Get Started with Deep Java Library

Learn to make a deep learning model using Deep Java Library. In this project, we'll train an image classifier on the CIFAR-10 dataset. The model will be created incrementally to illustrate the process as well as highlight important components of computer vision models.

You will learn to:

Make a deep learning model using DJL

Train a model on CIFAR-10 dataset

Fine tune the model to increase accuracy

Add different layers in a model

Skills

Machine Learning

Deep Neural Networks

Deep Learning Basics

Prerequisites

Basic understanding of Java

Basic understanding of Machine Learning

Technologies

Java

Deep Java Library

Project Description

Deep Java Library (DJL) aims to build deep learning models natively in Java while utilizing popular machine learning engines like MXNet, TensorFlow, PyTorch, and so on. A significant advantage of DJL is that it can run on ARM architecture, along with the common x86 architecture. Gradle’s native support allows the code to be executed and compiled on Android and other Java platforms.

We’ll create a deep learning model from scratch and train our dataset on the CIFAR-10 dataset. CIFAR-10 is a small dataset with ten classes: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck. This project serves as a stepping stone to using DJL so that we can create more complex and robust models.

Project Tasks

Building Blocks

Task 0: Introduction

Task 1: Convolution layer

Task 2: Activation Function

Task 3: Pooling Layer

Task 4: Flatten layer

Task 5: Dense Layer

Building the Classifier

Task 6: Import the Dataset

Task 7: Get Accuracy

Task 8: Model with Diagram

Task 9: Adjust the Epochs

Congratulations!