Azure AI Terms 01 Flashcards
(100 cards)
1
Q
CLU
A
Conversational language understanding. A feature in Azure Cognitive Service for Language.
2
Q
CNN
A
Convolutional neural network. A deep learning technique in AI used for image classification.
3
Q
GPT-3
A
Generative Pre-trained Transformer 3.
4
Q
LUIS
A
Language Understanding Service.
5
Q
MAE
A
Mean absolute error. When evaluating machine learning regression models, this measures how close the predictions are to the actual vaules; lower is better.
6
Q
mAP
A
Mean Average Precision. In AI classification, an overall metric that takes into account both precision and recall across all classes.
7
Q
MIND
A
Microsoft News Dataset. A large-scale dataset for news recommendation research that’s available in Azure Open Datasets.
8
Q
MLOps
A
Machine Learning Operations. The Machine Learning analog of DevOps, as used in Azure Machine Learning (Azure ML).
9
Q
MRF
A
Markov Random Field. For our purposes a technique sometimes used in computer vision and machine learning.
10
Q
MZOE
A
Mean zero one error. When evaluating machine learning regression models, 1 if the prediction was correct, otherwise 0.
11
Q
NER
A
Named entity recognition. A feature in Azure Cognitive Service for Language.
12
Q
NLU
A
Natural language understanding.
13
Q
NMT
A
Neural machine translation. The model used by Azure Translator.
14
Q
R-squared or R2
A
Coefficient of determination. When evaluating machine learning regression models, a measure of the variance from the mean in its predictions; the closer to 1, the better the model is performing.
15
Q
RAE
A
Relative absolute error. When evaluating machine learning regression models, the relative absolute distance between expected and actual values; lower is better.
16
Q
RMSE
A
Root mean squared error. When evaluating machine learning regression models, the square root of the average squared distance between the actual and the predicted values; lower is better.
17
Q
RSE
A
Relative squared error. When evaluating machine learning regression models, the total squared error of the predicted values by dividing by the total squared error of the actual values; lower is better.
18
Q
SACH
A
Structured Analysis of Competing Hypotheses.
19
Q
AI agent
A
Synonym for bot.
20
Q
AI enrichment
A
Embedded image and natural language processing in an AI knowledge mining pipeline that extracts text and information from content that can’t otherwise be indexed for full text search.
21
Q
Anomaly detection
A
Synonym for outlier detection.
22
Q
Attached Compute
A
In Azure Machine Learning (Azure ML), links to existing Azure compute resources, such as Virtual Machines or Azure Databricks clusters.
23
Q
AutoML
A
The automated machine learning process in Azure Machine Learning (Azure ML).
24
Q
Bot
A
Synonym for AI agent.
25
Classification
A type of supervised machine learning that predicts categories or classes.
26
Coefficient of determination (R-squared or R2)
When evaluating machine learning regression models, a measure of the variance from the mean in its predictions; the closer to 1, the better the model is performing.
27
Compute Cluster
In Azure Machine Learning (Azure ML), a scalable cluster of virtual machines for training the model and running experiments.
28
Compute Instance
In Azure Machine Learning (Azure ML), a development virtual machine used to process data and build the model.
29
Conversational AI
The process of building AI agents to take part in conversations with humans, commonly done with chatbots. See Azure Conversational Language Understanding.
30
Conversational language understanding (CLU)
A feature in Azure Cognitive Service for Language.
31
Convolutional neural network (CNN)
A deep learning technique in AI used for image classification.
32
Data asset
In Azure Machine Learning (Azure ML), the data for model training and other operations.
33
Entity
In Azure Cognitive Service for Language, the item to which the utterance refers. For example, if the utterance is "Turn on the light", the intent is something like ''TurnOn'' and the entity is the light.
34
Entity linking
In Azure Cognitive Service for Language, the process of identifying and disambiguating the entities found in a text.
35
Face detection
The process of identifying regions of an image that contain a human face, typically by returning bounding box coordinates. Contrast with facial analysis which identifies facial landmarks like nose and eyes, and facial recognition which identifies known individuals.
36
Facial analysis
The process of identifying facial landmarks like nose and eyes. Contrast with face detection which identifies regions of an image that contain faces, and facial recognition which identifies known individuals.
37
Facial recognition
The process of identifying known individuals from their facial features, based on multiple images of that individual which have already been classified. Contrast with face detection which identifies regions of an image that contain faces, and facial analysis which identifies facial landmarks like nose and eyes.
38
Feature engineering
In machine learning, the process of creating new features from raw data to increase the predictive power of the model.
39
Forecast
In machine learning, a synonym for prediction.
40
Image classification
A machine learning based form of computer vision in which a model is trained to categorize images based on the primary subject matter they contain. Contrast with object detection which classifies individual objects in the image.
41
Inference Cluster
In Azure Machine Learning (Azure ML), a deployment target for predictive services that uses a trained model in Azure Kubernetes Service (AKS).
42
Intent
In Azure Cognitive Service for Language, the purpose or goal expressed in a user's utterance, such as the "Turn on the light". In this case the intent is something like ''TurnOn'' and the entity is the light.
43
K-means clustering
A machine learning method that aims to partition //n// observations into //k// clusters in which each observation belongs to the cluster with the nearest mean (cluster centers or cluster centroid), serving as a prototype of the cluster.
44
Knowledge mining
Extracting information from large volumes of often unstructured data.
45
Lemmatization
In text analytics, a dictionary-based text normalization method that reduces a word to its canonical or dictionary form. Contrast with stemming which is a rule-based methdod for finding the root/base word.
46
Machine Learning Operations (MLOps)
The Machine Learning analog of DevOps, as used in Azure Machine Learning (Azure ML).
47
Markov Random Field (MRF)
For our purposes a technique sometimes used in computer vision and machine learning.
48
Mean absolute error (MAE)
When evaluating machine learning regression models, this measures how close the predictions are to the actual vaules; lower is better.
49
Mean Average Precision (mAP)
In AI classification, an overall metric that takes into account both precision and recall across all classes, similar to the F1 score.
50
Mean zero one error (MZOE)
When evaluating machine learning regression models, 1 if the prediction was correct, otherwise 0.
51
Microsoft News Dataset (MIND)
A large-scale dataset for news recommendation research that's available in Azure Open Datasets.
52
N-gram
In text analytics, a multi-term phrase used for frequency analysis. A two-word phrase is a bi-gram, a three-word phrase is a tri-gram, and so on.
53
Named entity recognition (NER)
A feature in Azure Cognitive Service for Language.
54
Neural machine translation (NMT)
The model used by Azure Translator.
55
Object detection
A machine learning based form of computer vision that classifies individual objects within the image and returns the coordinates of their bounding boxes, as opposed to image classification which categorizes the entire image based on the primary subject matter.
56
Ordered Categorical Variable
Another term for ordinal variable.
57
Ordinal Variable
A categorical variable with order but not amount such as one with "low", "medium", and "high" as values. Contrast with with nominal variable and continuous variable.
58
Outlier detection
Synonym for anomaly detection.
59
Overlap Threshold
In Azure Cognitive Service for Vision, the minimum percentage of overlap between predicted bounding boxes and ground truth boxes to be considered a correct prediction.
60
Phrase
In Azure Cognitive Service for Vision, a single result that describes an image. Each phrase has a confidence score. Contrast with "tag" which is a word that summarizes one attribute of an image.
61
Precision
When evaluating machine learning classification models, Precision = #True_Positive / (#True_Positive + #False_Positive). Measures how precise/accurate your model is. It's the ratio between the correctly identified positives (true positives) and all identified positives. The precision metric reveals how many of the predicted classes are correctly labeled. Contrast with recall, which is the ratio between the predicted true positives and what was actually tagged as positive. Precision and recall are combined in the F1 score and mean average precision (mAP).
62
Prediction
In machine learning, a synonym for forecast.
63
Probability Threshold
In Azure Cognitive Service for Vision, the minimum probability score for a prediction to be valid when calculating precision and recall.
64
Prosodic unit
When Azure Speech breaks down a phonetic transcription to synthesize speech, a prosodic unit is a phrase, clause, or sentence which is used to create phonemes that will be converted to audio format.
65
Recall
When evaluating machine learning classification models, Recall = #True_Positive / (#True_Positive + #False_Negative). This measures the model's ability to predict actual positive classes. It's the ratio between the predicted true positives and what was actually tagged as positive. The recall metric reveals how many of the predicted classes are correct. Contrast with precision, which is the ratio between the correctly identified positives (true positives) and all identified positives. Recall and precision are combined in the F1 score and mean average precision (mAP).
66
Regression
A type of supervised machine learning that predicts numeric values. See also time series forecasting.
67
Regression Tree
Decision tree model where the target variable can take continuous values. Contrast with "classification tree".
68
Reinforcement learning
A kind of machine learning technique that uses feedback to improve the outcomes from the model.
69
Relative absolute error (RAE)
When evaluating machine learning regression models, the relative absolute distance between expected and actual values; lower is better.
70
Relative squared error (RSE)
When evaluating machine learning regression models, the total squared error of the predicted values by dividing by the total squared error of the actual values; lower is better.
71
Residual values
When evaluating machine learning regression models, the difference between actual and predicted values.
72
Root mean squared error (RMSE)
When evaluating machine learning regression models, the square root of the average squared distance between the actual and the predicted values; lower is better.
73
Semantic segmentation
In Azure Cognitive Service for Vision, the ability to classify individual pixels in an image depending on the object that they represent.
74
Skillset
In an AI knowledge mining pipeline, the definition of the operations that extract and enrich data to make it searchable.
75
Smart tagging
A feature of Azure's Computer Vision service that suggests classes and bounding boxes for images being added to the training dataset.
76
Stemming
In text analytics, a rule-based text normalization method that produces variants of a root/base word. Contrast with lemmatization which is a dictionary-based method for reducing a word to its canonical or dictianary form.
77
Tag
In Azure Cognitive Service for Vision, a word that summarizes an attribute of an image. Contrast with "phrase" which is a short description of the entire image.
78
Time series forecasting
A type of supervised machine learning that predicts numeric values at a future point in time. See also regression.
79
Utterance
In Azure Cognitive Service for Language, an example of something a user might say such as "Turn on the light" which the application must interpret as an entity (the light) and an intent (something like ''TurnOn'').
80
Vectorized model
In text analytics, a model that captures semantic relationships between words by assigning them to locations in n-dimensional space, where commonly related words are closer to each other.
81
F1 score
When evaluating machine learning classification models, F1 Score = 2 * Precision * Recall / (Precision + Recall). It's a way of combining precision and recall into a single score, like mean average precision (mAP).
82
Confusion matrix
When evaluating machine learning classification models, a grid showing the number of true positives, true negatives, false positives, and false negatives.
83
F-score
When evaluating machine learning classification models, a combination of precision and recall as a weighted mean value. May be the same as F1 score.
84
Accuracy
When evaluating machine learning classification models, Accuracy = #True_Positives + #True_Negatives / #Predictions. Contrast with precision and recall.
85
Area under curve (AUC)
When evaluating machine learning classification models, AUC = true positive rate divided by true negative rate.
86
Azure Data Science Virtual Machine (DSVM)
A VM for Azure Machine Learning (Azure ML) that comes with ML tools and R and Python installed. It's intended for users of Microsoft SQL Server ML.
87
DSVM
Azure Data Science Virtual Machine. A VM for Azure Machine Learning (Azure ML) that comes with ML tools and R and Python installed. It's intended for users of Microsoft SQL Server ML.
88
Compute target
A compute resource in an Azure Machine Learning (Azure ML) workspace, such as a virtual machine or Azure Kubernetes Service (AKS) cluster.
89
Hyperparameter
A parameter for a machine learning model that controls how the training is done.
90
Inferencing
The use of a trained machine learning model to make predictions for new unseen data.
91
Open Neural Network Exchange (ONNX)
An open standard for machine learning interoperability.
92
ONNX
Open Neural Network Exchange. An open standard for machine learning interoperability.
93
Testing dataset
Synonym for validation dataset.
94
Validation dataset
Synonym for testing dataset.
95
Multi-service resource
In the context of Azure Cognitive Services, the Cognitive Services resource itself rather than a single-service resource like Custom Vision or Personalizer.
96
Single-service resource
In the context of Azure Cognitive Services, a specialized resource such as Text Analytics or Form Recognizer rather than the multi-service Cognitive Services resource.
97
Facial landmark
In facial analysis, the location of a part of a face such as the center of the left eye.
98
PosTag
Part of speech tagging
99
Semantic modeling
The process of identifying relationships between words.
100
Turn
In the Azure Bot service, one cycle of the bot receiving a message and sending a response back to the user.
