Microsoft AI-900 Quick Notes

Common Types of Computer Vision Workloads

• Image classification model helps to classify images based on their content.

• Object detection model helps to identify objects and their boundaries within the image.

• Semantic segmentation model helps classify pixels to the objects they belong to.

• Face detection is a Computer vision technique that helps detect and recognize people's faces.

• Image Analysis helps extract information from the images, tag them, and create a descriptive image summary.

1. To build a personal virtual assistant - Azure Bot Service

2. To Train a Model : You have to split your data into two sets: the first is for model training and the second for model testing. If you are using Automated machine learning, it automatically does that for you as part of data preparation and model training.

3. For Numeric Prediction : Regression Model

4. Azure Bot Service facilitates access to the knowledge base.

5. Custom vision service helps create your computer vision model.

6. Text Analytics helps analyze text documents, detect documents' language, extract key phrases, determine entities, and provide sentiment analysis.

7. Language Understanding Intelligent Service (LUIS) helps understand voice or text command.

Azure ML Studio has three main authoring tools on its home page:

1. Notebook

2. Automated ML

3. Designer

For language model training, we need to provide the following key elements: Entities, Intents, and Utterance. We can achieve this by using the Azure Cognitive service LUIS portal.

1. The Entity is the word or phrase that is the focus of the utterance, as the word "light" in the utterance "Turn the lights on.”

2. The Intent is the action or task that the user wants to execute. It reflects in utterance as a goal or purpose. We can define intent as "TurnOn" in the utterance "Turn the lights on.”

3. The Utterance is the user's input that your model needs to interpret, like "Turn the lights on" or "Turn on the lights".

Microsoft recognizes six principles of responsible AI:

• Fairness,

• Reliability and safety,

• Privacy and security,

• Transparency,

• Inclusiveness

• Accountability.

Object detection is the form of ML that helps to recognize objects on the images. Each recognizable object will be put in the bounding box with the class name and probability score.

Azure ML uses model evaluation for the measurement of the trained model accuracy. For regression models Evaluate Model module provides the following five metrics:

1. Mean absolute error (MAE),

2. Root mean squared error (RMSE),

3. Relative absolute error (RAE),

4. Relative squared error (RSE),

5. Coefficient of determination (R2).

Regression Model Metrics :

Root Mean Squared Error (RMSE) is the regression model evaluation metrics. It represents the square Root from the squared mean of the errors between predicted and actual values.

Mean absolute error (MAE) is the regression model evaluation metrics. It produces the score that measures how close the model is to the actual values — the lower score, the better the model performance.

Coefficient of determination or R2 is the regression model evaluation metrics. It reflects the model performance: the closer R2 to 1 - the better the model fits the data.

Classification Model Metrics :

Accuracy

Recall

Clustering model Metrics :

Number of Points

Combined Evaluation

Sentiment Analysis is a Text Analytics service that helps analyze text and returns sentiment scores (between 0 and 1) for each sentence.

A score close to 0 means a negative sentiment, and a score close to 1 means positive. In cases with a neutral or undefined sentiment, the score is 0.5.

When the application processes images, it uses Semantic segmentation to classify pixels that belong to the particular object (in our case, flooded areas) and highlights them.

• The process of normalization brings numeric features to a common scale.

• Feature engineering is the method of creating new features based on the existing ones.

  → After we bring data for model training or ingest data, the next stage is the Data transformation. Data transformation or data pre-processing usually includes the following steps: feature selection, data cleaning, and data normalization.

  → In Azure ML Designer, we need to drag-and-drop the "Select Columns in Dataset" module from the Data Transformation section. Then on the right-side panel, we can select all the features we want to use for the model training.

Custom vision is one of the Computer Vision tasks. Custom vision service helps create your own computer vision model. There are three main performance metrics for the Custom vision models: Precision, Recall, and Average Precision (AP).

* Precision defines the percentage of the class predictions that the model makes correct. For example, if the model predicts that ten images are bananas, and there are actually only seven bananas, the model precision is 70%.

* Recall l defines the percentage of the class identification that the model makes correct. For example, if there are ten apple images, and the model identifies only eight, the model recall is 80%.

* Average Precision (AP) is the combined metrics of both Precision and Recall.

The principle of Fairness directs AI solutions to treat everybody fairly, independently from gender, race, or any bias.

After we ingest the data, we need to do a data preparation or transformation before supplying it for model training. There are four typical steps for data transformation such as

• Feature selection,

• Finding and removing data outliers,

• Impute missing values, and

• Normalize numeric features.

  → Split data & ML algorithm selection is coming after data transformation.

  Computer Vision service is one of the main areas of Artificial Intelligence. It belongs to the group of Azure Computer vision solutions such as Computer Vision service, Custom Vision Service, Face service, and Form Recognizer.

  Computer Vision service works with images. This service brings sense to the image pixels by using them as features for ML models. These predefined models help categorize and classify images, detect and recognize objects, tag, and identify them. Computer Vision can "read" a text in images in 25 languages and recognize landmarks.

Form Recognizer service is one of the Azure Computer vision solutions additional to Computer Vision service, Custom Vision Service and Face service.

For automated document processing, Form Recognizer uses two models: Custom Model and a pre-build receipt model.

• With the Custom Model approach, you train the Form Recognizer model based on your own form and data. You just need only 5 samples of your form to start.

• A pre-build receipt model is a Form Recognizer default model that is trained to work with receipts. It helps recognize receipts and extract data from them.

The confusion matrix provides a tabulated view of predicted and actual values for each class. If we are predicting the classification for four classes, our confusion matrix will have a 4x4 size.

A binary confusion matrix is divided into four squares that represent the following values:

• True positive (TP) - the number of positive cases that the model predicted right.

• True negative (TN) - the number of negative cases that the model predicted right.

• False positive (FP) - the number of positive cases that the model falsely predicted right.

• False negative (FN) - the number of negative cases that the model falsely predicted right.

Recall metric defines how many positive cases that the model predicted are actually predicted right. We can calculate this metric using the following formula: TP/(TP+FN).

A formula for Accuracy metric calculation is (TP+TN)/Total number of cases.

A formula for Precision metric calculation is TP/(TP+FP).

A formula for F1 metric calculation is 2TP/(2TP+FP+FN).

Selectivity : this expression is for Selectivity (or true negative rate) metric calculation: TN/(TN+FP).

Area Under Curve (AUC) is the model performance metrics for classification models. For binary classification models, the AUC value of 0.5 represents the random predictions. The model predictions are the same as randomly selected values of "Yes" or "No."

If the AUC value is below 0.5, the model performance is worse than random. Ideally, the best-fitted model has a value of 1. Such an ideal model predicts all the values correctly.

Four types of Compute resources you can use in Azure Machine Learning Studio :

1. Compute Clusters

2. Kubernetes Clusters

3. Attached Compute

4. Compute Instances

An anomaly detection service ingests the data and automatically selects the best ML model for the identification of the possible data irregularities. The service alerts the customers as soon as such anomalies arise.

Feature selection helps us to narrow down the features that are important for our label prediction and discard all features that don’t play or play a minimal role in a label prediction. As a result, our trained model and prediction will be more efficient.

Model Explain Ability

Most businesses run on trust and being able to open th ML black box. This helps build transparency and trust. In heavily regulated industries like healthcare and banking, it is critical to comply with regulations and best practices. One key aspect of this is understanding the relationship between input variables (features) and model output. Knowing both the magnitude and direction of the impact each feature (feature importance) has on the predicted value helps better understand and explain the model. With model explain ability, we enable you to understand feature importance as part of automated ML runs.