Hands-on Workshop: Methods for Data Selection in Autonomous Vehicles

This is a handson lecture by a passionate community member who has a lot of experience with data and algorithms for self-driving cars.

Talk abstract:

When you train a neural network the most important thing is the data you train on. In this workshop I will go over several methods of active learning to select a better dataset to train your neural network on. At the end of the session I will also talk about the labeling process and the importance of the quality of your labels. The workshop will be very hands on, and at the end of the hour you will have several Jupyter Notebooks you can use in your own data selection pipeline.

00:00 Intro - What is active learning?
03:17 Roland Meert. The human brain is quick to adapt Traffic changes time, • A self-driving should be able handle unknown objects • The Al fleet should learn fast from situations! [1756.00]
05:02 Public datasets
05:42 Addressing the long-tail distribution of situations
11:22 The world will keep changing, and there is an infinite amount of edge cases
12:01 The human brain is quick to adapt
16:35 Selecting data before training a model
19:28 Active learning
25:43 Uncertainty sampling
29:33 Query by committee
33:05 Find similar samples
37:55 Putting it all together
39:28 Hands on experimentation!

[Chapters were auto-generated using our proprietary software - contact us if you are interested in access to the software]

This is a two part event - the first part is a talk, afterwards there is a 10 minute break and the second part is a hands-on session.

Presenter BIO:

Roland Meertens is product manager at Annotell, and is specialised in robotics projects. He set up machine learning projects all the way from sensor selection, data collection and labelling to deploying the model in production. Please approach him to talk about deep learning and neural networks.

-------------------------
Find us at:

[ddai]

Would love to hear any comment or input you have. Did you find the content interesting? Is there a place for improvement? Please leave any feedback you have here in a comment :)