Input is the information the model receives. In daily life, it is like what a person looks at before making a decision. For example, if you want to guess whether it will rain, the input could be dark clouds, strong wind, and humidity. In deep learning, input is simply what the model sees first.
Target is the correct answer the model is supposed to learn. It is like the answer printed at the back of a workbook. If the input is a photo of a cat, the target might be “cat.” If the input is a house description, the target might be its real price. The target is the truth the model is trying to match.
Prediction is the model’s answer after seeing the input. It is like a student giving an answer before checking the solution. If the model sees a house and says “€400,000,” that is its prediction. The prediction may be right or wrong, and learning happens by comparing it with the target.
Dataset is the full collection of examples used for learning. It is like a whole textbook full of practice questions and answers. In deep learning, each example usually contains an input and its target. The dataset is the material the model learns from.
Batch is a small group of examples taken from the dataset and processed together. It is like a student studying one page of exercises at a time instead of the whole textbook at once. Deep learning uses batches because working with smaller chunks is faster and more practical.
Epoch means one full pass through the training dataset. It is like reading through the entire textbook once from beginning to end. If the dataset is split into many batches, then finishing all those batches once is one epoch. Training usually needs many epochs.
Train set is the part of the data used to teach the model. It is like the exercises a student practices on before an exam. The model looks at this data, makes predictions, checks mistakes, and improves. This is where actual learning happens.
Validation set is a separate part of the data used to check how well the model is improving during training. It is like a mock quiz given during the semester. The model does not directly learn from it, but teachers and engineers use it to decide whether the model is getting better or starting to memorize too much.
Test set is the final set of unseen data used to measure the model at the end. It is like the real final exam. Since the model has not used this data during training or tuning, the test set gives a fair measure of how well the model performs on new examples.
Model is the system that turns input into prediction. It is like the student’s brain or method for solving problems. In deep learning, the model is usually a neural network. Its job is to find patterns in the input and use them to make useful predictions.
Parameter is an internal value inside the model that can change during learning. It is like a knob on a machine that can be adjusted to improve the output. In a student analogy, it is like adjusting habits or ways of thinking. In deep learning, parameters are the numbers the model learns.
Objective is the rule that tells the model what counts as doing well. It is like the scoring rule in an exam. If the model predicts badly, the objective gives a poor score; if it predicts well, the score improves. The objective gives the model a clear goal during training.