BHAD: Explainable unsupervised anomaly detection using Bayesian histograms
Alexander Vosseler

We present a Bayesian histogram anomaly detector (BHAD). BHAD scales linearly with the size of the data and allows a direct explanation of individual anomaly scores due to its simple linear form

Honey, I broke the PyTorch model >.< - Debugging custom PyTorch models in a structured manner
Clara Hoffmann

Honey, I broke the Pytorch model >.< No problem! In this talk, we'll build a toolbox to debug our models and prevent this from happening again -all by leveraging DL logic, synthetic data and pytest. Let's make our models unbreakable <3

Incorporating GPT-3 into practical NLP workflows
Ines Montani

Large language models like @OpenAI GPT-3 can complement existing machine learning workflows really well. You can get initial annotations from GPT-3, quickly fix them with an annotation tool like , and train a cheaper and better model.