Getting Started with Machine Learning

What's a better predictor of vehicle emissions than engine size? This hands-on talk shows you how to build a model in Python and find out.

#1about 3 minutes

The origins and evolution of machine learning

Machine learning evolved from modeling brain cells to powerful algorithms with the invention of backpropagation, large datasets like ImageNet, and the use of GPUs.

#2about 2 minutes

Core concepts of machine learning models

Machine learning is a subset of AI that uses statistics to find patterns, employing models like regression for numerical prediction and classification for assigning categories.

#3about 7 minutes

Building a model to predict CO2 emissions

A live coding demo shows how to use Pandas and Scikit-learn to train a random forest regressor on a Kaggle dataset for predicting vehicle CO2 emissions.

#4about 1 minute

Supervised, unsupervised, and reinforcement learning explained

The three main types of machine learning are explained, with reinforcement learning compared to getting a driver's license through environmental feedback.

#5about 3 minutes

Understanding deep neural networks and their challenges

Deep neural networks model the brain with layers and activation functions to handle complex data, but face challenges like overfitting, underfitting, and data bias.

#6about 5 minutes

Classifying images with noisy data using FastAI

This demo uses the FastAI framework to build an image classifier, demonstrating how to handle noisy data from web scrapes and interpret a confusion matrix.

#7about 1 minute

A look at advanced neural network architectures

An overview of specialized architectures includes Recurrent Neural Networks (RNNs) for sequential data, Transformers for language, and Autoencoders for data compression.

#8about 2 minutes

Applying machine learning in the automotive industry

Machine learning is used in the automotive sector for image segmentation in autonomous driving, predictive maintenance, and processing various sensor data.

#9about 2 minutes

The future of ML in quantum computing and biology

Exciting new applications for machine learning include optimizing quantum circuits with TensorFlow Quantum and predicting protein structures with AlphaFold.

#10about 5 minutes

Q&A on model reliability and explainable AI

The discussion addresses how to provide guarantees for model performance in the real world and the critical need for explainable AI to understand model failures.

#11about 9 minutes

Q&A on data, privacy, and model selection

This Q&A covers strategies for collecting diverse datasets, the impact of privacy regulations like GDPR, and how to choose the right model for a given task.