Class FederatedLearningMetadata

Namespace

AiDotNet.Models

Assembly

AiDotNet.dll

Contains metadata and metrics about federated learning training progress and results.

public class FederatedLearningMetadata

Inheritance

object

FederatedLearningMetadata

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

This class tracks various metrics throughout the federated learning process to help monitor training progress, diagnose issues, and evaluate model quality.

For Beginners: Metadata is like a training diary that records what happened during federated learning - how long it took, how accurate the model became, which clients participated, etc.

Think of this as a comprehensive training report containing:

• Performance metrics: Accuracy, loss, convergence
• Resource usage: Time, communication costs
• Participation: Which clients contributed
• Privacy tracking: Privacy budget consumption

For example, after training you might see:

• Total rounds: 50 (out of max 100)
• Final accuracy: 92.5%
• Training time: 2 hours
• Total clients participated: 100
• Privacy budget used: ε=5.0 (out of 10.0 total)

Fields

MetadataKey

The recommended key for storing federated learning metadata inside Properties.

public const string MetadataKey = "FederatedLearning"

Field Value

string

Properties

AccuracyHistory

Gets or sets the history of accuracy values across all rounds.

public List<double> AccuracyHistory { get; set; }

Property Value

List<double>

Remarks

For Beginners: A list showing how accuracy improved after each round.

For example: [0.60, 0.70, 0.78, 0.84, 0.88, 0.91, 0.93, 0.94, 0.945, 0.95] Shows accuracy improving from 60% to 95%

AggregationStrategyUsed

Gets or sets the aggregation strategy used during training.

public string AggregationStrategyUsed { get; set; }

Property Value

string

Remarks

For Beginners: Records which aggregation algorithm was used (FedAvg, FedProx, etc.). Important for reproducibility and understanding results.

AsyncModeUsed

Gets or sets the asynchronous federated learning mode used (if any).

public string AsyncModeUsed { get; set; }

Property Value

string

Remarks

For Beginners: If async FL is enabled, this records which mode was used (FedAsync or FedBuff). If training was synchronous, this is "None".

AverageClientsPerRound

Gets or sets the average number of clients selected per round.

public double AverageClientsPerRound { get; set; }

Property Value

double

Remarks

For Beginners: How many clients were active in each round on average.

For example:

• Round 1: 10 clients
• Round 2: 8 clients (some unavailable)
• Round 3: 10 clients
• Average: 9.3 clients per round

AverageRoundTimeSeconds

Gets or sets the average time per round in seconds.

public double AverageRoundTimeSeconds { get; set; }

Property Value

double

Remarks

For Beginners: How long each round takes on average. Useful for estimating how long future training runs will take.

For example:

• 50 rounds completed in 5000 seconds
• Average: 100 seconds per round
• Next training with 100 rounds will take ~10,000 seconds

CompressionEnabled

Gets or sets whether compression was enabled for client updates.

public bool CompressionEnabled { get; set; }

Property Value

bool

CompressionStrategyUsed

Gets or sets the compression strategy used for client updates (if enabled).

public string CompressionStrategyUsed { get; set; }

Property Value

string

Converged

Gets or sets whether training converged before reaching maximum rounds.

public bool Converged { get; set; }

Property Value

bool

Remarks

For Beginners: Convergence means the model stopped improving significantly. If true, training ended early because the model reached a good solution.

For example:

• Max rounds: 100
• Converged at round 50
• Converged = true
• Saved time by stopping early

ConvergenceRound

Gets or sets the round at which convergence was detected.

public int ConvergenceRound { get; set; }

Property Value

int

Remarks

For Beginners: Which round did the model stop improving significantly?

Useful for:

• Setting better MaxRounds for future training
• Understanding training dynamics
• Comparing different algorithms

DifferentialPrivacyEnabled

Gets or sets whether differential privacy was enabled.

public bool DifferentialPrivacyEnabled { get; set; }

Property Value

bool

Remarks

For Beginners: Records whether privacy mechanisms were active during training. Important for understanding any accuracy trade-offs.

FinalAccuracy

Gets or sets the final global model accuracy on validation data.

public double FinalAccuracy { get; set; }

Property Value

double

Remarks

For Beginners: Accuracy measures how often the model makes correct predictions. For example, 0.95 means the model is correct 95% of the time.

This is measured on validation data that wasn't used for training.

FinalLoss

Gets or sets the final global model loss value.

public double FinalLoss { get; set; }

Property Value

double

Remarks

For Beginners: Loss measures how far the model's predictions are from the true values. Lower loss indicates better performance.

For example:

• Initial loss: 2.5
• Final loss: 0.3
• The model has improved significantly

HeterogeneityCorrectionUsed

Gets or sets the heterogeneity correction method used (if any).

public string HeterogeneityCorrectionUsed { get; set; }

Property Value

string

HomomorphicEncryptionEnabled

Gets or sets whether homomorphic encryption was enabled.

public bool HomomorphicEncryptionEnabled { get; set; }

Property Value

bool

HomomorphicEncryptionModeUsed

Gets or sets the HE mode used (HEOnly or Hybrid).

public string HomomorphicEncryptionModeUsed { get; set; }

Property Value

string

HomomorphicEncryptionProviderUsed

Gets or sets the HE provider used.

public string HomomorphicEncryptionProviderUsed { get; set; }

Property Value

string

HomomorphicEncryptionSchemeUsed

Gets or sets the HE scheme used (CKKS or BFV).

public string HomomorphicEncryptionSchemeUsed { get; set; }

Property Value

string

LossHistory

Gets or sets the history of loss values across all rounds.

public List<double> LossHistory { get; set; }

Property Value

List<double>

Remarks

For Beginners: A list showing how loss changed after each round. Useful for plotting learning curves and diagnosing training issues.

For example: [2.5, 1.8, 1.2, 0.9, 0.7, 0.5, 0.4, 0.35, 0.32, 0.3] Shows steady improvement from 2.5 to 0.3

MetaLearningEnabled

Gets or sets whether federated meta-learning was enabled.

public bool MetaLearningEnabled { get; set; }

Property Value

bool

Remarks

For Beginners: Meta-learning in federated settings learns a global initialization that adapts quickly to each client.

MetaLearningInnerEpochsUsed

Gets or sets the inner (client) adaptation epochs used for meta-learning.

public int MetaLearningInnerEpochsUsed { get; set; }

Property Value

int

MetaLearningRateUsed

Gets or sets the meta learning rate used by the server update rule.

public double MetaLearningRateUsed { get; set; }

Property Value

double

MetaLearningStrategyUsed

Gets or sets the federated meta-learning strategy used (Reptile/PerFedAvg/FedMAML).

public string MetaLearningStrategyUsed { get; set; }

Property Value

string

Notes

Gets or sets additional notes or observations about the training run.

public string Notes { get; set; }

Property Value

string

Remarks

For Beginners: A freeform field for recording anything unusual or noteworthy that happened during training.

For example:

• "Client 5 dropped out after round 30"
• "Convergence was slower than expected"
• "High variance in client update quality"

PersonalizationEnabled

Gets or sets whether personalization was enabled.

public bool PersonalizationEnabled { get; set; }

Property Value

bool

Remarks

For Beginners: Personalization means some parameters can remain client-specific (or cluster-specific), improving accuracy on non-IID client data.

PersonalizationLocalAdaptationEpochs

Gets or sets the number of post-aggregation local adaptation epochs (if configured).

public int PersonalizationLocalAdaptationEpochs { get; set; }

Property Value

int

PersonalizationStrategyUsed

Gets or sets the personalization strategy used (FedPer, FedRep, Ditto, pFedMe, Clustered).

public string PersonalizationStrategyUsed { get; set; }

Property Value

string

PersonalizedParameterFraction

Gets or sets the fraction of parameters treated as personalized for head-split strategies.

public double PersonalizedParameterFraction { get; set; }

Property Value

double

PrivacyAccountantUsed

Gets or sets which privacy accountant was used for reporting.

public string PrivacyAccountantUsed { get; set; }

Property Value

string

ReportedDelta

Gets or sets the delta used when reporting ReportedEpsilonAtDelta.

public double ReportedDelta { get; set; }

Property Value

double

ReportedEpsilonAtDelta

Gets or sets the reported epsilon at the reported delta (when supported by the accountant).

public double ReportedEpsilonAtDelta { get; set; }

Property Value

double

Remarks

For Beginners: Some accountants (like RDP) can report a tighter epsilon for a given delta. This value is intended for reporting; it does not change how noise was applied.

RoundMetrics

Gets or sets the per-round detailed metrics.

public List<RoundMetadata> RoundMetrics { get; set; }

Property Value

List<RoundMetadata>

Remarks

For Beginners: Detailed information about each individual round, including which clients participated, their individual losses, communication costs, etc.

Useful for:

• Detailed analysis of training dynamics
• Identifying problematic clients
• Understanding convergence patterns

RoundsCompleted

Gets or sets the number of federated learning rounds completed.

public int RoundsCompleted { get; set; }

Property Value

int

Remarks

For Beginners: A round is one complete cycle where clients train and the server aggregates updates. This counts how many such cycles were completed.

SecureAggregationEnabled

Gets or sets whether secure aggregation was enabled.

public bool SecureAggregationEnabled { get; set; }

Property Value

bool

Remarks

For Beginners: Records whether client updates were encrypted during aggregation.

SecureAggregationMinimumUploaderCountUsed

Gets or sets the minimum uploader count used by secure aggregation (dropout-resilient modes).

public int SecureAggregationMinimumUploaderCountUsed { get; set; }

Property Value

int

SecureAggregationModeUsed

Gets or sets which secure aggregation mode was used.

public string SecureAggregationModeUsed { get; set; }

Property Value

string

Remarks

For Beginners: Different secure aggregation modes provide different drop-out tolerance. This value is "None" when secure aggregation is disabled.

SecureAggregationReconstructionThresholdUsed

Gets or sets the reconstruction threshold used by secure aggregation (dropout-resilient modes).

public int SecureAggregationReconstructionThresholdUsed { get; set; }

Property Value

int

ServerOptimizerUsed

Gets or sets the server-side federated optimizer used (FedOpt family).

public string ServerOptimizerUsed { get; set; }

Property Value

string

Remarks

For Beginners: Some federated learning variants apply an optimizer on the server after aggregation (for example, FedAvgM, FedAdam, FedYogi). If no server optimizer is used, this is "None".

TotalClientsParticipated

Gets or sets the total number of clients that participated across all rounds.

public int TotalClientsParticipated { get; set; }

Property Value

int

Remarks

For Beginners: How many different clients contributed to the model.

For example:

• 100 clients available
• 10 clients selected per round
• Over 50 rounds, might have 80 unique participants

TotalCommunicationMB

Gets or sets the total communication cost in megabytes.

public double TotalCommunicationMB { get; set; }

Property Value

double

Remarks

For Beginners: The total amount of data transferred between clients and server throughout training. Important for understanding bandwidth requirements.

For example:

• Each model update: 10 MB
• 10 clients per round
• 50 rounds
• Total: 10 MB × 10 × 50 × 2 (up and down) = 10,000 MB = 10 GB

TotalPrivacyBudgetConsumed

Gets or sets the total privacy budget (epsilon) consumed.

public double TotalPrivacyBudgetConsumed { get; set; }

Property Value

double

Remarks

For Beginners: If differential privacy is used, this tracks how much privacy budget has been spent. Privacy budget is finite - once exhausted, no more privacy guarantees.

For example:

• ε per round: 0.1
• 50 rounds completed
• Total consumed: 5.0
• If total budget is 10.0, have 5.0 remaining

TotalPrivacyDeltaConsumed

Gets or sets the total privacy delta consumed (basic composition reporting).

public double TotalPrivacyDeltaConsumed { get; set; }

Property Value

double

Remarks

For Beginners: Delta (δ) is the probability that the privacy guarantee fails. When using simple accounting, deltas add up across rounds.

TotalTrainingTimeSeconds

Gets or sets the total training time in seconds.

public double TotalTrainingTimeSeconds { get; set; }

Property Value

double

Remarks

For Beginners: The total wall-clock time from start to finish of federated learning, including all rounds, communication, and aggregation.