Mixed Deck — All AI Topics Flashcards

In natural language processing, what does 'tokenization' refer to?

Answer: Splitting raw text into smaller units such as words or subwords

Tokenization breaks text into tokens (words, subwords, or characters) that serve as the discrete input units for NLP models.

What is 'shadow mode' deployment in MLOps?

Answer: Sending production traffic to a new model without using its outputs to serve users

Shadow mode runs a new model on real traffic in parallel with the live model, comparing outputs without impacting users.

The system could analyze the potential outcomes of each move and forecast future plays by utilizing numerous deep neural networks, reinforcement learning, playing millions of games between its neural networks, and learning human techniques.

Answer: AlphaGo versus Lee Sedol

The description accurately details the capabilities and achievements of AlphaGo versus Lee Sedol. AlphaGo, developed by DeepMind, famously defeated the world champion Go player Lee Sedol using advanced techniques like deep neural networks and reinforcement learning, playing millions of games against itself to master the game.

Which of the following uses artificial intelligence?

Answer: Language understanding and problem-solving (Text analytics and NLP)

Artificial intelligence is extensively applied in areas requiring the understanding and processing of human language, such as text analytics and Natural Language Processing (NLP). These AI capabilities enable machines to interpret, analyze, and generate human language, facilitating applications like chatbots, sentiment analysis, machine translation, and information extraction.

Which Azure AI service provides pre-built models for extracting key-value pairs and tables from forms and documents?

Answer: Azure Document Intelligence

Azure Document Intelligence (formerly Form Recognizer) extracts key-value pairs, tables, and structured data from documents using pre-built and custom models.

Which SPARQL query keyword is used to retrieve triples from an RDF knowledge graph?

Answer: SELECT with WHERE patterns

SPARQL uses SELECT queries with WHERE clauses containing triple patterns (subject, predicate, object) to match and retrieve data from RDF graphs.

Which technique is used to visualize which parts of an input image most influence a CNN's classification decision?

Answer: Grad-CAM (Gradient-weighted Class Activation Mapping)

Grad-CAM uses gradients flowing into the final convolutional layer to produce a heatmap highlighting the regions most important to the prediction.

In the context of decision trees, what is information gain?

Answer: The reduction in entropy (uncertainty) in the target variable achieved by splitting on a given feature

Information gain measures how much a feature split reduces entropy (disorder) in the target variable — features with the highest information gain are chosen as split points to build a tree that separates classes most effectively.

What is the role of an 'enrichment pipeline' in Azure Cognitive Search?

Answer: It applies AI skills during indexing to extract and transform content

An enrichment pipeline (skillset) applies AI skills—OCR, entity extraction, translation—to raw content during indexing to augment the searchable index.

What does the 'grounding' technique in prompt engineering accomplish when working with Azure OpenAI?

Answer: Reduces the model's hallucination by injecting relevant source documents into the prompt context

Grounding provides the model with retrieved context from authoritative sources, anchoring its responses to verified information and reducing hallucination.

Which trait is frequently connected to artificial intelligence? (a) Consciousness and emotion

Answer: Limited scope and application in specific domains

Current artificial intelligence systems are primarily examples of 'narrow AI,' meaning they are designed to perform specific tasks within limited domains. Unlike human general intelligence, they lack consciousness, emotion, or the ability to apply knowledge broadly across different contexts. Their intelligence is specialized and confined to their training data and programmed objectives.

What is the purpose of 'model quantization' in the context of ML deployment?

Answer: To reduce model size and inference time by using lower-precision arithmetic

Quantization converts model weights from 32-bit floats to 8-bit integers, shrinking memory footprint and speeding up inference with minimal accuracy loss.

A search algorithm takes _____ as an input and returns _____ as an output.

Answer: Problem, Solution--

A search algorithm in AI takes a defined problem as its input, which includes an initial state, a set of possible actions, and a goal state. Its purpose is to explore the problem's state space and find a sequence of actions or a path that leads to a solution, which is then returned as the output. This process is fundamental to many AI applications.

You need to deploy a machine learning model so it can be consumed via a REST API with auto-scaling. Which Azure service is most appropriate?

Answer: Azure Machine Learning managed online endpoint

Azure Machine Learning managed online endpoints provide scalable, real-time REST API inference for deployed models.

Which of the following best describes an embedding in AI?

Answer: A dense vector representation of data in a continuous space

Embeddings map discrete objects (words, items) to dense vectors so that similar objects are close together in vector space.

Using Azure Form Recognizer, you may use machine learning technologies to create automated data processing software. Use Form Recognizer's Prebuilt Receipt model to read sales receipts in English. Please review the options below, then pick three from which Form Recognizer's Prebuilt Receipt model can extract data.

Answer: Quantity

Form Recognizer's Prebuilt Receipt model is designed to extract key data points from sales receipts. Among the options, it can reliably extract the Quantity of items purchased, the overall Receipt Type (e.g., sales receipt), and the Merchant Address. These fields are standard components of most sales receipts and are crucial for automated data processing.

What is 'multi-hop reasoning' in knowledge graphs and QA systems?

Answer: Answering a question by chaining multiple inference steps across graph edges

Multi-hop reasoning requires traversing multiple relationships in a knowledge graph or making several logical inferences to arrive at an answer that cannot be found in a single step.

You must recognize and remove any video clip deemed to contain adult material. You can use Azure's Video Indexer or Video Analyzer for Media. Which Element would you pick to view the required insights in the video and the time frames where these insights first appear?

Answer: Visual Content Moderation

To recognize and remove video clips containing adult material and view insights with time frames, Visual Content Moderation is the correct element. This feature within Azure Video Indexer specifically analyzes video content for adult or racy material. It provides detailed insights, including the exact time segments where such content is detected, enabling precise moderation.

Which knowledge representation technique uses probabilistic relationships between variables to model uncertainty?

Answer: Bayesian network

Bayesian networks use directed acyclic graphs to represent probabilistic dependencies between variables, enabling inference under uncertainty.

Your team wants to monitor model drift for a deployed Azure Machine Learning model over time. Which feature should you use?

Answer: Data drift detection in Azure Machine Learning

Azure Machine Learning's data drift detection compares incoming scoring data against the training dataset to identify distribution shifts.