Accelerating AI with Continuous Delivery

In this video series, you’ll:

  • Learn the core concepts of Machine Learning
  • Train an AI model to detect Cats and Dogs in images
  • Familiarize yourself with the common AI toolset: Kaggle, DVC & HuggingFace
  • Automate your whole model training with CI/CD

Meet the host

Boštjan Cigan

A senior engineer turned into DevRel with a history of working with startups and enterprises (Povio, Sportradar, Automattic, Qloo, Presearch), solving engineering challenges, leading and growing teams and a passion for community building and knowledge sharing (Thinkful, Smart Ninja).

Twitter
linkedin

1. What is Machine Learning?

We start with the core concepts of machine learning, covering neural networks and their roles in recognizing speech and images. This foundation prepares us for building an app capable of identifying dogs and cats from images.

2. Building and Training a Model using Kaggle

Using Kaggle, we’ll walk through creating a cat vs. dog classifier. This session covers the basic steps of machine learning – data preparation, model tuning and understanding evaluation metrics such as error rates and confusion matrices.

3. Cats and Dogs Model Classifier in Python

We’ll clone a demo repository and set everything up to work on our local machines. We will also explain the codebase through the process of model training.

4. Data Storage and Versioning with DVC

Get introduced to DVC, the leading tool for handling data in machine learning projects. We’ll set up a workflow that stores and versions your datasets and models in the cloud (AWS).

5. Deploying to HuggingFace

We will deploy our trained model and app on HuggingFace, showing how to use this open-source platform for sharing and collaborating on AI projects and also making your app run online.

6. Setting Up CI/CD for AI Model Training

Wrap up the series by adding continuous integration and delivery (CI/CD) into our machine learning workflow. We’ll automate the build, training and deploy phases which will give us the ability to track and select the best-performing models over time.