Title

Pretrained models

Models that have already been trained on a large dataset for a general task, improving accuracy and speed for new datasets.

Vision Transformers

A type of deep learning model that uses self-attention mechanisms to process images, capturing long-range dependencies and global context.

Fine-tuning

A transfer learning technique where parameters of a pretrained model are updated by training for additional epochs on a different task.

Fit one cycle

A training schedule that gradually increases and then decreases the learning rate during training.

Half-precision (to_fp16)

A technique using 16-bit floating-point numbers to speed up training and reduce memory usage, while possibly sacrificing precision.

Fastkaggle

A Python library that simplifies working with Kaggle competitions by automating tasks like data downloading and package installation.

CNN (Convolutional Neural Network)

A specialized type of neural network designed for processing grid-like data, particularly effective for image recognition.

Pooling layers

Layers that downsample feature maps produced by convolutional layers, reducing spatial dimensions and making the network more robust.

Learning Rate Finder

A technique used to determine an optimal learning rate for a specific model and dataset during training.

SGD (Stochastic Gradient Descent)

An iterative optimization algorithm that updates neural network weights by minimizing a cost function using a mini-batch of data.