Enum AutoMLTaskFamily

Namespace

AiDotNet.Enums

Assembly

AiDotNet.dll

Defines the high-level task family for an AutoML run.

public enum AutoMLTaskFamily

Fields

BinaryClassification = 1

Binary (two-class) classification.

DepthEstimation = 30

Depth estimation (predicting depth from 2D images).

Used for monocular/stereo depth prediction, 3D reconstruction.

GraphClassification = 9

Graph (whole-graph) classification.

GraphGeneration = 11

Graph generation.

GraphLinkPrediction = 10

Graph link prediction.

GraphNodeClassification = 8

Graph node classification.

ImageClassification = 17

Image classification.

ImageSegmentation = 19

Image segmentation.

MeshClassification = 26

Mesh classification (classifying 3D mesh objects).

Used for 3D shape recognition with mesh inputs.

Typical models: MeshCNN, SpiralNet++, DiffusionNet.

MeshSegmentation = 27

Mesh segmentation (per-face or per-vertex classification).

Used for mesh part segmentation, texture segmentation.

Typical models: MeshCNN, DiffusionNet.

MultiClassClassification = 2

Multi-class (single-label) classification.

MultiLabelClassification = 3

Multi-label classification (multiple labels can be true for one sample).

ObjectDetection = 18

Object detection.

PointCloudClassification = 21

Point cloud classification (classifying entire point clouds into categories).

Used for 3D object recognition, scene classification, and shape classification.

Typical models: PointNet, PointNet++, DGCNN.

PointCloudCompletion = 23

Point cloud completion (reconstructing missing parts of point clouds).

Used for 3D reconstruction from partial scans.

PointCloudSegmentation = 22

Point cloud segmentation (per-point classification/labeling).

Used for semantic segmentation of 3D scenes, part segmentation of objects.

Typical models: PointNet++, DGCNN with segmentation heads.

RadianceFieldReconstruction = 28

Neural radiance field reconstruction (novel view synthesis from images).

Used for 3D scene reconstruction and photorealistic rendering.

Typical models: NeRF, Instant-NGP, Gaussian Splatting.

Ranking = 6

Ranking (ordering items by relevance, e.g., search results).

Recommendation = 7

Recommendation (ranking or scoring items for users).

Regression = 0

Supervised regression (predicting continuous numeric values).

ReinforcementLearning = 20

Reinforcement learning.

SequenceTagging = 13

Sequence tagging (token-level labels like NER, POS).

SpeechRecognition = 16

Speech recognition (ASR).

TextClassification = 12

Text classification.

TextGeneration = 15

Text generation (language modeling / free-form generation).

ThreeDObjectDetection = 29

3D object detection (detecting and localizing objects in 3D space).

Used for autonomous driving, robotics, AR/VR applications.

TimeSeriesAnomalyDetection = 5

Time-series anomaly detection (detecting rare/abnormal events in time-ordered data).

TimeSeriesForecasting = 4

Time-series forecasting (predicting future values from past time-ordered values).

Translation = 14

Machine translation.

VolumetricClassification = 24

Volumetric classification (classifying 3D voxel grids).

Used for 3D medical imaging, CT/MRI analysis.

Typical models: 3D CNN, 3D U-Net, VoxelCNN.

VolumetricSegmentation = 25

Volumetric segmentation (per-voxel classification).

Used for organ segmentation, tumor detection in medical imaging.

Typical models: 3D U-Net, V-Net.

Remarks

AutoML uses the task family to choose sensible defaults for metrics, evaluation protocols, and candidate selection.

For Beginners: This is the kind of problem you're solving:

• Regression predicts numbers.
• BinaryClassification predicts one of two outcomes.
• MultiClassClassification predicts one of many outcomes.
• TimeSeriesForecasting predicts future values from past time-ordered values.
• ReinforcementLearning learns by interacting with an environment to maximize reward.