5) AI and ML

Vocabulary flashcards generated from the AI and Machine Learning lecture notes covering concepts, paradigms, algorithms, metrics, and project lifecycle terms.

Artificial Intelligence (AI)

A broad field focused on developing intelligent systems capable of performing tasks that typically require human intelligence, such as perception, reasoning, learning, problem solving, and decision-making.

Machine Learning (ML)

A type of AI focused on building methods that allow machines to learn from data to improve performance on tasks and make predictions without being explicitly programmed with rules.

MYCIN Expert System

An early AI medical diagnosis system developed in the 1970s that used a collection of over 500 rules to rank culprit bacteria and recommend drug treatments based on reported symptoms and test results.

Deep Learning (DL)

A subset of ML that utilizes multiple layers of artificial neurons and synapses to process complex patterns in large amounts of input data, typically requiring GPUs.

Generative AI (Gen-AI)

A subset of Deep Learning built on multi-purpose foundation models backed by neural networks that can perform a broad range of general tasks and create new content.

Transformer Model

An LLM architecture capable of processing entire sentences at once rather than word-by-word, assigning relative importance to specific words for faster, more coherent text processing.

GPT (Generative Pre-trained Transformer)

A transformer-based language model designed to generate human-like text or computer code based on input prompts.

BERT (Bidirectional Encoder Representations from Transformers)

A transformer-based model that processes text bidirectionally to understand language context and relationships.

RNN (Recurrent Neural Network)

A neural network architecture designed for sequential data such as time-series or text, commonly used in speech recognition.

ResNet (Residual Network)

A Deep Convolutional Neural Network (CNN) architecture primarily used for image recognition tasks, object detection, and facial recognition.

SVM (Support Vector Machine)

A machine learning algorithm used for both classification and regression tasks.

WaveNet

A deep generative model designed to synthesize raw audio waveforms, primarily used in speech synthesis.

GAN (Generative Adversarial Network)

A class of machine learning frameworks used to generate synthetic data such as images, videos, or audio resembling training data, useful for data augmentation.

XGBoost (Extreme Gradient Boosting)

An optimized, efficient implementation of the gradient boosting algorithm used for supervised learning tasks.

Labeled Data

Data that contains both input features and corresponding target output labels, used primarily in Supervised Learning.

Unlabeled Data

Data that consists only of input features without any associated target output labels, used in Unsupervised Learning.

Structured Data

Data organized in a formatted layout such as tabular rows and columns or time series datasets.

Unstructured Data

Data that does not conform to a predefined data model or tabular format, often consisting of text-heavy content, images, or audio.

Supervised Learning

An ML approach where an algorithm learns a mapping function from labeled training data to accurately predict outputs for unseen inputs.

Regression

A supervised learning task used to predict continuous numeric values within a range based on input features.

Classification

A supervised learning task used to predict discrete categorical labels or classes for input data.

Training Set

The portion of a dataset (typically 60–80%) used directly to train an ML model.

Validation Set

The portion of a dataset (typically 10–20%) used to tune model parameters, evaluate hyperparameter settings, and validate performance during development.

Test Set

An independent portion of a dataset (typically 10–20%) used exclusively to evaluate the final model's generalization accuracy.

Feature Engineering

The process of selecting, creating, transforming, and extracting meaningful variables from raw data using domain knowledge to improve model performance.

Unsupervised Learning

An ML approach aimed at discovering inherent structures, groupings, or relationships in unlabeled input data.

Clustering

An unsupervised learning technique used to group similar data points together based on feature similarity.

Association Rule Learning

An unsupervised learning method used to uncover interesting relationships and co-occurrences between items in transaction databases.

Anomaly Detection

An unsupervised learning technique designed to identify rare data points or observations that deviate significantly from standard patterns.

Semi-supervised Learning

An ML paradigm that combines a small set of labeled data with a large amount of unlabeled data using techniques like pseudo-labeling.

Self-Supervised Learning

A learning technique where a model formulates pretext tasks to generate pseudo-labels from unlabeled data on its own before solving downstream tasks.

Reinforcement Learning (RL)

A machine learning paradigm where an autonomous agent learns optimal decision-making by performing actions within an environment to maximize cumulative rewards.

Agent

In reinforcement learning, the learner or decision-making entity that interacts with the environment.

Environment

In reinforcement learning, the external system with which the agent interacts, transitions states, and receives feedback.

Policy

In reinforcement learning, the strategy or rule set that an agent uses to determine which action to take based on its current state.

RLHF (Reinforcement Learning from Human Feedback)

A method that incorporates human preference data into an RL reward model to align generative models and LLMs with human goals.

Overfitting

A model condition where a model memorizes noise and patterns in the training data too closely, resulting in strong training performance but poor performance on unseen evaluation data.

Underfitting

A model condition where a model is too simple or lacks essential features, causing it to perform poorly on both training and test data.

Bias

The error introduced by approximating a real-world problem with an overly simple model, where too much of this leads to underfitting.

Variance

The sensitivity of a model to small fluctuations or changes in the training set, where too much of this thing leads to overfitting.

Precision

A classification performance metric calculated as $\frac{\text{TP}}{\text{TP} + \text{FP}}$ ; preferred when false positives carry a high cost.

Recall

A classification performance metric calculated as $\frac{\text{TP}}{\text{TP} + \text{FN}}$ ; preferred when false negatives carry a high cost.

F1 Score

The harmonic mean of precision and recall calculated as $2\times\frac{\text{Precision} \times\text{Recall}}{\text{Precision} + \text{Recall}}$ , useful for evaluating balanced performance on imbalanced datasets.

AUC-ROC (Area Under the Receiver Operating Characteristic curve)

A performance metric ranging from 0 to 1 that evaluates classification models across multiple classification thresholds by plotting true positive rate against false positive rate.

MAE (Mean Absolute Error)

A regression metric measuring the average absolute magnitude of errors between predicted and actual values, calculated as $\frac{1}{n}\sum_{i=1}^{n}|y_{i}-\hat{y}_{i}|$ .

MAPE (Mean Absolute Percentage Error)

A regression metric measuring average prediction error as a percentage, calculated as $\frac{1}{n}\sum_{i=1}^{n}\left|\frac{y_i - \hat{y}_i}{y_i}\right|$ .

RMSE (Root Mean Squared Error)

A regression metric measuring the square root of average squared differences between predicted and actual values, calculated as $\sqrt{\frac{\sum_{i=1}^{n} (\hat{y}_i - y_i)^2}{n}}$ .

R² (R Squared)

A regression metric that quantifies the proportion of variance in the target variable explained by the inputs, with values closer to 1 indicating stronger fit.

Real-Time Inferencing

An inferencing mode where model predictions are generated immediately as data arrives, prioritizing speed and low latency.

Batch Inferencing

An inferencing mode where large amounts of data are processed all at once asynchronously, prioritizing overall accuracy and throughput over response speed.

Edge Inferencing

Running model predictions locally on low-power, localized devices (such as a Raspberry Pi) to achieve low latency, reduced compute overhead, and offline capabilities.

Hyperparameters

Configurable settings defined prior to model training that structure the model architecture and dictate the learning process.

Learning Rate

A hyperparameter that determines the step size taken when updating model weights during training optimization.

Batch Size

A hyperparameter specifying the number of training examples processed before updating the model's weights in one iteration.

Epoch

A hyperparameter metric defined as one full pass of the training algorithm over the entire training dataset.