PyData Helsinki Meetup #3 : xAI & NLP (January 2021)

We are delighted to announce the 3rd edition of PyData Helsinki meetups. We bring two amazing talks about NLP and xAI. And again, due to the current situation, the meetup will be held online.

The event will be held on Google Meet and it will be recorded. Please make sure your mic is MUTED when you join the meeting.

Online meetup link: https://meet.google.com/njx-ddsh-rui

Agenda:

• 6:30pm - 6:35pm : Greetings & Introduction
• 6:35pm - 7:00pm: Explainable AI: alternative methods.
• 7:00pm - 7:05pm: Q&A
• 7:05pm - 7:30pm: Zero-shot text classification without GPT-3.
• 7:30pm - 7:45pm: Q&A & closing.

-------------------------------------------

# Talk1: Explainable AI: alternative methods, by Sara Tähtinen

There's an increasing interest towards understanding the black box models and their predictions. Despite the hype, the field mostly concentrates only on the "main-stream" xAI: LIME, SHAP and Grad-CAM, all of them focusing on explaining post-model predictions. But the world of xAI is a lot larger. It starts by understanding the training data. It goes on with understanding where the model fails and are there some hidden biases inside. And finally, it focuses on local post-model explainability by one sample at the time. Many methods are still in research phase and hasn't have a breakthrough yet. But interesting solutions are available and I'm willing to share some great methods to promote alternative xAI solutions.

Sara Tähtinen has a strong mathematical background with research education (PhD in theoretical particle physics) and she is specialized in computer vision and deep learning. She has worked as machine learning developer and data scientist. Currently, she is working on a project to study and build ways of explaining deep learning models on image data. In the project, her role is to find out promising research papers and testing them out.

# Talk2: Zero-shot text classification without GPT-3, by Aapo Tanskanen

Do you need to classify text but don't have a labeled dataset for training your own classifier? Thanks to the recent advances in the NLP field it's possible to classify text without training specific classifiers every time. This presentation will give you a brief introduction of zero-shot text classification, how it works and compares to traditional text classification, and quick live demos.

Aapo Tanskanen works as a Data Scientist at Gofore with special interest in modern NLP and its advances. In addition to NLP, he has worked with data analytics, machine learning, chatbots and knowledge management.

-------------------------------------------

PyData is a community for developers and users of open-source data tools. PyData is an educational program of NumFOCUS, a 501(c)3 non-profit organization in the United States. The PyData Code of Conduct governs this meetup.