Assam HS Skill Subject Artificial Intelligence flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Artificial Intelligence

The field of building computer systems that perform tasks associated with human intelligence, including perception, reasoning, learning, language understanding, and decision-making.

Turing Test

A 1950 proposal by Alan Turing in which a machine is judged intelligent if a human interrogator cannot reliably distinguish its text responses from a person's.

Rational Agent

An agent that perceives its environment through sensors and chooses actions expected to maximize a defined performance measure, given its knowledge and available actions.

AI versus Machine Learning

AI is the broad goal of intelligent behavior; machine learning is a subset of AI where programs learn patterns from data instead of following only hand-written rules.

Mutable default argument pitfall

Default parameter values are evaluated once when a function is defined, so a default list is shared across calls; use None as the default instead.

NumPy ndarray

A homogeneous multidimensional array object that supports vectorized element-wise arithmetic, shape inspection, reshaping, and broadcasting without explicit Python loops.

pandas DataFrame

A two-dimensional labeled table with named columns and row indices, supporting filtering, grouping, joining, and missing-value handling for tabular datasets.

loc versus iloc

In pandas, loc selects rows and columns by label, while iloc selects them by integer position; mixing them up causes wrong rows to be returned.

Structured data

Data organized in a fixed schema such as rows and columns in a relational table, making it easy to query and feed into most machine learning models.

Median versus mean

The mean is the arithmetic average and is pulled by extreme values; the median is the middle ordered value and stays robust when outliers are present.

Standard deviation

The square root of variance, expressing the typical distance of values from the mean in the same units as the original data.

Scatter plot use

A chart plotting one numeric variable against another to reveal relationship direction, strength, clusters, and outliers between two continuous features.

Supervised learning

A learning setting where the model is trained on labeled examples pairing input features with a known target, aiming to predict the target for new inputs.

Unsupervised learning

A learning setting using unlabeled data to discover structure, such as grouping similar records with clustering or reducing dimensions with PCA.

Precision and recall

Precision is TP divided by TP plus FP, measuring how many predicted positives are correct; recall is TP divided by TP plus FN, measuring how many real positives were found.

Overfitting

A model fits training data noise too closely, producing low training error but poor performance on unseen test data; remedies include more data, simpler models, and regularization.

Train-test split

Dividing a dataset into a training portion used to fit the model and a held-out portion used to estimate how well it generalizes to new data.

Perceptron

The earliest neural unit: it computes a weighted sum of inputs plus bias and applies a step function, and it can only separate linearly separable classes.

ReLU activation

A function returning max(0, z); it is cheap to compute and reduces the vanishing gradient problem, making it a common default in hidden layers.

Softmax function

A function converting a vector of raw scores into probabilities between zero and one that sum to one, used for multi-class classification output layers.

Backpropagation

An algorithm that applies the chain rule backward through the network to compute the gradient of the loss with respect to every weight for gradient descent updates.

Learning rate

A hyperparameter scaling each gradient descent step; too large causes divergence or oscillation, too small makes training very slow.

Tokenization

Splitting raw text into smaller units such as words or subwords, the first structural step before stop-word removal, vectorization, or model input.

TF-IDF

A text vectorization scheme weighting each term by its frequency in a document multiplied by inverse document frequency, so common terms across documents count less.

Chatbot intent classification

The step that maps a user message to a predefined intent label, usually with a trained text classifier, before the bot selects a response or action.

Image data augmentation

Creating modified copies of training images through flips, rotations, crops, or brightness changes to enlarge the dataset and reduce overfitting in vision models.

Algorithmic bias

Systematic unfairness in model outputs arising from biased training data, sampling or labeling errors, or proxy variables correlated with protected attributes.

Data minimization

A privacy principle limiting collection and retention to personal data actually needed for a stated purpose, reducing harm if a breach occurs.

Informed consent

Permission given by a person after being told what data will be collected, why, how it will be used, and how they can withdraw it.

Explainability

The degree to which humans can understand why a model produced a given output, supported by tools such as feature importance and model cards.

Problem framing

The lifecycle stage converting a real need into a precise ML task by defining the objective, target variable, success metric, and baseline.

Model drift

Degradation of deployed model performance as real-world data distributions change over time, detected by monitoring inputs and outcomes and addressed by retraining.