Messages
The WNNLP'24 proceedings are now published, please have a look.
The workshop received 21 submissions (by 31 authors), of which all have been accepted for publication as part of the WNNLP 2024 proceedings (this volume). This constitutes a two-fold increase in comparison to WNNLP 2023, reflecting the rising interest to neural approaches in natural language understanding and generation.
The Programme Committee has selected the paper "Fact or Fiction?Exploring Diverse Approaches to Fact Verification with Language Models", by Tobias Opsahl, for the WNNLP 2024 Best Paper award. In this paper, the author applied novel methods (especially for subgraph extraction) not suggested in the track description. It allowed to outperform state of the art significantly.
The IN5550 course offered three obligatory assignment...
The exam webpage has now been updated with information about the presentations for our exam workshop on June 3rd. We copy the relevant update here for convenience:
All teams (whether PhD or MSc) should prepare a "lightning talk" for the final workshop on June 3rd at 12:15, including a few slides. PhD students will have 5 minutes each to present their paper; MSc teams will have 3 minutes each.
The track chairs will present a short task overview for each track, so you don't need to spend time and space on this in your presentation. Focus on the main contributions and results. Note that the time limit is strict due to our tight program, so be sure to not include more material than you can realistically cover.
The presentation slides should be pushed to the separate public (UiO-)git re...
Hi all, here's the leaderboard for the third obligatory assignment. The 3 teams of Master students with the best performance scores will get one bonus point, if not already at the maximum.
Congratulations!
| Team | Surprise metric (sentence transformers similarity) on the test set |
|---|---|
| Eivind Morris Bakke, Nora Winger Heggelund | 92.19 |
| Greg Taube | 91.53 |
| Caroline Kristin Vannebo, Kjetil K. Indrehus | 91.42 |
Next Monday, April 22nd, the IN5550/IN9550 course will again welcome a guest speaker. This time, it's Per Egil Kummervold from the AI lab of the National Library of Norway.
IMPORTANT: this is an in-person lecture, not a Zoom call! Please do come to the seminarrom Perl at 12:15, and you'll get a chance to talk to our speaker.
Per Egil is a well-known researcher in the field of Norwegian speech and language processing. In particular, he was heavily involved in training pioneering models like NB-BERT and NB-GPT-J.
But on Monday, he will talk about a more recent project at the National Library: a set of Norwegian text-to-speech models known as ...
In the course overview, you will now find a video and some selected readings for the coming lecture and QA session on neural machine translation, which will be held by Yves Scherrer. We encourage you to walk through the materials before the lecture.
Hi all, here's the leaderboard for the second obligatory assignment. The 3 teams of Master students with the best performance scores will get one bonus point, if not already at the maximum.
Congratulations!
| Team | F1 on the test set |
|---|---|
| Louis Give, Maria Antonietta Bruno and Timo Zaoral | 74.62 |
| Vebj?rn Haug K?sene | 71.28 |
| Eirik Eggset, Rayyan Ahmad Shah Syed and Torkild Finne |
71.22 |
The third obligatory assignment is now online, on the Git repository. It deals with natural language generation, in particular, abstractive summarization for English. You are going to employ large generative language models.
The assignment is due April 12, but feel free to start working on it, remember that the Easter holidays are coming. All the relevant topics have already been covered in the lectures, but feel free to ask anything on Mattermost, in GitHub issues, or at the next group on April 9.
Good luck!
Hi, here's the leaderboard for the first obligatory assignment. As promised, the 3 teams with the best performance scores will get a bonus point, if not already at the maximum.
Congratulations!
| Team | Macro F1 on the test set |
|---|---|
| Vebj?rn Haug K?sene | 0.85 |
| Louis Give, Maria Antonietta Bruno and Timo Zaoral | 0.85 |
| Aleksandar Davidov and Magnus S. Dhelie |
0.81 |
The second obligatory assignment is now online, on the Git repository. It will challenge you to use large pre-trained BERT language models for cross-lingual named entity recognition (NER), including a "surprise language".
The assignment is due March 15, but feel free to start working on it. We have already covered masked language models in the lecture this week, and will cover LLM fine-tuning next week.
You can expect hands-on exercises in using BERT models and HuggingFace Transformers library at the next group sessions.
Good luck!
The first obligatory assignment is now online, on the Git repository. It deals with building a neural document classifier using bags-of-words or word embeddings as features. It also touches upon recurrent neural networks (RNNs).
The assignment is due February 23, but feel free to start working on it. We have already covered linear classifiers, feed-forward neural networks and word embeddings in the lectures, and will cover RNNs next week.
We will provide extensive practice in building neural classifiers with PyTorch at the group sessions.
Hi all,
Thanks for attendance today! The slides for our first introductory lecture are now published, and you can find them in the course schedule. We also recommend to you to start with getting acquainted with the "Neural Network Methods for Natural Language Processing" textbook: please read the 1st chapter.
You are more than welcome to post your questions or comments to our Mattermost group chat.
This Thursday, the pre-recorded 2nd lecture will be published.
See you soon!
Welcome to our IN5550/IN9550 course which will guide you through deep learning applications to natural language processing!
The first introductory lecture this term will be held on Monday, January 15, at 12:15. We will go through course logistics (including routines for assignments and the final project-based exam) and motivate the now dominant use of neural architectures in Natural Language Processing (and most other sub-fields of Artificial Intelligence). The first lecture will be in-person.
Subsequent lectures (after the introductory one) will be provided to you in a pre-recorded format. Currently, the plan is that the lecture videos will be published every Thursday in the second half of the day. You can watch them whenever it is more convenient to you. Every Monday (at the official designated time slot of 12:15), we will have an interactive Q&A session in Z...