About
We are a new research group led by Wilker Aziz within ILLC working on probabilistic models for natural language processing.
News
- 2024
- Probabll is going to Malta, come talk to us about uncertainty in NLP. At EACL (main conference) we will be presenting about methodology to evaluate LM’s calibration against human uncertainty and confidence vs. label variation. At UncertaiNLP, Bryan will be talking about the inadequacy of the mode in neural text generators and the overlooked role of generalisation.
- 2023
- EMNLP 2023 paper: evaluating uncertainty in neural text generators against human production variability
- New pre-print on uncertainty representation in NLP (from theory to applications), a collaboration out of the Third ELLIS NLP Workshop (in Dagstuhl)
- 2022
- Two papers accepted at EMNLP 2022: sampling-based MBR for NMT and calibration of text classifiers under human disagreement
We are hiring, apply by August 21 2022!- Gonçalo M. Correia passed his PhD defence! Congratulations Dr. Correia for your awesome thesis on Learnable Sparsity and Weak Supervision for Data-Efficient, Transparent, and Compact Neural Models.
- Wilker gave a talk at Structure Prediction for NLP (slides, colab demo).
- Bryan gave a talk at the University of Copenhagen on Distribution-Aware Decoding for NMT.
- Our Deep Learning 2 course is ready!
- Sparse communication via mixed distributions has been accepted at ICLR2022!
- 2021
- 2 EMNLP2021 papers led by Nicola De Cao: Editing Factual Knowledge in Language Models and Highly Parallel Autoregressive Entity Linking with Discriminative Correction.
- In this followup to our Coling2020 paper we take MBR decoding from proof-of-concept to a robust and scalable decision rule.
- Probabll and DeepSPIN team up once more, this time we contribute a formal account to mixed random variables and take the first steps towards a theory of communication using sparse continuous random variables.
- PhD and postdoc positions at the newly established Mercury Machine Learning Lab (a collaboration among Booking.com, TUDelft, and UvA) including 1 PhD position in ML/NLP, together with Ivan Titov. Deadline: August 23, 2021.
- 2020
- Is MAP Decoding All You Need? The Inadequacy of the Mode in Neural Machine Translation is Coling2020’s best paper!
- Efficient Marginalization of Discrete and Structured Latent Variables via Sparsity is a NeurIPS2020 spotlight!
- How do Decisions Emerge in Neural Models? has been accepted at EMNL2020!
- The Power Spherical Distribution accepted at INNF+2020!
- Effective Estimation of DGLMs accepted at ACL2020!
- Latent morphology NMT is an ICRL2020 spotlight!
- VI tutorial is going to Alicante
- Code
January 2019
We are born!