AI-102 EXAM QUESTIONS 2024 MICROSOFT AZURE AI 102 CERTIFICATION COURSE PART-4

This video is the fourth part of a question series for AI 1002, thanking subscribers for their consistent engagement and valuable feedback, particularly Hasid Singh for suggestions on rearranging concepts and including drag-and-drop and fill-in-the-blanks question formats. The presenter assures that all core concepts will be covered, even if the question format differs. **Question 31: Multi-turn Conversation and Waterfall Dialogues** The first question addresses multi-turn conversations, where an AI and user engage in continuous dialogue, maintaining context. A practical example using ChatGPT illustrates this. The question asks how to initiate a waterfall dialogue, which involves processing results from previous steps. The options are "begin," "replace," "cancel," and "prompt." * **Begin:** This option initiates a new conversation flow from the start, creating a new instance. This is a correct method for starting a dialogue. * **Replace:** This is used to switch between or replace current dialogues, which is not the requirement here. * **Cancel:** This is not relevant as the goal is to start a new dialogue, not end the current one. * **Prompt:** This is crucial for asking the user for information, enabling the multi-turn conversation. Initiating a dialogue alone isn't enough; you need to prompt the user for input. Therefore, "begin" and "prompt" are the correct choices for initiating a waterfall dialogue that carries context and prompts for user input. **Question 32: Azure AI Services for Content Moderation** The second question concerns implementing advanced algorithms to flag potentially offensive, dangerous, or unwanted content using Azure AI services. The options are Azure AI Video Indexer, Azure AI Vision, Azure AI Content Safety, and Azure AI Language. * **Azure AI Video Indexer:** This service is for analyzing audio and video files, not directly for flagging text or image content. * **Azure AI Vision:** While it can process images and text, it's not specialized for content moderation compared to other options. * **Azure AI Content Safety:** This is the most accurate and specialized service for detecting and flagging offensive, dangerous, and unwanted content, identifying potential risks and threats. * **Azure AI Language:** This service focuses on Natural Language Processing (NLP) for understanding text but isn't designed for flagging offensive content. Thus, **Azure AI Content Safety** is the correct answer. **Question 33: Text-to-Speech Pronunciation and Volume Control** This question asks about the technology to control pronunciation and volume in text-to-speech (TTS) conversations. * **Azure Speech Recognition:** This service recognizes speech, it does not control TTS output. * **HTML:** While HTML can embed audio, it lacks the functionality to control specific speech characteristics like pronunciation and volume. * **SSML (Speech Synthesis Markup Language):** This is a markup language specifically designed for speech synthesis. It allows fine-grained control over pronunciation, volume, pitch, speed, and emphasis, making the generated speech more natural and precise. * **XML:** This is a general-purpose markup language for structuring data and does not have built-in features for speech properties. Therefore, **SSML** is the correct technology. **Question 34: Azure Speech Service Configuration for Different Use Cases** This question presents a scenario where a user needs to configure the Azure Speech service for three different use cases: outputting audio to a wave file, outputting to a speaker, and outputting as an in-memory stream. The task is to map these use cases to four variables: `synthesizer`, `speech_config`, `audio_config`, and `null`. The four variables provided are: 1. `synthesizer`, `speech_config`, `null`, `audio_config` 2. `synthesizer`, `speech_config`, `audio_config`, `speech_config` 3. `synthesizer`, `speech_config`, `audio_config`, `audio_config` 4. `synthesizer`, `speech_config`, `speech_config`, `null` Let's analyze the use cases and typical configurations: * **Use Case 1: Output audio to a wave file:** To save audio to a file, both `speech_config` (for the subscription and voice settings) and `audio_config` (to specify the output file path and format, like WAV) are required. The presenter identifies variable 3 as a potential match, suggesting `synthesizer`, `speech_config`, `audio_config`, `audio_config`. However, the provided answer suggests a specific mapping to `variable 3` which seems to imply a structure like `synthesizer`, `speech_config`, `audio_config`, `audio_config`. Looking at the provided options for the variables, option 3 seems to be the closest if we interpret the structure as `synthesizer`, `speech_config`, `audio_config`, `audio_config`. The presenter later clarifies that for saving to a file, both `speech_config` and `audio_config` are needed. * **Use Case 2: Output to a speaker:** When outputting directly to a speaker, the audio is played in real-time. You need `speech_config` but not a specific `audio_config` for file output. The presenter identifies variable 4 as a potential match: `synthesizer`, `speech_config`, `speech_config`, `null`. This implies that only `speech_config` is passed, and the audio output is handled by the system's default audio device. * **Use Case 3: Output as an in-memory stream:** For an in-memory stream, you need `speech_config` and an `audio_config` that handles streaming. Passing `null` for the audio configuration in this context often signifies that the audio processing will be managed as a stream directly. The presenter identifies variable 1 as the correct match: `synthesizer`, `speech_config`, `null`, `audio_config`. This suggests that while `audio_config` is present, the `null` might indicate a specific streaming handler or that the `audio_config` itself is set up for streaming. The presenter confirms that the correct mapping is: * Use Case 1 (Wave file): Variable 3 (assuming it maps to `synthesizer`, `speech_config`, `audio_config`, `audio_config`) * Use Case 2 (Speaker): Variable 4 (assuming it maps to `synthesizer`, `speech_config`, `speech_config`, `null`) * Use Case 3 (In-memory stream): Variable 1 (assuming it maps to `synthesizer`, `speech_config`, `null`, `audio_config`) **Question 35: Bot Framework Dialogues for User Input and Results** This question, similar to an earlier one, asks which dialogue type to use in an Azure Bot application to ask for user input and return a result, enabling seamless context switching. * **Input Dialogue:** Captures user input but doesn't handle asking for it or returning a result. * **QnA Maker Dialogue:** Integrates a QnA service based on predefined knowledge bases, not for general user input and result processing. * **Action Dialogue:** Triggers specific behaviors based on user interaction but doesn't directly handle asking for input and returning results. * **Prompt Dialogue:** This dialogue type is designed to initiate asking for user input, capture that input, and then process it to return a result. It integrates the input dialogue's function and handles the overall flow of getting input and producing an output. Therefore, the **Prompt Dialogue** is the correct choice. **Question 36: Creating a Resource for Sentiment Analysis and OCR with a Single Key Endpoint** The user needs to create an Azure resource for sentiment analysis and OCR, requiring a single key endpoint. This implies using a multi-service Azure Cognitive Services resource. The question involves completing an HTTP request to create this resource. * **Line 1 (Management URL):** The request needs to be directed to the correct Azure management endpoint. For creating resources, the path typically starts with the subscription. The options are `subscriptions/{subscriptionId}` or `tenants/{tenantId}`. Since the goal is to create a resource within a subscription, `subscriptions/{subscriptionId}` is appropriate. * **Line 2 (Resource Provider):** The resource to be created performs sentiment analysis and OCR, which fall under Azure Cognitive Services. The correct resource provider for this is `Microsoft.CognitiveServices`. Thus, the correct completion is `subscriptions/{subscriptionId}` for the first part and `Microsoft.CognitiveServices` for the second part. **Question 37: Monitoring System for Engine Sensor Data and Alerts** The system needs to monitor engine sensor data (rotation speed, angle, temperature, pressure) and generate alerts for atypical values. This points towards anomaly detection. * **Application Insights in Azure Monitor:** Primarily for application performance monitoring and diagnostics, not direct sensor data anomaly detection. * **Metric Alerts in Azure Monitor:** Can trigger alerts based on predefined metric thresholds, but may not capture complex atypical patterns. * **Multivariate Anomaly Detection:** This is ideal for analyzing complex relationships between multiple sensor readings (like speed, angle, temperature, pressure) to identify unusual patterns that might not be apparent from individual metrics. * **Univariate Anomaly Detection:** Detects anomalies based on a single data stream, which is less suitable for complex, multi-sensor systems. Given the multiple sensor inputs, **Multivariate Anomaly Detection** is the most appropriate solution. **Question 38: Securing Azure Language Service Access** The requirement is to ensure that only resources within a specific virtual network (`VNet1`) can access a Language Service resource (`T1`). * **Language Service Container:** Not relevant for network access control. * **Virtual Network Setting for T1:** Configuring virtual network settings for the Language Service resource (`T1`) is the correct approach. This allows you to specify which virtual networks or subnets are permitted to access the service, effectively creating a private endpoint or service endpoint. Therefore, configuring the **virtual network setting for T1** is the correct action. **Question 39: Responsible AI Principles and Sentiment Analysis for Bonuses** This question involves using sentiment analysis results from surveys to calculate customer service staff bonuses, while adhering to Microsoft's Responsible AI principles. The principles are Fairness, Reliability & Safety, Privacy & Security, Inclusiveness, Transparency, and Accountability. * **Add a human review and approval step before making decisions affecting staff financial situation:** This directly aligns with the **Accountability** principle, ensuring a human is responsible for final decisions impacting staff. * **Include sentiment analysis results only when surveys return a low confidence score:** This is unfair as it ignores high-confidence positive feedback and is not consistent with **Fairness**. * **Use all surveys, including those from customers who requested data deletion:** This violates **Privacy & Security** principles by not respecting data deletion requests. * **Publish raw survey data to a central location and provide staff access:** This compromises **Privacy & Security** by exposing sensitive customer data. Therefore, adding a **human review and approval step** is the correct action to ensure compliance with Responsible AI principles. **Question 40: Exporting Custom Vision Model to a Disconnected Network** A custom vision project performing object detection in the general domain needs to be exported for use on a network disconnected from the internet. * **Change Classification Type:** Not the primary issue; the domain is key. * **Change Domain to Compact:** For offline use, the model must be exported to a compact domain, which is optimized for edge devices and offline scenarios. This is the crucial first step. * **Create a new classification model and retrain:** This is not necessary if the existing model can be adapted. * **Retrain the model and then export the model:** After changing the domain to compact, the model needs to be retrained within that domain to optimize it for offline performance. Following retraining, the model can be exported. The correct sequence of actions is: 1. Change the domain to **Compact**. 2. **Retrain the model** using the compact domain. 3. **Export the model** for offline use.