Examquestions Flashcards
What is Artificial Intelligence?
Machines that perform jobs that mimic human behavior
What is Machine Learning?
Machines that get better at a task without explicit programming
What is deep learning?
Machines that have an artificial neural network inspired by the human brain to solve complex problems
What are key elements to AI?
Machine Learning: foundation of an AI system to learn and predict human behaviour
Anomaly detection: detect outliers
Computer Vision: see like a human
Natural language processing: process human language with context
Conversational AI: hold a conversation like a human
What is a dataset?
logical grouping of units of data that have the same data structure to train
What is Data labeling?
identifying raw data and adding labels to provide context to help the machine learn
What is Supervised Learning?
You know the labels and have a precise output you want
make a prediction
What is Unsupervised learning?
labels are not known and the outcome does not need to be precise
recognize a structure or pattern
What is Reinforcement Learning?
No data and a ML model generates data to reach a goal
Game AI
What are Neural Networks?
algorithm to represent a brain
connections between neurons are weighted with multiple layers
What is deep learning?
A neural network with 3 or more layers
What is feed forward?
Neural Network where connections between nodes dont form a cycle
What is backpropagation? (BP)
backward movement though a neural network to adjust weights to improve the outcome on the next iteration - this is how it learns
What is the ground truth?
the answer/reality you want to reach with the model
What is a loss function?
compares the ground truth to the prediction to determine the error rate
Why are GPUs used for machine learning?
they can perform parallel operations on multiple sets of data - can have thousands or processor cores
What is forecasting?
makes a future prediction with relevant data
“analysis of trends”
What is a prediction?
makes a future prediction without relevant data
uses statistics to predict future outcomes
more guessing than forecasting
What are internal Evaluation metrics for an ML model?
Accuracy, F1 score, Precision, Recall
What are the external Evaluation metrics for an ML model?
metrics used to evaluate the final prediction of the ML model
What is Jupyter Notebook/Lab?
web-based application for authoring documents with live-code
text
visualizations etc.
What is Regression?
process of finding a function to correlate a labeled labeled dataset into continuous variable/number - make a prediction into the future
distance from the vector to the regression line is called error
What is classification?
finding a function to divide a labeled dataset into classes/categories
What is clustering?
grouping unlabeled data based on similarities and differences
What is a confusion matrix?
visualize the model predictions vs ground truth labels (actual)
also called error matrix
What Anomaly detection?
process of finding an outlier in a dataset
What is Computer Vision?
gain level of understanding digital images and video
Convolutional neural network: image and video recognition
recurrent neural network: handwriting or speech recognition
Semantic Segmentation: identify segments or objects
Image analysis: analyse image or video and apply label
What is natural language processing?
understand the context of a corpus
analyzes text and interpret etc.
What is conversational AI?
Chatbots, Voice Assistants etc.
Online Customer Support
HR processes
Health Care
IoT etc.
What is Responsible AI? What are the principles?
ethical, transparent and accountable AI
- Fairness - should treat all people fairly
- Reliability and Safety
- Privacy and Security
- Inclusiveness
- Transparency
- Accountability
What is Azure Cognitive Services?
comprehensive family of AI services and cognitive APIs to help you build intelligent apps
deploy with containers, with not much expertise
What is part of Azure Cognitive Services?
Decision:
Anomaly Detector
Content Moderator
Personaliser
Language:
Language Understanding
QnA Maker
Text Analytics
Translator
Speech:
Speech to Text
Text to Speech
Speech Translation
Speaker Recognition
Vision:
Computer Vision
Custom Vision
Face Detection
What is Knowledge Mining?
Knowledge Mining is a discipline that uses a combination of intelligent services to quickly learn from vast amounts of information
What are text analytics and what is part of it?
Sentiment Analysis
Opinion Mining
Key Phrase Extradiction
Language Detection
Named entity recognition
What are important infos to the Form Recognizer Service?
trained with own data
only needs five sample input forms to start
can use it to extract data from forms
What are the key parts LUIS reads out of a text?
Intention
Entities what parts of the intent is used to determine the answer
utterance: the user input
What is the Azure Bot Service?
Intelligent, serverless bot service that scales on demand - is used for creating, publishing and managing bots
What is Bot Framework Composer?
opensource IDE for developers to create conversational experiences (bots)
has templates and can author, test, provision and manage solutions
What is Azure Machine Learning Services?
Service to run workloads
Jupyter Notebooks
Azure Machine Learning Python: code
MLOps: automation for creation
Azure Machine Learning Designer: no code
Data Labeling Service: ask humans to label data
Responsible Machine Learning
Whats part of author in Azure Machine Learning Studio?
Notebooks: write python code to built models
AutoML: auto building of models (limited)
Designer: No Code Designer to construct ML
Whats part of Assets in Azure Machine Learning Studio?
Datasets
Experiments
Pipelines
Models
Endpoints
What is part of “manage” in Azure Machine Learning Studio?
Compute
Environments
Datastores
Data Labeling
Linked Services
What are the four kinds of compute to choose in Azure Machine Learning Studio?
Compute Instances: development workstations
Compute Clusters: scalable virtual machines
Inference Clusters: deployment targets for services
Attached Compute: links to exisiting azure resources
What options are there for labeling in Machine Learning Studio?
Human labeling
Machine learning assisted data labeling
What are the two ways to build pipelines?
Azure Machine Learning Designer
Azure Machine Learning Python SDK
What is Model Registry?
management tool for your models
What is Azure ML Endpoints?
service that allows you to deploy machine learning models as a web service
If you deploy a model over Endpoints what Compute types are used?
Containers or Kubernetes
What is AutoML?
automates the creation of an ML model
What is Data Guard Rails in AutoML
service that ensures high quality for the input data
What scaling or normalization techniques are available in AutoML?
StandardScaleWrapper: normalizes values for better performance
MinMaxScalar: transforms features according to min and max value
RobustScalar: Scales features by their quantile range
Principal component analysis: reduces the labels
TruncatedSVDWrapper: reduces complexity but without centering the data first
SparseNormalizer: similar
What is the Model Selection within AutoML?
service that helps you select the best performing statistical model
What are classification metrics?
For well-balanced datasets:
accuracy
average_precision_score_weighted
norm_macro_recall
precision_score_weighted
imbalanced:
AUC_weighted
What is a well-balanced/imbalanced dataset?
well-balanced: evenly distributed values
imbalanced: some values very often, others are scarce
What are primary metrics for Regression Scenarios?
large range:
spearman_correlation
r2_score
small range:
normalized_root_mean_squared_error
normalized_mean_absolute_error
What is Model Validation?
Compares how good the model “learned” by comparing the results from a training dataset with a test dataset
What is custom vision?
no-code service to build Classification and Object Detection ML Models
Custom Vision - What Classification Domains are recommended for which size of datasets?
bigger datasets:
General
General A1
smaller datasets:
General A2
Custom Vision - How many images do you need per label to train?
15
Custom Vision - What are evaluation metrics?
Precision: accuracy
Recall: How many relevant iteams returned?
Average Precision
For object detection: mean average precision
What should Kubernetes and what should Containers be used for?
Kubernetes: Deployment
Containers: Dev/Testing
What are labels and features?
Feature: Nebenwerte um auf den zu schätzenden Wert zu schließen
Label: der Wert der geschätzt wird
What is the difference between clustering and categorizing?
Cluster: ähnliche Werte zusammen in Kategorien bringen
Categorizing: Werte anhand von festen Labeln zuordnen
Was sind Metriken um Regression zu bewerten?
R2
RMSE
What Metric for classification models?
true positive rate
Kann Custom Vision service für Videos benutzt werden?
Nein
