10th November 2026

Lecture (60 m), Loss functions activity (30 m), Project time (30 m)

Slides 📝 

https://docs.google.com/presentation/d/1HB9moyC7jRsqq9zSWyN0oIIDsknoYWN3k6t6PbYNDPk/edit?usp=sharing 

Activity

Thinking through loss functions individual -> groups 

For each scenario, think about what loss function would train the model to the desired behavior. Describe the loss function in words, then try to write it as math. 

Train a classifier to detect 3 bird species. The labeled data contains 10,000 Hadada Ibis, 2,000 African Sacred Ibis, and 5 Southern Bald Ibis labels.  

I have 50,000 hours of unlabeled audio data. I want to discover the unique sound types in the data. 

I have labeled song recordings for 20 individual Levaillant’s Cisticola. I want to train a classifier to assign a song to one of these individuals, or tell me that the song belongs to none of the 20 known individuals. 

Credits:

Top image: This image was created with the assistance of DALL·E 3.