Secure Multi-Party Computation (MPC)

This is the fundamental guarantee of MPC. It ensures that no party learns anything more about another party's private input than what can be inferred from the output of the computed function itself. The protocol is designed so that intermediate computation states (shares, messages) reveal zero information about the original secrets.

• Formal Definition: A protocol is private if the view of any party (their input, randomness, and received messages) can be simulated using only their own input and the final output. This simulation-based security proves that the protocol leaks nothing extra.
• Example: In a joint salary average computation, each company learns only the final average, not any other company's individual salary figure.

This property guarantees that the protocol computes the exact, agreed-upon function and that all honest participants receive the correct output. It must hold even in the presence of malicious participants who may try to submit invalid data or deviate from the protocol.

• Robustness: In many MPC models, if all parties follow the protocol, the correct result is always produced.
• Guaranteed Output Delivery (GOD): A stronger form ensures that honest parties always receive an output, even if malicious parties abort. Weaker models may allow abort, but not incorrect results.
• Example: In a secure auction, correctness ensures the true highest bid is identified and the winner is correctly determined, preventing a malicious party from claiming victory with a lower bid.

This property ensures that parties must commit to their inputs before learning anything about the inputs of others. It prevents an adversary from choosing their input as a function of another party's input, which could leak information or manipulate the result.

• Formalization: Often enforced by a commitment phase at the beginning of the protocol, where inputs are cryptographically locked in.
• Importance: Without this, in a millionaire's problem, the second party could tailor their input based on a guess about the first party's wealth, iteratively probing to discover the exact value.
• TinyML Context: In Federated Edge Learning (a form of MPC), this property ensures a device's model update is based solely on its local data, not influenced by updates from other devices it has seen.

MPC protocols are proven secure under specific, formal adversarial models that define the adversary's capabilities. The two primary models are:

• Semi-Honest (Passive) Adversaries: Parties follow the protocol but try to learn extra information from the messages they receive. Security guarantees privacy.
• Malicious (Active) Adversaries: Parties may arbitrarily deviate from the protocol—sending wrong messages, aborting, or injecting false inputs. Security must guarantee both privacy and correctness.

Additional dimensions include:

• Threshold (t-out-of-n): Security is guaranteed as long as fewer than t parties are corrupted.
• Static vs. Adaptive: Whether the adversary chooses which parties to corrupt before the protocol starts or adaptively during execution.

Fairness ensures that if one party learns the output, then all honest parties learn the output. It prevents a situation where a malicious party learns the result (e.g., the winner of an auction, the result of a business calculation) and then aborts the protocol to prevent others from learning it.

• Challenge: Achieving perfect fairness in asynchronous networks is impossible in general (Cleve's impossibility result).
• Solutions: Many practical protocols adopt partial fairness or leverage gradual release mechanisms where the output is revealed piece-by-piece, or use a trusted third party as a fallback arbiter.
• TinyML Context: Critical in secure sensor fusion where multiple devices contribute data; one device should not gain a tactical advantage by learning the fused result first and then disconnecting.

These are strong liveness properties that define what happens if parties abort.

• Guaranteed Output Delivery (GOD): Honest parties always receive the correct output, regardless of malicious party behavior (including abort). This often requires a majority of honest parties.
• Robustness: The protocol computes the correct output and all parties receive it. If a party aborts, they are simply excluded, and the computation proceeds without them.
• Abort: A weaker but common model where the protocol may terminate without output if a malicious party aborts. Security is maintained (privacy), but liveness is not.

These properties are crucial for business-critical or financial MPC applications where a result is legally or operationally required.

Banks can collaboratively detect cross-institutional fraud patterns and money laundering networks using MPC. Each bank contributes its private transaction records to a secure computation that identifies linked fraudulent activities across accounts held at different institutions, all while keeping individual customer transaction data confidential.

• Key Use Case: Identifying sophisticated, coordinated fraud rings that operate across multiple financial entities.
• Technical Mechanism: A secure computation of linked transactions or anomaly scores, often using secret sharing schemes.
• Business Impact: Increases detection rates for systemic fraud while maintaining strict compliance with customer privacy regulations and competitive secrecy.

Companies use MPC to optimize logistics and procurement through secure multi-party auctions and supply chain analytics. Participants can determine the winning bid in an auction without revealing individual bid amounts, or calculate optimal joint inventory levels without disclosing true demand forecasts or cost structures.

• Key Use Case: A sealed-bid auction where the winner and price are determined without revealing non-winning bids.
• Example: The Danish Sugar Beet Auction is a historic early example where farmers and a processor used MPC to determine a fair market clearing price.
• Benefit: Enables efficient markets and collaborations between otherwise competing entities by removing the risk of exposing strategic business data.

MPC enables privacy-preserving authentication and credential checks. For example, a user can prove they are over a certain age to a service provider without revealing their exact birth date, or prove their salary is within a range for a loan application without disclosing the exact figure. This is achieved by performing the verification logic (e.g., age >= 21) on encrypted inputs.

• Key Use Case: Password breach checking where a user can check if their password has been leaked in a public database without sending the password itself to the checker.
• Standardization: Protocols like Privacy Pass and concepts for verifiable credentials leverage MPC principles.
• Benefit: Minimizes data exposure in authentication flows, reducing the impact of potential data breaches at the verifier.

In Cyber-Physical Systems (CPS) like smart grids or connected vehicles, MPC allows multiple sensors or devices owned by different entities to compute a joint result without leaking proprietary or sensitive raw sensor streams. For instance, multiple energy providers can compute optimal grid load balancing using private generation and consumption data.

• TinyML Application: Multiple constrained microcontroller-based sensors in a factory can jointly compute an aggregate safety metric (e.g., average toxic gas level) without any sensor revealing its precise location or reading to the others.
• Challenge: Requires lightweight cryptographic MPC variants suitable for the severe resource constraints of MCUs.
• Security Goal: Protects against both network-based eavesdropping and compromised nodes within the participant group.

Homomorphic Encryption is a cryptographic scheme that allows computations to be performed directly on encrypted data. Unlike MPC, which involves multiple parties, homomorphic encryption typically involves a single party performing operations on ciphertexts. The encrypted result, when decrypted, matches the result of operations performed on the plaintext.

• Key Distinction from MPC: Enables a single untrusted server to compute on encrypted client data without ever decrypting it.
• Use Case: Privacy-preserving cloud analytics where data owners encrypt data before sending it to a cloud service for processing.
• Performance Consideration: Historically computationally intensive, but newer schemes (e.g., CKKS for approximate arithmetic) are making it more feasible for specific use cases.

Federated Learning is a decentralized machine learning paradigm where a global model is trained across multiple decentralized edge devices holding local data samples, without exchanging the data itself. Instead, devices compute model updates (e.g., gradients) which are sent to a central server for aggregation.

• Relationship to MPC: MPC can secure the federated learning process. While federated learning avoids sharing raw data, model updates can still leak information. MPC protocols can be used to perform secure aggregation of these updates, ensuring the central server never sees individual contributions.
• TinyML Context: Enables on-device learning and model personalization across a fleet of microcontrollers while preserving data privacy at the source.

Differential Privacy is a rigorous mathematical framework that quantifies and bounds the privacy loss incurred by an individual when their data is included in a statistical analysis or dataset. It works by adding carefully calibrated noise to the output of a computation.

• Synergy with MPC: MPC and differential privacy are often used together in a layered defense. MPC can securely compute a function over private data, and differential privacy can then be applied to the final, aggregated result before it is released. This protects against privacy leakage from the output itself.
• Formal Guarantee: Provides an epsilon (ε) privacy budget, offering a quantifiable, mathematical promise of privacy that holds regardless of an adversary's auxiliary information.

A Zero-Knowledge Proof is a cryptographic protocol where one party (the prover) can prove to another party (the verifier) that a given statement is true, without conveying any information beyond the validity of the statement itself.

• Complementary Role to MPC: While MPC is about computing on secret data, ZKPs are about verifying properties of secret data. In an MPC system, ZKPs can be used to ensure participants are correctly following the protocol (malicious security) without revealing their private inputs.
• TinyML Application: Can prove that an inference was performed correctly by a constrained device (e.g., "I ran the correct model on the sensor data") or that private input data meets certain criteria, without revealing the data.

A Trusted Execution Environment is a secure, isolated area of a main processor (e.g., ARM TrustZone, Intel SGX) that provides hardware-enforced protection for the execution of sensitive code and the handling of confidential data, separate from the device's standard operating system.

• Alternative/Complement to MPC: A TEE provides a trusted hardware solution for isolating computation, whereas MPC provides a cryptographic software solution. They can be combined: sensitive components of an MPC protocol (e.g., key generation) can be executed inside a TEE for enhanced security.
• Trade-off: TEEs rely on the security of the specific hardware and its manufacturer, while MPC's security is based on cryptographic assumptions and does not require specialized hardware (though it is more computationally expensive).

Secret Sharing is a fundamental cryptographic primitive where a secret (e.g., a cryptographic key) is divided into multiple parts, called shares, which are distributed among participants. The secret can only be reconstructed when a sufficient number of shares are combined.

• Foundation of MPC: Most MPC protocols are built using secret sharing schemes (e.g., Shamir's Secret Sharing). Each party's private input is secretly shared among all parties. The MPC protocol then performs computations directly on these shares, never reconstructing the original inputs in the clear.

• Threshold Schemes: Uses a (t, n)-threshold where any t out of n shares can reconstruct the secret, but any fewer reveal nothing. This provides resilience against participant drop-out or compromise.