Table of Contents

Class MetaLearningTask<T, TInput, TOutput>

Namespace
AiDotNet.Data.Structures
Assembly
AiDotNet.dll

Represents a single meta-learning task for few-shot learning, containing support and query sets.

public class MetaLearningTask<T, TInput, TOutput> : IMetaLearningTask<T, TInput, TOutput>

Type Parameters

T

The numeric data type used for calculations (e.g., float, double).

TInput

The input data type (e.g., Matrix<T>, Tensor<T>, double[]).

TOutput

The output data type (e.g., Vector<T>, Tensor<T>, double[]).

Inheritance
MetaLearningTask<T, TInput, TOutput>
Implements
IMetaLearningTask<T, TInput, TOutput>
Inherited Members

Examples

// Create a 5-way 3-shot task
// Support set: 5 classes × 3 shots = 15 examples
// Query set: 5 classes × 10 queries = 50 examples

var supportX = new Tensor<double>(new[] { 15, 784 }); // 15 images, 784 pixels each
var supportY = new Tensor<double>(new[] { 15, 5 });   // 15 one-hot labels, 5 classes
var queryX = new Tensor<double>(new[] { 50, 784 });   // 50 query images
var queryY = new Tensor<double>(new[] { 50, 5 });     // 50 query labels

var task = new MetaLearningTask<double>
{
    SupportSetX = supportX,
    SupportSetY = supportY,
    QuerySetX = queryX,
    QuerySetY = queryY
};

// Use with MAML, Reptile, or SEAL algorithms
var innerLoss = model.Train(task.SupportSetX, task.SupportSetY);
var outerLoss = model.Evaluate(task.QuerySetX, task.QuerySetY);

Remarks

In meta-learning, particularly few-shot learning, a task is a small classification or regression problem sampled from a larger dataset. Each task contains a support set (for adapting/learning) and a query set (for evaluating the adaptation).

For Beginners: Meta-learning is about "learning to learn" - training a model to quickly adapt to new tasks with only a few examples. Think of it like learning a language: - Traditional learning: Learn one specific language from thousands of examples - Meta-learning: Learn the general skill of language acquisition so you can learn new languages faster

A MetaLearningTask represents one mini-problem in this process:

  • Support Set: A few labeled examples the model can study (like a mini training set)
  • Query Set: Examples to test how well the model adapted (like a mini test set)

For example, in 5-way 3-shot classification:

  • The support set has 5 classes with 3 examples each (15 total examples)
  • The query set has examples from those same 5 classes to test performance

The model learns from many such tasks, developing the ability to quickly adapt to new tasks.

Thread Safety: This class is not thread-safe. Create separate instances for concurrent access.

Performance: This is a lightweight container class with O(1) property access. Memory usage depends on the size of the tensors stored.

Properties

Metadata

Gets or sets the additional metadata about the task.

public Dictionary<string, object>? Metadata { get; set; }

Property Value

Dictionary<string, object>

Name

Gets or sets an optional name or identifier for the task.

public string? Name { get; set; }

Property Value

string

NumQueryPerClass

Gets or sets the number of query examples per class.

public int NumQueryPerClass { get; set; }

Property Value

int

Remarks

The number of examples in the query set for each class. Used for evaluating performance after adaptation.

NumShots

Gets or sets the number of shots (examples per class) in the support set.

public int NumShots { get; set; }

Property Value

int

Remarks

In N-way K-shot learning, this represents K (number of examples per class). For example, in 5-way 1-shot learning, NumShots = 1.

NumWays

Gets or sets the number of ways (classes) in this task.

public int NumWays { get; set; }

Property Value

int

Remarks

In N-way K-shot learning, this represents N (number of classes per task). For example, in 5-way 1-shot learning, NumWays = 5.

QueryInput

Gets the input features for the query set (alias for QuerySetX).

public TInput QueryInput { get; }

Property Value

TInput

QueryOutput

Gets the target labels for the query set (alias for QuerySetY).

public TOutput QueryOutput { get; }

Property Value

TOutput

QuerySetX

Gets or sets the input features for the query set.

public TInput QuerySetX { get; set; }

Property Value

TInput

Input data containing the examples used for evaluating task adaptation. Shape depends on TInput type, typically larger than the support set size.

Remarks

For Beginners: These are the examples used to test how well the model adapted to this task. Think of them as the "quiz questions" after studying the practice problems (support set).

Continuing the 5-way example:

  • After studying the 3 examples per class (support set)
  • The model is tested on new, unseen examples from those same classes
  • Typically 10-15 queries per class (50-75 total examples)

The model must correctly classify these new examples based only on what it learned from the few support examples. This tests the model's ability to quickly adapt.

QuerySetY

Gets or sets the target labels for the query set.

public TOutput QuerySetY { get; set; }

Property Value

TOutput

Output data containing the true labels for evaluating predictions on the query set. Shape depends on TOutput type.

Remarks

For Beginners: These are the correct answers for the query set examples. They're used to calculate how accurately the model can classify new examples after adapting to the task using only the support set.

The query accuracy is the key metric in few-shot learning:

  • High query accuracy = Model successfully adapted from few examples
  • Low query accuracy = Model struggled to generalize from limited data

During meta-training, the model learns to maximize query set performance across many different tasks, developing strong few-shot learning abilities.

SupportInput

Gets the input features for the support set (alias for SupportSetX).

public TInput SupportInput { get; }

Property Value

TInput

SupportOutput

Gets the target labels for the support set (alias for SupportSetY).

public TOutput SupportOutput { get; }

Property Value

TOutput

SupportSetX

Gets or sets the input features for the support set.

public TInput SupportSetX { get; set; }

Property Value

TInput

Input data containing the examples used for task adaptation. Shape depends on TInput type (e.g., Matrix<T>, Tensor<T>, double[]).

Remarks

For Beginners: These are the input examples the model can study to adapt to this task. Think of them as the "practice problems" before a quiz. In a 5-way 3-shot image classification task: - You have 5 different categories (ways) - With 3 example images each (shots) - Total: 15 example images to learn from

The model looks at these examples to quickly understand what makes each class unique.

SupportSetY

Gets or sets the target labels for the support set.

public TOutput SupportSetY { get; set; }

Property Value

TOutput

Output data containing the labels corresponding to each example in the support set. Shape depends on TOutput type (e.g., Vector<T>, Tensor<T>, double[]).

Remarks

For Beginners: These are the correct answers for each example in the support set. They tell the model which category each example belongs to.

For example, in animal classification:

  • If SupportSetX contains images of cats, dogs, and birds
  • SupportSetY tells which images are cats (label 0), dogs (label 1), or birds (label 2)

Labels can be:

  • Class indices: [0, 0, 0, 1, 1, 1, 2, 2, 2, ...] for 3-shot
  • One-hot encoded: [[1,0,0], [1,0,0], [1,0,0], [0,1,0], ...] for 3-shot

TaskId

Gets or sets an optional task identifier.

public int? TaskId { get; set; }

Property Value

int?
Edit this page