IndexKernel<T>

Models & Types · AiDotNet.Kernels

Index kernel for multi-task/multi-output Gaussian Processes.

Implements: IKernelFunction<T>

For Beginners

The IndexKernel enables GPs to model multiple related tasks or outputs simultaneously, sharing information between them.

In multi-task learning, you have multiple related prediction tasks. For example:

  • Predicting exam scores for multiple subjects (math, science, english)
  • Forecasting sales for multiple products
  • Estimating properties of multiple chemical compounds

The IndexKernel models the correlation between tasks using a covariance matrix: k_task(t, t’) = B[t, t’]

Where B is a positive semi-definite “task covariance matrix” that captures:

  • B[t, t]: The variance of task t
  • B[t, t’]: The covariance between tasks t and t’

When combined with an input kernel (e.g., RBF), the full multi-task kernel is: k((x, t), (x’, t’)) = k_input(x, x’) × k_task(t, t’) = k_input(x, x’) × B[t, t’]

This allows the GP to:

  • Share information between similar tasks
  • Leverage data from data-rich tasks to help data-poor tasks
  • Learn task relationships from data

How It Works

Usage patterns:

  1. Initialize with a task covariance matrix (if known)
  2. Initialize randomly and optimize the task covariance
  3. Use with ProductKernel to combine with input kernel

Constructors

Constructor Summary
IndexKernel(double[] taskCovariance) Initializes an IndexKernel with a specified task covariance matrix.
IndexKernel(int numTasks, int? rank, double variance, int? seed) Initializes an IndexKernel with specified number of tasks using random initialization.

Properties

Property Summary
int NumTasks { get; } Gets the number of tasks.

Methods

Method Summary
IndexKernel<T> Independent(int numTasks, double variance) Creates an IndexKernel with identity task covariance (independent tasks).
IndexKernel<T> UniformCorrelation(int numTasks, double correlation, double variance) Creates an IndexKernel with uniform task correlation.
T Calculate(Vector<T> x1, Vector<T> x2) Calculates the task kernel value between two task indices.
double CalculateForTasks(int task1, int task2) Gets the task covariance for explicit task indices.
double GetTaskCorrelation(int task1, int task2) Gets the correlation between two tasks.
double[] GetTaskCovariance() Gets a copy of the task covariance matrix.