Embarking on your journey to conquer the AI-900: Microsoft Azure AI Fundamentals certification? This comprehensive study guide is designed to equip both technical and non-technical developers with essential knowledge about Azure AI services, including machine learning, computer vision, and natural language processing. As AI technologies, including generative AI and conversational AI, gain momentum, mastering these concepts is crucial for success in today’s job market. In this ultimate guide, we’ll delve into the key topics covered in the AI-900 exam, ensuring you’re well-prepared to navigate through AI workloads, principles, and the Azure Open AI Service. Whether you’re aiming to enhance your understanding of Azure services or boost your career with valuable AI certification, this post has everything you need for exam success!
The AI-900: Microsoft Azure AI Fundamentals certification is a standard Microsoft certification for technical and non-technical developers. It covers a wide range of topics that fall under the umbrella of Azure services relating to machine learning and other AI concepts. The certification has been gaining popularity due to the increasing conversations around AI, generative AI, conversational AI, and so on.
You can find the Microsoft Learn webpage for the certification here.
This post is going to cover the three biggest topics that are on the exam. I am confident in saying that if you understand the topics covered here then you will be more than successful on exam day. The three main topics we will be covering are the AI Workloads, the AI Principles, and Azure Open AI Service.
AI Workloads
The AI-900 certification exam evaluates the learner’s ability to describe AI workloads and considerations, and recognize elements of popular AI workloads such as computer vision, prediction/forecasting, conversational AI, and more. The exam is designed to test the learner’s knowledge of fundamental principles of machine learning on Azure. Also, some of the topics related to Microsoft Azure services. Fortunately, we are going to cover each one here.
There are Four Workloads but each workload has several “sub-workloads” underneath it’s umbrella. The questions ask you which workload, or sub-workload, would be needed to accomplish a specific ask.
The Four Workloads are: Machine Learning, Computer Vision, Natural Language Processing, and Knowledge Mining.
Machine Learning
The first workload we’ll cover is machine learning. Machine learning is a type of computer science that helps computers learn and make decisions on their own. It’s like teaching a computer to recognize patterns and make predictions based on what it has learned from data. Machine learning is a process of creating predictions by analyzing previous data, whether it’s numeric or not. It aims to make accurate predictions about future potential outputs.
Azure Machine Learning Services is a cloud-based platform that enables you to create, design, test, and publish your AI models. This is all done in the Machine Learning Designer, an interactive space where you can design, create, and test these models.
Machine Learning breaks up into regression, classification, clustering and anomaly detection.
Regression
Regression is a statistical technique that analyzes past numerical data to predict a numerical future. The stock market is a great example of this. The model looks at trends to determine the most likely prediction for the future. Once you build and test your model, you will receive scoring metrics that indicate how well it performed. These metrics include the Root Mean Squared Error (RMSE), Relative Squared Error, and Coefficient of Determination (R-squared). A helpful study hack is that all of these abbreviations start with an “R”. The Coefficient of Determination is “R^2” and Root Mean Square Error is “RMSE”. If they start with an R, they relate to regression!
Classification
Classification is a machine learning technique that involves assigning each data point to a specific class. If you see anything related to a Boolean, an option set, or a pick list, the answer is probably classification. For example, imagine you are a teacher and you have different data about each of your students. A classification model can predict whether each student passes or fails. This model can output the actual letter grade if it is designed to classify that way. It will give each student a classification label from that option set.
After testing your classification model, it will give you a confusion matrix, which summarizes the performance of the model on a set of test data.
It adds each data point to one of four boxes. The example used in the learning module involves a hospital with patients, where the classification model is trying to classify whether the patients were sick. The top left square, true positive, shows the count of records where the model predicted that the patient was sick and they were sick. The bottom right square is the true negative, which shows the count of records where the model predicted healthy and the patients were healthy.
The bottom left square is the false negative, where the model predicted that the patient was not sick, but they actually were. The false positives are in the top right, where the model predicted or classified that the patient is sick, but in fact, they were not. This is a false positive
Clustering
Clustering is a machine learning technique that involves breaking your data into segments. A good example of when this might be used in the real world is to segment your customers for marketing analysis or to separate your customers based on geographical locations so that you can split them into regions. This would be effective using clustering .
The difference between clustering and classification is that clustering does not necessarily apply a label to it. It does not say, “Hey, these are healthy patients, these are our sick patients” or “Here are our passing students and our failing students”. Instead, it simply groups the data points that have similar inputs into groups.
Anomaly Detection
Anomaly detection is a machine learning technique that identifies data points, events, or observations that do not conform to the expected pattern of a given group. It is like a fraud detection system that alerts you when something unusual happens. For example, when you are out of the country and your credit card gets swiped, you get a text from your bank. This is an example of anomaly detection. The model looks at your previous data and checks if something fits outside the standard deviation of your typical information. If it does, it alerts the anomaly detected model.
Computer Vision
Computer vision is a field of artificial intelligence that enables computers to interpret the world and make predictions based on inputs from cameras, videos, or photos. It has four main workloads: computer vision, custom vision, form recognizer, and document analysis.
Computer Vision
Some of its capabilities include object detection, which can determine if an object is in an image, tell you the location of the object in the image, the number of objects in an image, the color, or the resolution of the image. It can even read numbers from photos or videos. For example, say you need to track runners by their numbers in a marathon. You could use computer vision to read those numbers and keep their times. Computer vision can analyze images and videos and has preset capabilities such as automatically identifying brand logos in a photo without additional design or testing.
Custom Vision
Custom Vision is a cloud-based service that enables you to build, deploy, and improve your own image identifier models. It has many of the same capabilities as computer vision. However, you will need to use your own images to test and validate the model.
Custom Vision is a great tool for businesses that want to create custom image classifiers using their own data. It allows you to specify your own labels and train custom models to detect them. Once you’ve trained your model, you can test, retrain, and eventually use it in your image recognition app to classify images or detect objects.
Face
Azure AI Face is a cloud-based service that enables you to build facial detection and recognition solutions. Facial detection detects if there is a face in an image, while facial recognition determines whose face it is. The more photos you provide, the more accurate the model’s predictions will be. This model can even detect faces and recognize faces with things like hats and sunglasses on.
Facial recognition has many applications, such as unlocking your phone with your face, identifying people in security footage, or verifying someone’s identity for account recovery. However, it also raises concerns about privacy and surveillance. Therefore, it’s important to use facial recognition responsibly and ethically by following Microsoft’s AI Principles up next.
Document Intelligence
Document Intelligence is a machine learning technique that involves scanning documents or receipts. Then automatically extracting that information to put it where it needs to be . A common example of this is expense reports and receipts . I feel like I got several questions in my practice problems, or on my exam that were related to Document intelligence. If you need to rescan a receipt, you can use the Document Intelligence workload .
Natural Language Processing
Natural Language Processing (NLP) is a field of artificial intelligence that focuses on interpreting either written or spoken language and providing a response . There are four main models under NLP: language, translator, speech, and bot service .
Language
Azure AI Language provides you with the ability to understand and analyze text to build intelligent applications. A good example of Azure AI Language service is sentiment analysis. Sentiment analysis determines how happy or unhappy the writer of a text is based on natural language. A good use case for this is looking at customer reviews, where you can use the language service to determine the sentiment analysis of all those reviews.
Translator
Translator is going to be used to determine the language that something is written in. You can also translate the language from something to another.
Speech
Speech is a machine learning technique that has two capabilities: speech analysis and speech synthesis. This involves converting speech to text, while speech synthesis involves converting text to speech. Speech analysis can be used for things like automatic closed captions, where it determines the language and creates the captions. For speech synthesis, a great example would be where automated machines are reading a script aloud.
If something is spoken language that is different from what actually translates, it’s not going to be translator; it’s going to be speech because it was spoken.
Bot Service
Azure AI Bot Service is a cloud-based service that provides a bot framework to create and manage bots. It integrates back-end services like language by connecting to channels for web chat, emails, teams, and others. While it may sound similar to Power Virtual Agents, there is a difference. Bot Services is more technical as opposed to Power Virtual Agent, which is practice around being low-code.
Knowledge Mining
Azure AI Knowledge Mining is an emerging discipline in artificial intelligence (AI) that uses a combination of intelligent services to quickly learn from vast amounts of information. It allows organizations to deeply understand and easily explore information, uncover hidden insights, and find relationships and patterns at scale. This can be used to extract information from large volumes of often unstructured data to create a searchable knowledge store.
Microsoft AI Principles
Microsoft has six overarching principles that outline their approach to the pursuit of new AI technology and ensure that the pursuit of them remains responsible. They believe that when you create technology, you must ensure that the technology is developed and used responsibly. These six AI principles are founded on two perspectives: ethical and explainable. By adhering to these principles, Microsoft commits to building AI systems that are trustworthy, transparent, and accountable.
You can find more information on all of the principles outlined below in the Microsoft Documentation here.
Fairness
The fairness principle is founded on the belief that an AI system should treat all people fairly. This involves ensuring that the AI system allocates opportunities, resources, or information in ways that are just. If you get a question that suggests the output of an AI system is biased, that is going to be in violation of the fairness principle. Fairness safeguards against systems that are created using already biased data and their outputs then being biased as well.
Reliability and Safety
Reliability and safety is a principle that ensures AI systems always perform as they are expected and safely for any of the users that are involved. An example of how to practice the reliability and safety principle would be implementing safeguards into your AI system. Safeguards that would not provide an output if fundamental input data was not provided. For instance, if you are using an AI system to predict the diagnosis of different patients, but if that patient’s temperature was never taken, you might want to create a safeguard like not to diagnose a patient if there’s no temperature taken. Because this may be something that is fundamental to the diagnosis, it’s exercising reliability and safety to have the system not provide an output.
Privacy and Security
Privacy and Security has everything to do with protecting the data that is either built or is currently being used by the AI system.
Inclusiveness
Inclusiveness is a principle that exercises AI systems should empower and engage all people, even with different abilities . Not to be confused with fairness, inclusiveness has to do with the users of the system. The fairness principle has to do with more of the output . An example of inclusiveness could be if a user is blind, the output would then be reading aloud using speech synthesis as opposed to just transcribing the text on the screen . By adhering to this principle, AI systems can be designed to be more accessible and inclusive for all users.
Transparency
Transparency is a principle that states AI systems should be understandable . If you want an AI system to produce a document that states why it came to its conclusion, this exercise is related to the transparency principle . This involves considering how people may misunderstand, misuse, or incorrectly estimate the capabilities of the system .
Accountability
The final AI principle is accountability, which pertains to the creators of a system and all people who are generally responsible for the use and practice of AI systems. It also involves making sure that people continuously oversee the systems to ensure that they are still reliable, safe, private, secure, and fair.
Azure Open AI Service
Azure OpenAI Services is a cloud-based service that provides advanced language models such as GPT-4, GPT-3, Codex, DALL-E, and Whisper models with the security and enterprise promise of Azure. It is a partnership between Microsoft and OpenAI, the founders of Chat-GPT. Azure Open AI Services currently has four main capabilities.
- Access to Powerful Language Models
- Fine Tuning – allows you to fine-tune modules with your data and hyperparameters to increase the accuracy of output
- Responsible AI – built-in features to help you ensure you’re using AI responsibly
- Enterprise grade Security – runs on the Azure global infrastructure to meet your production needs such as critical enterprise security compliance, and regional availability
The last thing to talk about related is the Azure OpenAI Studio. This is a web-based interface where you can build, test, and deploy your models for public consumption. Within the studio, you will have what’s called playgrounds, which are interfaces that are created with a low or no-code approach to test your models. You can play around and quickly iterate and experiment with different capabilities. There’s even an assistant drop-down there to help you get started and answer any questions that you have while building.
In Conclusion
The AI-900: Microsoft Azure AI Fundamentals certification is a standard Microsoft certification that covers a wide range of topics related to Azure services and machine learning. The exam is designed for both technical and non-technical developers and requires preparation. Although the exam is not super challenging, it will ask you about general principles and which Azure service to use.
Considering you have worked through this post, you are already very prepared for the exam. To continue to prepare, I always recommend using tools such as the Microsoft learning modules, which provide guided articles created by Microsoft that cover everything the exam is going to cover. Secondly, practice problems, which are an invaluable resource for studying. Check out this post here to see my fail-proof Microsoft exam strategy!
Happy Testing!