SLPs and Machine

Flashcard 1

Concept: Machine Learning (ML)

Definition: A branch of computer science focused on the development of models and algorithms that “learn” from data in order to improve predictive performance1 .

Example: Developing a model that learns from speech recordings and other patient data to predict whether a patient will respond to a specific type of aphasia therapy2 ....

Flashcard 2

◦

Concept: Artificial Intelligence (AI)

◦

Definition: A broader field than ML that focuses on equipping machines with capacities to approximate humanlike intelligence (e.g., communication, reasoning, perception)1 .

◦

Example: Creating a system that can understand spoken language and provide automated feedback to a person practicing speech sounds1 ....

•

Flashcard 3

◦

Concept: Algorithm

◦

Definition: A sequence of rules executed by a computer to solve a problem5 ....

◦

Example: A one-layer decision tree that classifies a patient as having aphasia if their Western Aphasia Battery Aphasia Quotient is less than or equal to 93.7, and within normal limits otherwise5 .

•

Flashcard 4

◦

Concept: Dataset

◦

Definition: A collection of data from which ML models learn5 ....

◦

Example: A dataset containing information about multiple patients, including their aphasia severity, brain connectivity measures, and whether or not they improved after speech-language therapy5 .

•

Flashcard 5

◦

Concept: Model

◦

Definition: A program that can find patterns or make decisions from previously unseen data. It represents what algorithms learned from training data6 ....

◦

Example: A trained neural network that, when given a new patient's speech data, predicts the severity of their dysarthria1 ....

•

Flashcard 6

◦

Concept: Feature

◦

Definition: An independent property used as input to the model6 ....

◦

Example: In a model predicting aphasia severity, scores from different subtests of a language assessment (e.g., naming, repetition) could be features8 .

•

Flashcard 7

◦

Concept: Label/Target

◦

Definition: The predetermined outcome to be predicted by the model6 ....

◦

Example: In a classification task, the label could be the diagnostic category (e.g., fluent vs. non-fluent aphasia). In a regression task, the target could be a continuous test score (e.g., WAB-R AQ)8 ....

•

Flashcard 8

◦

Concept: Training

◦

Definition: The process whereby the model learns the hidden pattern and relationship existing in the features and how the features are related to target value11 ....

◦

Example: Feeding a model a dataset of speech samples and corresponding diagnostic labels so that the model can learn to associate specific speech patterns with different diagnoses11 .

•

Flashcard 9

◦

Concept: Testing

◦

Definition: The process of evaluating final performance of a model with optimal configuration on some other, previously unseen data12 ....

◦

Example: After training a model to classify speech disorders, evaluating its accuracy on a new set of speech samples from patients the model has never seen before13 .

•

Flashcard 10

◦

Concept: Classification

◦

Definition: The goal of a classification task is to predict a discrete label from two or more categories based on input features9 ....

◦

Example: Predicting whether a child has dyslexia or not based on their performance on literacy assessments9 .

•

Flashcard 11

◦

Concept: Regression

◦

Definition: The goal of a regression task is to predict a continuous numeric value based on input features10 ....

◦

Example: Predicting a patient's score on the Western Aphasia Battery–Revised Aphasia Quotient (WAB-R AQ) based on neuroimaging data10 ....

•

Flashcard 12

◦

Concept: Supervised Learning

◦

Definition: Supervised learning algorithms are characterized by labeled input data, meaning the training data includes both features and the correct output (label or target)16 ....

◦

Example: Training a model to classify different types of speech sounds using audio recordings that have been manually labeled with the correct sound16 .

•

Flashcard 13

◦

Concept: Unsupervised Learning

◦

Definition: Unsupervised learning algorithms are trained to detect patterns and similarities in unlabeled data sets, where the training data only contains features without predefined outcomes17 ....

◦

Example: Using clustering algorithms to identify natural groupings of patients with aphasia based on their connected speech characteristics, without pre-defined diagnostic labels19 ....

•

Flashcard 14

◦

Concept: Neural Network

◦

Definition: A powerful ML method inspired by the human brain, consisting of inter-connected groups of artificial neurons that can learn complex functions17 ....

◦

Example: Using a deep neural network to automatically classify different types of voice disorders based on acoustic features of voice recordings21 ....

•

Flashcard 15

◦

Concept: Evaluation Metric

◦

Definition: A measure used to quantify the performance of an ML model based on its predictions compared to the actual outcomes23 ....

◦

Example: Using accuracy (the proportion of correctly predicted diagnoses) to evaluate the performance of a model that classifies speech disorders, or using R-squared (the proportion of variance explained) to evaluate a model that predicts aphasia severity25 ....