AI 용어집
인공지능 완전 사전
Image classification
Computer vision task consisting of assigning a predefined label to an entire image according to its main content, using machine learning algorithms.
Convolutional Neural Network
Deep learning architecture specialized in image processing, using convolutional layers to automatically extract hierarchical features.
Convolution
Mathematical operation applying a filter (kernel) over an image to detect specific patterns like edges, textures, or shapes.
Pooling
Dimensionality reduction operation that samples extracted features, allowing to reduce computational complexity while preserving essential information.
Softmax
Output activation function converting logits into probability distribution over classes, ensuring that the sum of probabilities equals 1.
Confusion matrix
Performance evaluation table showing the model's correct and incorrect predictions for each class, allowing identification of systematic errors.
One-hot encoding
Vector representation of class labels where each vector contains a single element equal to 1 and all others equal to 0.
ResNet architecture
Family of deep neural networks using residual connections to enable training of very deep networks without performance degradation.
Fully connected layers
Final layers of a CNN where each neuron is connected to all neurons of the previous layer to combine extracted features into final predictions.
Cross-entropy loss
Standard loss function for multi-class classification problems measuring the divergence between predicted probabilities and true labels.
Stratified sampling
Sampling method preserving class distribution when splitting the dataset into training, validation, and test sets.
Grad-CAM
Visualization technique generating heatmaps to interpret CNN decisions by identifying image regions influencing the classification.
Top-k accuracy
Performance metric considering a prediction correct if the true label appears among the model's k most probable predictions.