Medical Recommendations Server. More...

#include <server.h>

Inheritance diagram for SeComLib::SecureRecommendations::Server:

Public Types
typedef std::vector < Paillier::Ciphertext >	EncryptedSvmValues
	Vector of encrypted SVM evaluations.

typedef std::vector < Paillier::Ciphertext >	EncryptedClusterVotes
	Container for the encrypted cluster votes.

Public Member Functions
	Server (const PaillierPublicKey &key)
	Constructor. More...

	~Server ()
	Destructor.

void	SetClient (const std::shared_ptr< const Hub > &client)
	Sets a reference to the client. More...

void	Initialize ()
	Initializes the internal SVM instances. More...

void	InteractiveSecureDivision (const BigInteger &numerator, SecureSvm::EncryptedVector &denominators) const
	Performs the interactive secure division protocol. More...

void	GetAccuracyPredictions (Server::EncryptedClusterVotes &clusterVotes, SecureSvm::EncryptedVector &safetyPredictions, const TestDataRow &medicalRelevanceTestData, const std::map< std::string, TestDataRow > &safetyTestData) const
	Procedure which computes (and returns) the cluster votes for the first block and the medical safety predictions for accuracy analysis. More...

void	GetPerformancePredictions (SecureSvm::EncryptedVector &firstTwoBlocksPredictions, SecureSvm::EncryptedVector &SafetyBlockPredictions, const TestDataRow &medicalRelevanceTestData, const std::map< std::string, TestDataRow > &safetyTestData) const
	Procedure which computes (and returns) the predictions generated by the first two blocks and the safety block predictions for performance analysis. More...

std::vector< std::string >	GetSafetyBlockSvmsUnsafeClasses () const
	Returns the unsafe classes for each safety SVM. More...

std::deque< std::string >	GetSafetyBlockModelFiles () const
	Returns the names of the model files for the safety block. More...

void	DebugValue (const Paillier::Ciphertext &value) const
	Sends a request to the client to debug an encrypted value. More...

Private Types
typedef std::vector< std::pair < size_t, size_t > >	PermutationVector
	Define the permutation map container.

Private Member Functions
void	loadMedicalRelevanceSvmModels (const std::string &modelsDirectory)
	Loads the Medical Relevance block SVM models. More...

void	loadSafetySvmModels (const std::string &modelsDirectory)
	Loads the Safety block SVM models. More...

Server::EncryptedClusterVotes	getTotalClusterVotes (Server::EncryptedSvmValues &votes) const
	Returns the total number of votes per cluster. More...

void	interactiveSignEvaluation (Server::EncryptedSvmValues &data) const
	Performs the interactive sign evaluation protocol inplace. More...

void	interactiveMaximumEvaluation (Server::EncryptedClusterVotes &data) const
	Performs the interactive maximum evaluation protocol inplace. More...

	Server (Server const &)
	Copy constructor - not implemented.

Server	operator= (Server const &)
	Copy assignment operator - not implemented.

Private Attributes
std::weak_ptr< const Hub >	client

Paillier	cryptoProvider
	The crypto provider.

PaillierPublicKey	clientPublicKey
	The client public key.

SecureSvm::KernelTypes	kernel
	The SVM kernel type.

unsigned int	contentItemCount
	The number of content items.

std::vector< BigInteger >	dummyContentItems
	Dummy vector of content items.

std::vector< Paillier::Ciphertext >	dummyEncryptedPreferenceScores
	Dummy vector of encrypted preference scores.

unsigned int	blindingFactorSize
	The maximum size of the blinding factors.

std::string	modelFileExtension
	The model files extension.

std::string	medicalRelevanceModelsDirectory
	Medical relevance models folder.

std::string	safetyModelsDirectory
	The full path to the safety block models directory.

std::deque< std::string >	safetySvmModelFiles
	The safety block model files.

std::vector< std::unique_ptr < SecureSvm > >	medicalRelevanceSvms
	Medical Relevance Block SVM models.

std::vector< std::unique_ptr < SecureSvm > >	safetySvms
	Safety Block SVM models.

unsigned int	medicalRelevanceClusterCount
	Number of clusters for the Medical Relevance Block.

Server::EncryptedClusterVotes	encryptedClusterVotes
	Stores the number of votes received for each cluster after evaluating the SVMs.

Paillier::Ciphertext	encryptedZero
	Precompute [0] for optimization purposes.

Paillier::Ciphertext	encryptedOne
	Precompute [1] for optimization purposes.

Detailed Description

Medical Recommendations Server.

Definition at line 62 of file secure_recommendations/server.h.

Constructor & Destructor Documentation

SeComLib::SecureRecommendations::Server::Server ( const PaillierPublicKey & key )

Constructor.

Initializes the client's public key

Precomputes required data

Parameters

key	The client public key

Initialize the medical relevance SVM models parameters

Initialize the safety SVM models parameters

Precompute [0] and [1] for optimization purposes

Definition at line 41 of file secure_recommendations/server.cpp.

Member Function Documentation

void SeComLib::SecureRecommendations::Server::DebugValue ( const Paillier::Ciphertext & value ) const

Sends a request to the client to debug an encrypted value.

Parameters

value the value to debug

Definition at line 277 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::GetAccuracyPredictions	(	Server::EncryptedClusterVotes &	clusterVotes,
		SecureSvm::EncryptedVector &	safetyPredictions,
		const TestDataRow &	medicalRelevanceTestData,
		const std::map< std::string, TestDataRow > &	safetyTestData
	)		const

Procedure which computes (and returns) the cluster votes for the first block and the medical safety predictions for accuracy analysis.

Determines the cluster with the highest number of votes.

Given N, the number of clusters, for each cluster we must perform binary clasification and then sum the votes for each cluster. The cluster having the highest number of votes is determined with an interactive protocol.

Evaluates the safety block SVMs.

Parameters

clusterVotes	returns a vector containing [1] at the index of the cluster with the maximum number of votes and [0] everywhere else
safetyPredictions	returns a vector which, for each safety SVM, contains [1] if the SVM produced a positive value and [0] otherwise
medicalRelevanceTestData	the medical relevance test data
safetyTestData	the safety test data

Exceptions

Missing safety test data

First, we evaluate all the medical relevance SVMs

We also evaluate the safety SVMs (we will pop these out later, because we want to perform the interactive sign evaluation in a single step)

Todo:: Throw a custom exception here...

Ask the client to determine the sign with an interactive secure permutation algorithm, such that for each value we will contain either [0] or [1]

Compute the total number of votes for each cluster (we add the elements on each column together)

Ask the client to evaluate the cluster with the maximum number of votes The results vector will contain [1] at the index of the maximum and [0] at the other indexes

Definition at line 166 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::GetPerformancePredictions	(	SecureSvm::EncryptedVector &	firstTwoBlocksPredictions,
		SecureSvm::EncryptedVector &	safetyPredictions,
		const TestDataRow &	medicalRelevanceTestData,
		const std::map< std::string, TestDataRow > &	safetyTestData
	)		const

Procedure which computes (and returns) the predictions generated by the first two blocks and the safety block predictions for performance analysis.

Evaluates the output of the first two blocks:

Produces a vecor of sum_j(dummyEncryptedContentItems[i] * vote[j])
Adds the above vector homomorphically to the dummyEncryptedPreferenceScores and obtains the firstTwoBlocksPredictions
Sets the first 5 elements of the dummyEncryptedSafetyScores to the value obtained by evaluating the safety SVM

Parameters

firstTwoBlocksPredictions	returns a vector containing the combined predictions of the first two blocks
safetyPredictions	returns a vector which, for each safety SVM, contains [1] if the SVM produced a positive value and [0] otherwise
medicalRelevanceTestData	the medical relevance test data
safetyTestData	the safety test data

Evaluate the medical relevance votes and the safety prediction

Compute the outptut of the first two blocks

Definition at line 233 of file secure_recommendations/server.cpp.

std::deque< std::string > SeComLib::SecureRecommendations::Server::GetSafetyBlockModelFiles ( ) const

Returns the names of the model files for the safety block.

Returns: a vector containing the names of the model files for the safety block

Definition at line 270 of file secure_recommendations/server.cpp.

std::vector< std::string > SeComLib::SecureRecommendations::Server::GetSafetyBlockSvmsUnsafeClasses ( ) const

Returns the unsafe classes for each safety SVM.

Returns: a vector containing the set of unsafe classes for each safety SVM

Definition at line 257 of file secure_recommendations/server.cpp.

Server::EncryptedClusterVotes SeComLib::SecureRecommendations::Server::getTotalClusterVotes ( Server::EncryptedSvmValues & votes ) const

private

Returns the total number of votes per cluster.

Parameters

votes a vector containing the encrypted values predicted by the SVMs

Returns: Vector containing the votes for each cluster

We want to compute the sum of elements on each column, so we iterate over the lines in the inner loop

We computed only the SVM predictions for cluster(i, j), where i < j, because for i > j, prediction(i, j) = 1 - prediction(j, i)

Definition at line 324 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::Initialize ( )

Initializes the internal SVM instances.

This method needs to be called first!

Loads the medical relevance and the safety SVMs.

Generates dummy data for the content items, the preferences block and the safety block.

Load medical relevance block SVMs

Load safety block SVMs

Generate a dummy vector of content items with values from 0 to 10 and scale them with 10.

Generate a dummy vector of encrypted preferences block scores with values from 1 to 10.

Todo:: Is the range really from 1 to 10 or is it 0 to 10?

Definition at line 78 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::interactiveMaximumEvaluation ( Server::EncryptedClusterVotes & data ) const

private

Performs the interactive maximum evaluation protocol inplace.

Overwrites the input data with a vector containing [1] at the location of the maximum and [0] everywhere else.

Permutes the values.

Blinds the values multiplicatively and additively \( [w_i r_1 + r_2] = [w_i]^{r_1} r_2 \).

Calls the client's EvaluateMaximum method.

Reverses the permutation.

Parameters

data	a vector of encrypted cluster votes

Get a random permutation

Permute the input vector

Multiplicatively blind the values with a random factor \( r_1 > 0 \)

Additively blind the values with a random factor \( r_2 \)

Interact with the client

Reverse the permutation

Definition at line 407 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::InteractiveSecureDivision	(	const BigInteger &	numerator,
		SecureSvm::EncryptedVector &	denominators
	)		const

Performs the interactive secure division protocol.

Blinds each denominator multiplicatively with a factor \( r_i \), such that it does not generate overflows.

Permutes all the denominators randomly.

Sends the permuted vector of blinded denominators to the client and receives the results of \( 1 / d_i \).

Undoes the random permutation and homomorphically multiplies each result by \( r_i \), in order to get the desired values.

Parameters

numerator	the value of the numerator
denominators	vector of encrypted denominators. Will be overwritten by the kernel values

Blind the encrypted denominators. Store the blinding factors.

\( r_i \) must always be > 0!

Get a random permutation

Permute the blinded encrypted denominators.

Interact with the client

Reverse the permutation

Undo the blinding

Definition at line 119 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::interactiveSignEvaluation ( Server::EncryptedSvmValues & data ) const

private

Performs the interactive sign evaluation protocol inplace.

Overwrites the input data with a vector of votes ([0] or [1]) for each value in the input vector.

Permutes the values.

Multiplicatively blinds the values: \( [f_i(x)]^{r_i} \).

Calls the client's EvaluateSign method.

Reverses the permutation.

Parameters

data	a vector containing the encypted values predicted by the SVMs

Get a random permutation

Permute the input vector

Multiplicatively blind the values with a random factor \( r \)

Interact with the client

Reverse the permutation

Definition at line 374 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::loadMedicalRelevanceSvmModels ( const std::string & modelsDirectory )

private

Loads the Medical Relevance block SVM models.

The SVMs are binary classifiers between two clusters.

Since the SVMs are symmetric ( \( SVM_{jvi} = 1 - SVM_{ivj} \)), we only load the ones where \( i < j \)

Parameters

modelsDirectory the path to the medical relevance SVM models directory

Definition at line 288 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::loadSafetySvmModels ( const std::string & modelsDirectory )

private

Loads the Safety block SVM models.

Parameters

modelsDirectory the path to the safety SVM models directory

Todo:: Throw a custom error.

Definition at line 305 of file secure_recommendations/server.cpp.

void SeComLib::SecureRecommendations::Server::SetClient ( const std::shared_ptr< const Hub > & client )

Sets a reference to the client.

Associates the server with the client instance

Parameters

client a client instance

Definition at line 67 of file secure_recommendations/server.cpp.

Member Data Documentation

std::weak_ptr<const Hub> SeComLib::SecureRecommendations::Server::client

private

A reference to the client. Use weak_ptr here in order to avoid a circular deadlock with the client which holds a reference back to the server

Definition at line 106 of file secure_recommendations/server.h.

The documentation for this class was generated from the following files:

Public Types

Public Member Functions

Private Types

Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation