ARTIFICIAL INTELLIGENCE Course Notes (DLBDSEAIS01)

Practice flashcards covering key concepts from the AI course notes, including history, modern AI, reinforcement learning, NLP, and computer vision.

When and where was the term 'artificial intelligence' first coined?

In 1956 at the Dartmouth Summer Research Project on Artificial Intelligence (the Dartmouth Conference).

Who coined the term 'artificial intelligence'?

John McCarthy.

What is the purpose of the Turing Test?

To determine whether a machine's conversational ability is indistinguishable from a human.

What are the three core components of an expert system?

Knowledge base, inference engine, and user interface.

What is the period referred to as Nascent Artificial Intelligence and its focus?

1956–1974; the symbolic AI and knowledge representation era.

What caused the first AI winter (1974–1980)?

Over-optimistic expectations (e.g., machine translation) and insufficient data/computing power.

What caused the second AI winter (1987–1993)?

Collapse of the Lisp machine market and limits in expert systems’ scalability and reliability.

What is an AI winter?

Periods when interest, research activity, and funding in AI decreased.

Name three factors identified as critical for AI progress.

Algorithms/experience, computing capacity, and data availability.

What are two landmark early expert systems and their domains?

Dendral (organic chemistry) and MYCIN (medical diagnoses).

What are the three main types of expert-system knowledge representations?

Case-based, Rule-based, and Decision-tree representations.

Which institutions and researchers were pivotal in early AI history ?

Dartmouth College and MIT were key early centers; notable researchers include Turing, McCarthy, Minsky, and Chomsky.

What is the difference between symbolic AI and connectionist approaches?

Symbolic AI uses explicit rules and logic; connectionist approaches use neural networks and learning from data.

What event around 2012 signaled a resurgence of AI progress?

The rise of deep learning and connectionist approaches achieving state-of-the-art results.

What are ANI and AGI?

ANI: Artificial Narrow Intelligence (weak AI); AGI: Artificial General Intelligence (strong AI, domain-independent).

Name the major application industries highlighted for AI adoption in modern AI systems.

High Tech/Telecom, Automotive/Assembly, Financial Services, Business/Legal/Professional Services, Healthcare/Pharma, Consumer Goods/Retail.

What data splits are typically used to evaluate AI models?

Training set, Development/Validation set, and Test set.

What metrics are commonly used for binary classification?

Accuracy, Precision, Recall, and F-score.

What is a confusion matrix?

A table classifying results into True Positives, False Positives, True Negatives, and False Negatives.

What are the core components of reinforcement learning as defined in the notes?

Agent, Environment, States, Actions, Rewards, Policy, and Value.

What does the Markov property imply in MDPS?

Future states depend only on the current state and action, not on past states.

What is the Bellman equation used for in RL?

Computing the expected reward of taking an action in a state, including immediate and future rewards.

What is the Q-learning update formula?

Q(s,a) := r + γ max_a' Q(s',a').

What is Temporal Difference (TD) learning?

A model-free method that updates predictions based on the difference between successive predictions.

What is the role of exploration vs exploitation in Q-learning?

Exploration chooses random actions to learn about the environment; exploitation uses the Q-table to choose the best-known action.

What are the three NLP subdomains introduced in the notes?

Speech recognition, Natural Language Understanding (NLU), and Natural Language Generation (NLG).

What is TextRank and its purpose?

An unsupervised extractive text summarization method based on sentence similarity and graph ranking.

What is Word2Vec and its two training models?

Word2Vec is a neural-network-based word embedding method with CBOW (predicts a word from context) and Skip-gram (predicts context words from a word).

What are the common word-vector and document-vector techniques mentioned?

Word2Vec, TF-IDF, and GloVe.

What are USE and BERT in sentence representations?

USE (Universal Sentence Encoder) and BERT are pre-trained models for obtaining sentence embeddings; BERT uses masked language modeling and next-sentence prediction.

What is TF-IDF and what do TF and IDF stand for?

TF-IDF weights terms by their frequency in a document (TF) and inversely by how common the term is across documents (IDF).

What is semantic segmentation?

Pixel-level labeling where each pixel is assigned to a class (e.g., chair, table, background).

What role do CRFs play in segmentation?

Conditional Random Fields refine segmentation maps by modeling label dependencies.

What are common CV feature detectors and descriptors mentioned?

Edges, corners, and blobs; detectors like Canny edge detector and Harris corner detector; descriptors like BRIEF, ORB, SIFT, SURF.

What is the general CNN-based approach to semantic segmentation?

A CNN for feature extraction with a deconvolution/upsampling path to produce per-pixel class labels.

What is camera calibration and which parameters does it estimate?

Calibration estimates intrinsic (focal length, optical center, distortion) and extrinsic (rotation/translation) parameters.

What is the Brown-Conrady model used for?

Modeling and correcting lens distortion in camera calibration.

What is Zhang’s checkerboard calibration method?

A flexible camera calibration approach using checkerboard patterns across multiple images to estimate intrinsic/extrinsic parameters and distortion.

What is the difference between intrinsic and extrinsic camera parameters?

Intrinsic: internal camera properties like focal length and distortion; Extrinsic: camera's orientation and position in the world.

Why is feature matching important and what distances are used?

To find corresponding features across images; Euclidean distance for real-valued descriptors, Hamming distance for binary descriptors like BRIEF; FLANN for efficient matching.

What are some practical CV use cases for semantic segmentation?

Autonomous driving (vehicles, lanes, pedestrians), geosensing (land use), pose estimation, and medical imaging.

What is pivot machine translation used for?

Pivot machine translation is a technique that uses a third language as a bridge to translate text or speech between two languages that do not have a direct translation pair.

What is binary classification in machine learning?

Binary classification is a task in machine learning where the goal is to categorize input data into one of two distinct classes (e.g., spam or not-spam, true or false, disease or no-disease).

What is the purpose of the Harris Corner Detector?

Used in computer vision to identify corners and other salient features in an image, which are stable points useful for tasks like feature tracking, image registration, and object recognition.