored for machine learning scale these techniques to real-
world deployments handling massive volumes of sensitive data across
organizational boundaries. MPC addresses privacy challenges through its
provable security definitions, heralding a new paradigm of collaborative
analytics respecting information ownership.
Distributed systems spanning multiple organizations handle increasingly
sensitive data that needs joint analysis and insights while maintaining
privacy. Secure multi-party computation (MPC) enables this by allowing
distributed parties to compute functions over their private inputs without
revealing anything beyond the outputs. MPC finds applications in finance,
healthcare, supply chains and other sectors involving collaborative data
analytics across borders. This paper discusses cryptographic protocols
underlying MPC and reviews representative techniques employed to ensure
privacy and security in distributed systems.
Privacy Requirements for MPC
Key privacy requirements for MPC protocols include:
- Input privacy: No party should learn anything about others’ private
inputs beyond what can be inferred from their own inputs and the
function output.
- Output privacy: For functions like comparisons, the output too should
reveal minimum information.
- Semi-honest and malicious security: Protocols should be secure under
passive attacks by semi-honest adversaries following the protocol, and
active attacks by malicious adversaries.
- Collusion resistance: No set of colluding parties should gain privacy
breach beyond individual capabilities.
- Circular security: Even if intermediate values are revealed in multi-
round protocols, the final inputs should remain private.
Protocols aim to achieve these through secure computation using
cryptographic constructions that simulate a “trusted third party”
functionality in a decentralized model.
Yao’s Garbled Circuits
Yao’s protocol is one of the earliest and most practical MPC techniques. It
works by garbling or encrypting a boolean circuit representation of the
function using free-XOR technique and oblivious transfer.
- Circuit encryption uses random keys so no information about
inputs/outputs leak beyond what can be computed from output.
- Inputs are mapped to keys which participant obtains through 1-2 OT to
keep inputs private.
- Evaluation reveals only keys and not the original inputs. Output
consists of keys decrypting to result.
Optimizations include fixed-key garbling, row reduction, halfgates etc. It
provides security against passive attacks and has been deployed for
applications like private set intersection.
GMW and BGW Protocols
The GMW protocol by Goldreich-Micali-Wigderson was one of the first general
MPC protocols and introduced the idea of secret sharing. Inputs are secret
shared among parties who locally perform operations and reconstruct output
shares.
Ben-Or, Goldwasser and Wigderson (BGW) protocol concurrently introduced
verifiable secret sharing addressing active attacks, by requiring parties to
prove they followed protocol using zero-knowledge proofs over secret shares.
- Inputs are shared additively/multiplicatively among parties who
perform local computations.
- Multiplication needs multiplication triples generated during pre-
processing.
- Reconstruction reveals output but not individual inputs under
appropriate threshold.
Variants like SPDZ provide improved efficiency and fault tolerance for
arithmetic circuits. They work for both honest-majority and dishonest-
majority settings but have higher overhead than garbled circuits.
Oblivious Transfer Extensions
Oblivious transfer (OT) forms a core building block in most MPC protocols to
transfer inputs privately. The cost of cryptographic OT scales poorly, so OT
extensions are used as optimizations.
- Parties engage in series of OT calls using short authentication strings
which are correlated to subsequent transactions, rather than fresh keys
each time.
- Only the first interaction uses public-key based OT primitives like Rabin
OT or RSA OT while others exploit the correlations.
OT-extensions including the NUOT and RandOT protocols significantly
improve OT costs and enable practical secure computation for large
domains. Network coding variants like OT-combine further optimize
bandwidth usage.
Secure Computation for Machine Learning
MPC enables distributed training of machine learning models like neural
networks, SVM over privacy-sensitive datasets across organizations:
- Secret shares of inputs are used with techniques like Yao’s garbled
circuits and SPDZ-like multiplication triples to evaluate activations
layer-by-layer privately.
- Gradients are computed locally on shares and aggregated. Updates are
applied using MPC primitives.
- FlexSC and ABY frameworks enable training commonly used ML models
with low latency compared to general MPC techniques.
Differentially private approaches incorporate calibrated noise to enable
model analytics while provably preventing membership inference attacks.
Blockchain techniques may also be integrated for immutable audit logs
during distributed ML computations.
Applications and Real-world Deployments
Some real-world uses of MPC protocols for data analytics across industries
and organizations:
- Banks conduct credit scores and loan approvals on joint customer
database without data sharing in SecureDec.
- Hospitals collaborate on genomic studies and clinical trials via private
record matching in PSI.
- Government agencies run analytics on covid spread and contact
tracing preserving citizen privacy.
- Enterprises jointly optimize supply chains minimizing customer/product
privacy risks using MPC supply chain management platforms.
- Partners jointly optimize Internet ad auction by running second price
auction on private bids with techniques from P4P.
MPC deployments address privacy mandates while unlocking business value
from distributed analytics at global scales. Standardization efforts further
drive adoption in privacy-critical applications.
Conclusion
Cryptographic protocols like Yao’s garbled circuits, GMW, BGW and their
variants provide secure foundations for MPC enabling distributed parties to
jointly compute over private inputs. Optimizations including OT extensions
and protocols tailored for machine learning scale these techniques to real-
world deployments handling massive volumes of sensitive data across
organizational boundaries. MPC addresses privacy challenges through its
provable security definitions, heralding a new paradigm of collaborative
analytics respecting information ownership.
