ML terminologies
| Term . | Description . |
|---|---|
| AI | AI refers to the simulation of human intelligence in machines that are programmed to think and act like humans. These machines use algorithms and data to perform tasks, such as recognizing patterns, learning from experience, making decisions, and solving problems. |
| ML | ML is a subfield of AI that involves training algorithms to make predictions or decisions based on data. ML algorithms use statistical models and algorithms to analyze data, learn from that data, and then make a prediction or classification about new data. The goal of ML is to automatically improve the performance of the algorithm over time by learning from the input data. This allows the algorithm to become more accurate in its predictions and decisions as it is exposed to more data. |
| Deep learning | Deep learning is a subset of ML that uses artificial neural networks with multiple layers to process and analyze complex data. These deep neural networks are trained to identify patterns in large amounts of data, such as images, speech, and text, and then make predictions or decisions based on that data. Deep learning has led to breakthroughs in many applications of AI, including computer vision, speech recognition, and natural language processing. |
| Transfer learning | Transfer learning is an ML technique where a model trained on one task is reused as the starting point for a model on a related task. The idea is that the knowledge gained from solving one problem can be useful for solving a similar problem, allowing for faster training times and improved performance compared with training a model from scratch. Transfer learning is commonly used in computer vision and natural language processing, where models trained on large, general-purpose data sets can be fine-tuned for specific tasks with smaller amounts of data. |
| Supervised learning | Supervised learning is a type of ML technique in which an algorithm learns to make predictions or decisions by training on labeled data. The goal of supervised learning is to learn a mapping function that can accurately predict output values for new input data. The learning process involves adjusting the parameters of the algorithm based on the errors between the predicted and actual output values. The most common types of supervised learning algorithms are regression and classification. Common types of supervised learning algorithms are decision trees, linear regression, logistic regression, random forest, support vector machine, and others. |
| Unsupervised learning | Unsupervised learning is a type of ML technique in which an algorithm learns to identify patterns and relationships in data without being explicitly trained on labeled data. In unsupervised learning, the algorithm is presented with a set of input data, and it learns to discover patterns and structure within the data on its own. Some common types of unsupervised learning algorithms are clustering, dimensionality reduction, and anomaly detection. |
| Semisupervised learning | Semisupervised learning is a type of ML technique that combines both labeled and unlabeled data for training a model. In semisupervised learning, the algorithm is presented with a small amount of labeled data and a large amount of unlabeled data, and it learns to make predictions by using both types of data. |
| CNN | CNN is a type of deep learning artificial neural network used for image and video recognition, as well as natural language processing tasks. It is designed to process data through multiple layers of arrays, called convolutions, which learn features from the input data, reducing its dimensionality and allowing for pattern recognition. The use of pooling layers and fully connected layers allows a CNN to make predictions based on the features it has learned, making it a powerful tool in computer vision and NLP. |
| RNN | RNNs are a type of artificial neural network used for processing sequential data such as time series, natural language text, and speech. Unlike traditional feedforward neural networks, RNNs have a feedback loop that allows information to persist, allowing the network to maintain information from past inputs and use it in conjunction with current inputs in making predictions. This makes RNNs well suited for tasks in which the current output is dependent on the previous inputs, such as language generation, speech recognition, and machine translation. |
| NLP | NLP is a field of AI concerned with enabling computers to understand, interpret, and generate human language. NLP encompasses a wide range of tasks, including text classification, sentiment analysis, machine translation, named entity recognition, and question answering, among others. To perform NLP tasks, techniques from computational linguistics, computer science, and ML are combined to develop algorithms that can process and analyze large amounts of text and speech data. The goal of NLP is to enable computers to understand and process human language in a way that is similar to how humans do, making it a key component in the development of artificial general intelligence. |
| GNNs | GNNs are a type of neural network used for processing data structured as graphs, such as social networks, molecule structures, and road networks. GNNs are designed to handle the non-Euclidean structure of graph data by updating node representations based on the representations of their neighboring nodes, and by propagating information through edges in the graph. This allows GNNs to learn rich representations of the graph structure, making them useful for tasks such as node classification, graph classification, and link prediction. GNNs have been shown to outperform traditional neural networks on many graph-based tasks and have become a rapidly growing area of research in the field of deep learning. |
| Algorithm | An algorithm is a set of instructions or rules that a computer follows to perform a specific task. In ML, algorithms are used to analyze data and make predictions based on that data. |
| Model | A model is a representation of a system or process that can be used to make predictions. In ML, models are trained on data and then used to make predictions about new data. |
| Parameter | A parameter is a configuration variable that is internal to the model and whose value is learned from the data during training. It is a part of the model that can be adjusted to optimize the performance of the algorithm. For example, in linear regression, the parameters are the coefficients of the model that are learned during training to minimize the error between the predicted values and the actual values. In neural networks, the parameters include the weights and biases that are learned through backpropagation during training. |
| Hyperparameter | A hyperparameter is a setting or configuration that is external to the model and is used to control its learning process. Unlike model parameters that are learned during training, hyperparameters are set before the learning process begins and remain constant throughout training. Examples of hyperparameters include the learning rate of the model, the number of hidden layers in a neural network, the number of decision trees in a random forest model, and regularization parameters. The selection of appropriate hyperparameters can have a significant impact on the performance of the model. |
| Training data | Training data is a set of data that is used to train an ML model. The model uses the training data to learn how to make predictions. In this data set, the features and outcomes are known, and therefore, the model can learn how to predict the outcomes from the features. |
| Validation data | Validation data are a set of data that are used to evaluate the performance of an ML model. The model is tested on the validation data to see how well it makes predictions. |
| Test data | Test data are a set of data that provide a final, real-world check of an unseen data set to confirm that the ML algorithm was trained effectively. Preferably, the test data set should represent an external, multi-institutional data set. Furthermore, the outcomes are known in this data set but are not used in training and therefore can be used to test the model. |
| Overfitting | Overfitting occurs when an ML model is too closely fit to the training data, resulting in poor performance on new data. |
| Underfitting | Underfitting occurs when an ML model is too simple to capture the complexity of the data, resulting in poor performance on both the training and validation data. |
| Bias | Bias occurs when an ML model makes systematic errors in its predictions. Bias can result from errors in the data or from a poorly designed model. |
| Accuracy | Accuracy is a measure of how well an ML model makes predictions compared with the actual value. |
| ROC/AUC | ROC is a commonly used evaluation metric in binary classification problems. It plots the true-positive rate against the false-positive rate at various thresholds, providing a visual representation of the classifier’s performance. AUC is a single number summary of the ROC curve, representing the overall performance of a classifier. It measures the area under the ROC curve and ranges from 0 to 1, with a higher AUC indicating a better classifier. An AUC of 0.5 indicates a classifier with no discrimination power, whereas an AUC of 1 means perfect discrimination. |
| Precision | Precision is a measure of how many of the predictions made by an ML model are correct. |
| Recall | Recall is a measure of how many of the actual values are correctly predicted by an ML model. |
| F1 score | F1 score is a commonly used metric to evaluate the performance of binary and multiclass classification algorithms. It is the harmonic mean of precision and recall, in which precision is the number of true-positive predictions divided by the sum of true-positive and false-positive predictions, and recall is the number of true-positive predictions divided by the sum of true-positive and false-negative predictions. The F1 score ranges from 0 to 1, with a higher score indicating better performance. It balances precision and recall, making it a useful metric when the cost of false negatives and false positives is not equal. |
| Term . | Description . |
|---|---|
| AI | AI refers to the simulation of human intelligence in machines that are programmed to think and act like humans. These machines use algorithms and data to perform tasks, such as recognizing patterns, learning from experience, making decisions, and solving problems. |
| ML | ML is a subfield of AI that involves training algorithms to make predictions or decisions based on data. ML algorithms use statistical models and algorithms to analyze data, learn from that data, and then make a prediction or classification about new data. The goal of ML is to automatically improve the performance of the algorithm over time by learning from the input data. This allows the algorithm to become more accurate in its predictions and decisions as it is exposed to more data. |
| Deep learning | Deep learning is a subset of ML that uses artificial neural networks with multiple layers to process and analyze complex data. These deep neural networks are trained to identify patterns in large amounts of data, such as images, speech, and text, and then make predictions or decisions based on that data. Deep learning has led to breakthroughs in many applications of AI, including computer vision, speech recognition, and natural language processing. |
| Transfer learning | Transfer learning is an ML technique where a model trained on one task is reused as the starting point for a model on a related task. The idea is that the knowledge gained from solving one problem can be useful for solving a similar problem, allowing for faster training times and improved performance compared with training a model from scratch. Transfer learning is commonly used in computer vision and natural language processing, where models trained on large, general-purpose data sets can be fine-tuned for specific tasks with smaller amounts of data. |
| Supervised learning | Supervised learning is a type of ML technique in which an algorithm learns to make predictions or decisions by training on labeled data. The goal of supervised learning is to learn a mapping function that can accurately predict output values for new input data. The learning process involves adjusting the parameters of the algorithm based on the errors between the predicted and actual output values. The most common types of supervised learning algorithms are regression and classification. Common types of supervised learning algorithms are decision trees, linear regression, logistic regression, random forest, support vector machine, and others. |
| Unsupervised learning | Unsupervised learning is a type of ML technique in which an algorithm learns to identify patterns and relationships in data without being explicitly trained on labeled data. In unsupervised learning, the algorithm is presented with a set of input data, and it learns to discover patterns and structure within the data on its own. Some common types of unsupervised learning algorithms are clustering, dimensionality reduction, and anomaly detection. |
| Semisupervised learning | Semisupervised learning is a type of ML technique that combines both labeled and unlabeled data for training a model. In semisupervised learning, the algorithm is presented with a small amount of labeled data and a large amount of unlabeled data, and it learns to make predictions by using both types of data. |
| CNN | CNN is a type of deep learning artificial neural network used for image and video recognition, as well as natural language processing tasks. It is designed to process data through multiple layers of arrays, called convolutions, which learn features from the input data, reducing its dimensionality and allowing for pattern recognition. The use of pooling layers and fully connected layers allows a CNN to make predictions based on the features it has learned, making it a powerful tool in computer vision and NLP. |
| RNN | RNNs are a type of artificial neural network used for processing sequential data such as time series, natural language text, and speech. Unlike traditional feedforward neural networks, RNNs have a feedback loop that allows information to persist, allowing the network to maintain information from past inputs and use it in conjunction with current inputs in making predictions. This makes RNNs well suited for tasks in which the current output is dependent on the previous inputs, such as language generation, speech recognition, and machine translation. |
| NLP | NLP is a field of AI concerned with enabling computers to understand, interpret, and generate human language. NLP encompasses a wide range of tasks, including text classification, sentiment analysis, machine translation, named entity recognition, and question answering, among others. To perform NLP tasks, techniques from computational linguistics, computer science, and ML are combined to develop algorithms that can process and analyze large amounts of text and speech data. The goal of NLP is to enable computers to understand and process human language in a way that is similar to how humans do, making it a key component in the development of artificial general intelligence. |
| GNNs | GNNs are a type of neural network used for processing data structured as graphs, such as social networks, molecule structures, and road networks. GNNs are designed to handle the non-Euclidean structure of graph data by updating node representations based on the representations of their neighboring nodes, and by propagating information through edges in the graph. This allows GNNs to learn rich representations of the graph structure, making them useful for tasks such as node classification, graph classification, and link prediction. GNNs have been shown to outperform traditional neural networks on many graph-based tasks and have become a rapidly growing area of research in the field of deep learning. |
| Algorithm | An algorithm is a set of instructions or rules that a computer follows to perform a specific task. In ML, algorithms are used to analyze data and make predictions based on that data. |
| Model | A model is a representation of a system or process that can be used to make predictions. In ML, models are trained on data and then used to make predictions about new data. |
| Parameter | A parameter is a configuration variable that is internal to the model and whose value is learned from the data during training. It is a part of the model that can be adjusted to optimize the performance of the algorithm. For example, in linear regression, the parameters are the coefficients of the model that are learned during training to minimize the error between the predicted values and the actual values. In neural networks, the parameters include the weights and biases that are learned through backpropagation during training. |
| Hyperparameter | A hyperparameter is a setting or configuration that is external to the model and is used to control its learning process. Unlike model parameters that are learned during training, hyperparameters are set before the learning process begins and remain constant throughout training. Examples of hyperparameters include the learning rate of the model, the number of hidden layers in a neural network, the number of decision trees in a random forest model, and regularization parameters. The selection of appropriate hyperparameters can have a significant impact on the performance of the model. |
| Training data | Training data is a set of data that is used to train an ML model. The model uses the training data to learn how to make predictions. In this data set, the features and outcomes are known, and therefore, the model can learn how to predict the outcomes from the features. |
| Validation data | Validation data are a set of data that are used to evaluate the performance of an ML model. The model is tested on the validation data to see how well it makes predictions. |
| Test data | Test data are a set of data that provide a final, real-world check of an unseen data set to confirm that the ML algorithm was trained effectively. Preferably, the test data set should represent an external, multi-institutional data set. Furthermore, the outcomes are known in this data set but are not used in training and therefore can be used to test the model. |
| Overfitting | Overfitting occurs when an ML model is too closely fit to the training data, resulting in poor performance on new data. |
| Underfitting | Underfitting occurs when an ML model is too simple to capture the complexity of the data, resulting in poor performance on both the training and validation data. |
| Bias | Bias occurs when an ML model makes systematic errors in its predictions. Bias can result from errors in the data or from a poorly designed model. |
| Accuracy | Accuracy is a measure of how well an ML model makes predictions compared with the actual value. |
| ROC/AUC | ROC is a commonly used evaluation metric in binary classification problems. It plots the true-positive rate against the false-positive rate at various thresholds, providing a visual representation of the classifier’s performance. AUC is a single number summary of the ROC curve, representing the overall performance of a classifier. It measures the area under the ROC curve and ranges from 0 to 1, with a higher AUC indicating a better classifier. An AUC of 0.5 indicates a classifier with no discrimination power, whereas an AUC of 1 means perfect discrimination. |
| Precision | Precision is a measure of how many of the predictions made by an ML model are correct. |
| Recall | Recall is a measure of how many of the actual values are correctly predicted by an ML model. |
| F1 score | F1 score is a commonly used metric to evaluate the performance of binary and multiclass classification algorithms. It is the harmonic mean of precision and recall, in which precision is the number of true-positive predictions divided by the sum of true-positive and false-positive predictions, and recall is the number of true-positive predictions divided by the sum of true-positive and false-negative predictions. The F1 score ranges from 0 to 1, with a higher score indicating better performance. It balances precision and recall, making it a useful metric when the cost of false negatives and false positives is not equal. |
AI, artificial intelligence; AUC, area under the curve; CNN, convolutional neural network; GNN, graph neural network; NLP, natural language processing; RNN, recurrent neural network; ROC, receiver operating characteristic.