Active Learning in Speech Command Classification

These flashcards cover key concepts and terminology related to active learning in the context of speech command classification.

Active Learning

A machine learning approach where the algorithm selectively queries labels for data points to efficiently learn from limited labeled data.

Command Speech Classification

The process of categorizing spoken commands into pre-defined labels or classes.

Dataset Decomposition

The process of splitting a dataset into training, testing, and active learning subsets.

Baseline

A reference point established by training a model on a subset of data to evaluate future improvements.

Random Sampling

A strategy in which random subsets of data are used for training to evaluate model performance.

Uncertainty Sampling

An active learning strategy where the model selects instances it is least confident about to label.

Diversity Sampling

An active learning technique aimed at selecting diverse examples to maximize learning from a limited dataset.

ResNet-18

A convolutional neural network model used for image classification tasks, known for its deep residual learning framework.

Spectrogram

A visual representation of the spectrum of frequencies in a sound signal as they vary with time.

Colab and Google Drive Integration

Linking Google Colab to Google Drive to store large files and save progress during data processing.

Label Frequency Dictionary

A data structure used to track the number of samples associated with different labels in a dataset.

Accuracy Rate

A performance metric indicating the proportion of correct predictions made by a model.

Client Budget Constraints

Limitations on the amount of labeled data that can be annotated based on the client's financial resources.

Training Dataset

A subset of data used to train a model, allowing it to learn patterns and make predictions.

Testing Dataset

A set of data used to evaluate the model's performance after it has been trained.

DataLoader

A utility that provides an iterable over a dataset, often used in training and testing machine learning models.