The Flipped Classroom model for teaching Conditional Random Fields in an NLP course

This is a presentation that I gave regarding my experience in using the Flipped Classroom method for teaching Conditional Random Fields in an NLP course. Please find an article about that experience and the teaching material in the following link: https://www.aclweb.org/anthology/2021.teachingnlp-1.13/

Transcript

1. Manex Agirrezabal Centre for Language Technology Department of Nordic Studies and Linguistics Supervisors: Lis Lak Risager Patrizia Paggio TLHE program project The Flipped Classroom model for teaching Conditional Random Fields in an NLP course
2. Outline • Motivation • Context • Evaluation form • Activities • Results and discussion • Conclusion 27/05/2021 2
3. Outline • Motivation • Context • Evaluation form • Activities • Results and discussion • Conclusion 27/05/2021 3
4. Intro: What did I do? Why did I do it? • Flipped classroom method for teaching Conditional Random Fields in a Natural Language Processing course • Evaluate method • Spoiler! It works! And it is quite rewarding J. • Why? • I believe students reach a higher understanding of a topic 27/05/2021 4
5. Flipped classroom 27/05/2021 5
6. Outline • Motivation • Context • Evaluation form • Activities • Results and discussion • Conclusion 27/05/2021 6
7. Context: Program • The IT & Cognition program at UCPH • International and interdisciplinary program • Small group of students • Two years -> four semesters • Three areas: • Natural Language Processing • Image Processing • Cognitive Science • Language Processing I and II • Semester 1 and 2 27/05/2021 7
8. Context: Course and lecture • Intended Learning Outcomes (courses): • LP1: Basic knowledge about different tasks relevant to Natural Language Processing and their relationship to current society • LP2: Development of more advanced algorithms and their application in more specific tasks • Intended Learning Outcomes (lecture) • What are CRFs and why are they better than MEMMs? • Understand the Label Bias problem • Be able to apply CRFs for their own work 27/05/2021 8
9. Outline • Motivation • Context • Evaluation form • Activities • Results and discussion • Conclusion 27/05/2021 9
10. Pre classroom knowledge 1) Do you know what a MEMM is (Maximum Entropy Markov Model)? 2) Do you know what a CRF is (Conditional Random Field)? 3) Do you know what the Label Bias problem is? 4) Do you feel capable of using a CRF for your own problems, such as developing a Named Entity Recognition system? 27/05/2021 10
11. 27/05/2021 11
12. Outline • Motivation • Context • Evaluation form • Activities • Results and discussion • Conclusion 27/05/2021 12
13. Structure • Before class • Watch two video lectures that we made • MEMMs • CRFs • Read article about CRFs (Lafferty et al., 2001) 27/05/2021 13
14. Activities • Four activities • Increasing complexity, based on Bloom’s taxonomy 27/05/2021 14 Activity 1 Activity 2 Activity 3 Activity 4
15. Activities • In the next week, we worked on the practical aspect. How to use CRFs for training a Named Entity Recognizer, for example. 27/05/2021 15 Image from https://opendatascience.com/named-entity-recognition-milestone-models- papers-and-technologies/
16. Outline • Motivation • Context • Evaluation form • Activities • Results and discussion • Conclusion 27/05/2021 16
17. Evaluation (form) 27/05/2021 17 1) Do you know what a MEMM is (Maximum Entropy Markov Model)? 2) Do you know what a CRF is (Conditional Random Field)? 3) Do you know what the Label Bias problem is? 4) Do you feel capable of using a CRF for your own problems, such as developing a Named Entity Recognition system? 5) Do you feel that this structure (teaching style) is more rewarding? 6) Do you feel that this structure is more demanding (mentally)
18. Results and discussion 27/05/2021 18
19. Discussion • Evaluation seems positive • Learning goals were satisfied • 7/15 students felt this method is rewarding • Teachers required time for preparation is higher. 27/05/2021 19 Further comments hard not to be able to ask questions when you watch the lecture homework distribution was far from being optimal group work worked better it was nice to have time to understand the code and exercises
20. Outline • Motivation • Context • Evaluation form • Activities • Results and discussion • Conclusion 27/05/2021 20
21. Conclusion • Flipped classroom: relevant method for complex topics • Homework as lecture: One step beyond in the understanding of CRFs • Paper online! https://arxiv.org/pdf/2105.07850.pdf 27/05/2021 21
22. 27/05/2021 22
23. Manex Agirrezabal Centre for Language Technology Department of Nordic Studies and Linguistics Supervisors: Lis Lak Risager Patrizia Paggio TLHE program project The Flipped Classroom model for teaching Conditional Random Fields in an NLP course