Azure AI Engineer Flashcards

You are creating an AI system that will examine several millions of images. You must suggest a method for keeping the photos. The answer must be to reduce expenses. Which storage option would you recommend?

Answer: Azure Blob Storage

Azure Blob Storage is a highly scalable and cost-effective object storage solution specifically designed for unstructured data like images, videos, and documents. For storing millions of images, its tiered storage options (hot, cool, archive) allow for significant cost savings by moving less frequently accessed data to cheaper tiers. This makes it an ideal choice for large-scale image storage when cost reduction is a primary concern.

You launch an application that analyzes the sentiment of data kept in Azure Cosmos DB. You just loaded a lot of data into the database. The information belonged to a client called Geeks, Ltd. You find that Geeks data queries take a while to finish. The answer must be to reduce expenses. What is the best approach to achieving the objective?

Answer: Change the partioning strategy

In Azure Cosmos DB, an inefficient partitioning strategy can lead to 'hot partitions' where a disproportionate amount of data or queries are directed to a single partition, causing performance bottlenecks and increased costs. By optimizing the partitioning strategy, data can be distributed more evenly across partitions, improving query performance for specific clients and potentially reducing the need for higher throughput (RUs), thereby lowering expenses.

You are creating an AI-based picture categorization solution in Azure. It would help if you decided which processing platform would enable you to upgrade the logic in the future. Without batching, the answer must have the lowest latency for inferencing. Which compute target should you choose?

Answer: Field-programmable gate arrays (FPGAs)

FPGAs are reconfigurable hardware accelerators that offer extremely low latency for inferencing, especially for single-item (non-batched) predictions, due to their ability to execute custom logic directly in hardware. Their reconfigurability also allows for future upgrades to the AI model's logic without requiring a complete hardware redesign. This combination of low latency and flexibility makes FPGAs an excellent choice for real-time AI inferencing where speed and adaptability are critical.

You must create an interactive website to upload photographs and pose several predetermined questions in response to each image. Which services should you use?

Answer: Dynamically ask questions based on an uploaded image: Azure Bot Service Analyze and classify an image: Computer vision

To achieve an interactive website that processes uploaded images and responds with questions, Azure Computer Vision is essential for analyzing and classifying the image content. Azure Bot Service can then be used to build the conversational interface, dynamically asking predetermined questions based on the insights gained from the Computer Vision API. This combination provides both the image understanding and the interactive dialogue capabilities required.

You set up an extensive data workload infrastructure. It would help if you used Microsoft Machine Learning Server and Azure HDInsight. You want to launch rx function calls concurrently using RevoScaleR compute contexts. What are the three different computing contexts that the machine learning server supports?

Answer: 1. SQL 2. Spark 3. Local parallel

Microsoft Machine Learning Server (formerly R Server) with RevoScaleR provides various compute contexts for executing `rx` functions, enabling scalable and parallel processing. The three primary distributed and parallel compute contexts supported for `rx` function calls are SQL Server (for in-database analytics), Apache Spark (for big data processing on clusters like HDInsight), and local parallel (for leveraging multiple cores on a single machine). These contexts allow for efficient execution of machine learning tasks on different data platforms.

You intend to roll out an application that can recognize images. The application will use two Azure Blog storage stores called Blob1 and Blob2 to store picture data. You must suggest a security measure that complies with the enumerated conditions. - Access to Blob1 must be controlled by using a role - Access to Blob2 must be time-limited and constrained to specific operations

Answer: 1. Azure Active Directory (Azure AD) 2. Shared Access Signature (SAS)

Azure Active Directory (Azure AD) provides robust role-based access control (RBAC) for Azure resources, making it suitable for controlling access to Blob1 using roles. For Blob2, a Shared Access Signature (SAS) is the perfect solution as it allows granting time-limited permissions to specific storage resources and operations, without sharing account keys. This combination addresses both security requirements effectively.

You have a solution that utilizes an Azure Kubernetes Service cluster with five nodes. There is one N-series virtual computer in the group. Once every day and infrequently on demand, an Azure batch AI operation runs. When the cluster is not in use, you must suggest a method for keeping it configured. The solution may incur no computing expenses. What should the recommendation contain? Answer: Change the partitioning property

Answer: FALSE

The proposed answer, 'Change the partitioning property,' is irrelevant to reducing computing expenses for an Azure Kubernetes Service (AKS) cluster when it's not in use. Partitioning is a concept related to data storage (like Cosmos DB) or distributed processing, not directly to AKS cluster compute costs. To incur no computing expenses for an AKS cluster when idle, one would typically scale down the node pools to zero or deallocate the cluster, not change a partitioning property.

Is a machine learning technique where a neural network learns to predict outcomes or categorize data using labeled datasets. Answer: Supervised machine learning

Answer: True

The definition accurately describes supervised machine learning. In supervised learning, models are trained on labeled datasets, meaning each input data point is paired with its correct output or category. The goal is for the model, often a neural network, to learn the mapping from inputs to outputs so it can predict outcomes or classify new, unseen data.

A type of neural network architecture that is popular in LLM research that uses self-attention mechanisms to process input data.

Answer: What is Transformers

The description precisely defines the Transformer architecture, which revolutionized natural language processing and is the foundation for most modern Large Language Models (LLMs). Transformers are characterized by their innovative use of self-attention mechanisms, allowing them to weigh the importance of different parts of the input sequence and process data in parallel, leading to significant advancements in tasks like machine translation and text generation.

Is a branch of machine learning that employs neural networks to simulate and resolve challenging issues.

Answer: Deep Learning

Deep Learning is a specialized subfield of machine learning that specifically utilizes artificial neural networks with multiple layers (hence 'deep') to learn complex patterns from data. It is particularly effective at tackling challenging problems in areas like image recognition, natural language processing, and speech recognition, often outperforming traditional machine learning methods on large, complex datasets.

Are composed of layers of interconnected nodes that analyze and transform data, and are modeled after the composition and operation of the human brain.

Answer: neural networks

This definition perfectly describes neural networks, which are the core building blocks of deep learning. They are computational models inspired by the structure and function of the human brain, consisting of interconnected nodes (neurons) organized in layers. These networks process and transform data by passing signals through these connections, adjusting weights and biases to learn patterns and make predictions.

The process of adapting an LLM for a specific task or domain by training it on a smaller, relevant dataset.

Answer: What is fine tuning

Fine-tuning is a common and powerful technique in machine learning, especially with Large Language Models (LLMs). It involves taking a pre-trained model and further training it on a smaller, task-specific dataset to adapt its knowledge and capabilities to a particular domain or application. This process allows the model to specialize and perform better on specific tasks without needing to train a model from scratch.

The machine learning technique in which an agent learns to make decisions in an environment to maximize a reward signal. Application of this technique: robotics and game playing

Answer: Reinforcement machine learning

This definition accurately describes Reinforcement Learning (RL). In RL, an agent learns to interact with an environment by taking actions, receiving feedback in the form of rewards or penalties, and then adjusting its strategy to maximize cumulative rewards over time. This trial-and-error learning paradigm is particularly well-suited for applications like robotics, game playing, and autonomous systems, where agents need to make sequential decisions.

To find patterns in a data set, machine learning uses data and algorithms. It can handle smaller data sets and is better suited for simpler jobs. The model's training also doesn't take very long. Deep learning, on the other hand, makes use of artificial neural networks to identify patterns in a set of data. Deep learning, as opposed to machine learning, is better at complex tasks and utilizes larger data sets as opposed to smaller ones. The model's training takes a long period as well. Deep learning is intended for complicated tasks that must be solved well, whereas machine learning can be utilized for smaller, simpler tasks. This is the major distinction between machine learning and deep learning.

Answer: ML Vs. DL

The provided text explicitly outlines the key distinctions between Machine Learning (ML) and Deep Learning (DL). It discusses their respective strengths regarding data set size, suitability for task complexity, training duration, and the fundamental architectural difference of deep learning utilizing artificial neural networks. Therefore, the text serves as a direct comparison of ML versus DL.

A natural language processing program that, by utilizing straightforward pattern matching methods, could carry on a conversation with a person

Answer: ELIZA - AIL of MIT

ELIZA, developed at MIT in the 1960s, was one of the earliest natural language processing programs designed to simulate human conversation. It achieved this by using simple pattern matching and substitution rules to respond to user input, often mimicking a Rogerian psychotherapist. This groundbreaking program demonstrated the potential for human-computer interaction through natural language, fitting the description perfectly.

When supervised machine learning is used, the computer forecasts future data based on historical data. The computer learns from its error and performs better the next time based on how accurate the forecast was.

Answer: What is Supervised Machine Learning?

The description accurately defines supervised machine learning. It highlights the core principle of learning from historical, labeled data to make predictions about future data. The mention of the computer learning from its errors and improving based on prediction accuracy is characteristic of the iterative training process in supervised learning, where models are optimized to minimize discrepancies between predicted and actual outcomes.