Privacy by Design

A rigorous mathematical framework that guarantees the output of a data analysis or machine learning algorithm does not reveal whether any specific individual's data was included in the input dataset. It adds calibrated statistical noise to queries or model updates to provide a quantifiable privacy guarantee, expressed as epsilon (ε).

• Core Mechanism: Adds noise to aggregate outputs or gradients during training.
• Key Use: Enabling privacy-preserving analytics and federated learning where raw data cannot be shared.
• Example: Apple uses differential privacy in iOS to collect usage statistics without identifying individual users.

A cryptographic protocol that allows one party (the prover) to prove to another party (the verifier) that a statement is true without revealing any information beyond the validity of the statement itself. This enables verification of computations or credentials without exposing the underlying private data.

• Core Mechanism: Uses complex mathematical constructions (e.g., zk-SNARKs, zk-STARKs) to generate a proof of correct execution.
• Key Use: Privacy-preserving authentication, confidential transactions in blockchain, and proving model inference was performed correctly without revealing the model or input data.
• Example: A user can prove they are over 21 without revealing their exact birth date or passport number.

A form of encryption that allows specific types of computations to be performed directly on ciphertext, generating an encrypted result which, when decrypted, matches the result of the operations as if they had been performed on the plaintext. This enables data to be processed while remaining encrypted.

• Core Mechanism: Uses lattice-based cryptography to support mathematical operations on encrypted data.
• Key Use: Secure outsourcing of data processing to untrusted clouds (e.g., performing analytics on encrypted health records).
• Limitation: Currently, fully homomorphic encryption is computationally intensive, making it practical primarily for specific, high-value use cases.

A decentralized machine learning approach where a global model is trained across multiple decentralized edge devices or servers holding local data samples, without exchanging the data itself. Instead, only model updates (e.g., gradients) are shared with a central coordinating server for aggregation.

• Core Mechanism: Iterative training rounds where devices compute updates on local data, send encrypted updates to a central server, which aggregates them to improve the global model.
• Key Use: Training predictive text models on user smartphones or developing diagnostic models across multiple hospitals without centralizing patient data.
• Privacy Consideration: Often combined with differential privacy or secure aggregation to prevent leakage from the shared updates.

A cryptographic technique that enables a group of parties to jointly compute a function over their private inputs while keeping those inputs concealed from each other. The protocol ensures that no party learns anything beyond the final output of the computation.

• Core Mechanism: Uses secret sharing and garbled circuits to distribute computation across parties.
• Key Use: Privacy-preserving data mining, secure auctions, and genomic analysis where multiple institutions want to compute on their combined datasets without revealing individual records.
• Example: Two banks can compute their combined number of defaulting customers for a shared region without revealing each other's customer lists.

A core principle of Privacy by Design that dictates that systems should limit the collection, processing, and retention of personal data to what is directly relevant and necessary to accomplish a specified purpose. It is a proactive constraint on data handling.

• Implementation Techniques:
  • Purpose Specification: Clearly defining the use case before collecting any data.
  • Collection Limitation: Only gathering fields essential for the task.
  • Storage Limitation: Implementing automatic data deletion policies after the purpose is fulfilled.
  • Pseudonymization: Replacing direct identifiers with artificial keys.
• Regulatory Driver: A fundamental requirement of regulations like the GDPR (Article 5).