GEELEC: FUNDAMENTALS OF AI DEVELOPMENT

Supervised Learning

• The most common form of AI learning

Supervised Learning

Learns from labeled data: the model sees inputs and the correct

answers

Supervised learning

Learns to predict the correct output when given new inputs

Unsupervised Learning

The model gets raw data without labels

Unsupervised learning

It finds patterns, clusters, or groupings in the data on its own

Unsupervised

Used in recommendation systems, anomaly detection, customer

segmentation, and more

Semi Supervised learning

Combines a small amount of labeled data with larger set of unlabeled

data during training

Semi Supervised Learning

Useful when labeling data is expensive or impractical, such as in

speech recognition tasks where only a fraction of audio data is

transcribed

Model

is a system that produces outputs based on learning patterns from input data

Model

It is built by training a machine learning algorithm with relevant data.

Data

The raw material that teaches the model

Features

Key characteristics extracted from the data to help the model learn (e.g., keywords

in a message

Algorithms

The recipe that processes the data to find patterns

Labels

In supervised learning, these are the answers the model should learn to predict

Training Process

The loop where the algorithm learns from data and adjust itself

Data

Features

Algorithms

Labels

Training Process

Ingredients for a Model

Data collection

Gathering as many examples as possible that reflect the real-world

situations the model will face.

Pre Processing

Cleaning (removing duplicates, correcting errors), transforming (converting

text into numbers), and standardizing data formats. Also includes

normalizing values and handling missing data.

Model Initialization

Choose the algorithm and set initial values like learning rate or number of

layers. This step sets up the architecture that the training will optimize.

Training

Feed the preprocessed, labeled data into the model. The model will make predictions and compare them to the correct answers.

Loss calculation

The model calculates how far off it was from the correct answer using a

“loss function”.

Optimization

Based on the loss, the model adjusts its parameters to improve future

predictions. This is where learning actually happens.

Iteration

The training process runs many times through the full dataset (called

epochs) to refine predictions and reach stable, high performance.

Data collection

Pre processing

Model Initialization

Training

Loss calculation

Optimization

Iteration

Training the model

Good, representative data

Diverse examples

that match the

real-world use case

Balance Training

Enough positive

and negative cases

Avoiding Overfitting

Don’t memorize the

answers, learn the

patterns

Performance

Monitoring

Use metrics

(accuracy, loss) to

track progress

Good, Representative Data

Balance Training

Avoiding Overfitting

Performance Monitoring

What makes the training successful

Inference

is the stage where a trained model makes predictions using what it has already

learned.

Inference

It applied knowledge to new, unseen data.

Inference

Happens in real-time (e.g., chatbot) or in batches (e.g., email classification).

Inference

The model is not learning anymore - just generating outputs.

Validation

helps check if a model can make a good prediction on new data - not just memorize

training examples.

Validation

Common practice: Dataset are split into training, validation, and testing sets..

Validation

data is unseen during training and is used to fine-tune model hyperparameters.

Validation

It helps adjust the model and avoid memorizing (overfitting) or not learning enough

(undefitting).

Scientists

Cleans, prepares, and

explores data; finds

trends and builds

dataset

Data Engineer

Builds the pipeline that

trains and serves

models in production

environments.

AI ethicist

Guides fairness,

transparency,

accountability, and

ensures systems do no

harm.

AI Trainer

Breaks down AI

concepts for learners,

emphasizes

responsible use, and

helps bridge technical

and ethical gaps