What we will cover in the course

Week by week

The main topics we will cover during this five-week course are as follows:

Week 1 – data collection and annotation:

• Choosing the problem – what type of question am I trying to answer?
• Data collection – things to consider
• Annotation – attaching meaning to your data

Week 2 – understanding and working with data:

• Organising your data – keeping track of large datasets
• Data division – splitting into training and test subsets
• Expanding your dataset – augmentation and synthetic data
• Releasing data – sharing your data

Week 3 – choosing and using models:

• Choosing an environment – software and programming tools
• Model selection – network architectures and transfer learning
• Training and inference – what is the difference?
• Improving performance – troubleshooting and parameter tuning

Week 4 – trusting results:

• How well can your model generalise on new data?
• Interpreting output – use of different performance metrics
• Improving results and ethical considerations

Week 5 – practical tips and tricks:

• Software and hardware tips – e.g. GPU acceleration
• Writing and reading machine learning papers – key information to provide
• Engaging with computer science – sharing expertise across disciplines

Week 1 in more detail

In broad terms, we will start off in week one by discussing a crucial step in any machine learning project, data collection.

This will include consideration of what type of problem you are trying to solve, or the question you want to answer. Is it classification? Or regression? Or something else?

Next, we follow up with some discussion of the raw data collection process itself, with particular focus on the taking of photographs to produce image data.

Finally, we look at data annotation, where you are able to attach meaning to the data you have collected, from which machine learning models can then learn.

Week 1 learning outcome

• Design experiments that will collect good quality data for use in machine learning and deep learning models