Data Science & AI Flashcards

Standard Deviation/What does larger values mean?

A measure of data spread; large values = high variability.

Variance of 0

All values are identical.

Gaussian Distribution

Bell curve.

% within 1 SD in normal distribution

68%.

Continuous Variable

Height, temperature.

Discrete Variable

Number of students.

Expected Value of Die Roll

3.5

Importance of Gaussian Distribution

Used in modeling errors and natural variation.

68-95-99.7 Rule

Percent within 1, 2, 3 SDs.

Outliers Affect Mean

Skew results upward/downward.

Parameter vs Statistic

Parameter = population, Statistic = sample.

Positive Covariance

Variables increase together.

Best Visualization for Distribution

Histogram or boxplot.

Best Chart for Categories

Bar chart.

Boxplot vs Histogram

Boxplot = summary stats; histogram = frequency distribution.

Multivariate Data Example

Data with multiple features, e.g., height and weight.

Multiple Linear Regression Use

Predict continuous outcomes.

Why Clean Data

Removes errors, improves accuracy.

Duplicate Record Issue

Distorts analysis results.

Low-Quality Data Issue

Produces unreliable AI results.

Imputation Meaning

Filling in missing data.

Classification Algorithms

Decision trees, logistic regression, SVM.

Decision Trees Use

Model decision paths.

Feature Selection Role

Removes irrelevant data to improve performance.

Correlation vs Causation

Correlation = relationship, causation = direct effect.

Data Normalization Importance

Scales data for fair model comparison.

Real-World Data Science Use

Predict sales, detect fraud, diagnose diseases.

Purpose of SQL

Query and manage databases.

SQL Query for All Employees

SELECT * FROM employees;

Library for Data Manipulation

Pandas.

NumPy Used For

Numerical computation.

Pandas Used For

DataFrames, data manipulation.

PyTorch Used For

Deep learning models.

Data Wrangling

Transforming and cleaning raw data.

R vs Python

R = stats focus; Python = general-purpose.

Relational Database

Organizes data in linked tables.

Primary key

Unique identifier per record.

Generative AI

AI that creates new data (text, images, etc.).

Generative AI examples

ChatGPT, DALL·E.

Limitation of generative AI

Can generate incorrect or biased content.

Ethical concern

Deepfakes, misinformation.

Computer vision

AI for image/video understanding.

NLP used for

Language processing tasks.

LLM

Large Language Model.

Hallucination in LLMs

AI generating false or made-up info.

Bias in LLMs

Reflects biased training data.

Symbolic vs neural AI

Symbolic = rules, Neural = learned patterns.

Machine learning

AI that learns from data to make predictions.

Training vs testing vs validation

Train = learn, Test = evaluate, Validation = tune.

Supervised learning feature

Labeled data.

Unsupervised learning example

K-means clustering.

Reinforcement learning

Learning via rewards/punishments.

Real-world reinforcement example

Game-playing AI, robotics.

Neural network

Layers of interconnected nodes mimicking the brain.

Deep learning

Neural networks with many layers.

Overfitting

Model memorizes data; prevented with regularization.

Learning rate

Controls speed of model updates.

Speech recognition

Technology that converts spoken language into text.

Knowledge representation

A method for encoding information for AI reasoning.

Reasoning in AI

Using logic or inference to draw conclusions from data.

Example of reasoning in AI

Diagnosing diseases based on symptoms.

Data privacy

Protecting personal data from unauthorized access.

AI transparency importance

Ensures trust and accountability in AI systems.

Algorithmic bias

Systematic unfairness due to biased data or design.

AI ethics

Guidelines for responsible development and use of AI.

Example of AI ethical concern

Facial recognition used without consent.

Data literacy

Ability to read, work with, analyze, and argue with data.

Raw data

Unprocessed information collected from sources.

Dataset

A structured collection of related data.

Data visualization importance

Helps interpret patterns and insights easily.

Metadata

Data describing other data, such as format or source.