(The picture is from 2012 or so. I might or might not have aged in the meantime.)
Germán Kruszewski
I am a Marie Skłodowska Curie Action (MSCA) Postdoctoral Fellow hosted by the COLT Lab at the Universitat Pompeu Fabra, researching open-ended evolution through artificial chemistry methods and its applications to language understanding. For this, I am building on my earlier work on Combinatory Chemistry. I am also in a sabbatical from my position as a Senior Scientist at NAVER LABS Europe, where I have worked on aligning language models to preferences and other topics connected to controlled generation.
Before, I was a Postdoctoral Researcher at Facebook AI Reserch where I have studied state-of-the-art Natural Language Processing systems in relation to their language learning skills, and sought to develop a more holistic understanding of intelligence, by trying to model the emergence of complex adaptive systems in silico.
Previously, I did my PhD in Computational Linguistics under the direction of Marco Baroni. I worked in the area of Distributional Semantics, and more specifically, in understanding the strengths and limitations of distributional models when trying to account for the richness of human conceptual knowledge. To this end, I've explored how these semantic representations can be used in inference. [Toggle mini-CV]
-
-
Naver Labs Europe
2020 - Present
Senior Scientist
-
Facebook AI Research
2017 - 2020
Postdoctoral Researcher
-
Facebook AI Research
Summer 2016
Summer intern under the direction of Tomas Mikolov
-
IBM Watson
Summer 2015
Summer intern in the Watson Concept Expansion team.
-
University of Trento
Fall 2012 - Fall 2016
PhD Student at the Center for Mind/Brain Sciences of the University of Trento. Working on inference with Distributional Semantic Models. Advised by Marco Baroni.
-
INRIA/LORIA
Fall 2011 - Fall 2012
Research assistant working under the direction of Claire Gardent.
-
University of General Sarmiento
2009 - 2011
T.A./Researcher
-
University of Buenos Aires
2003 - 2011
Licenciado en Ciencias de la Computacion (Computer Science MSc equivalent). Thesis title: Choice Based Revenue Management in hotel industry (topics: revenue management, choice models, capacity control). Advised by Gustavo Vulcano.
Publications
Should you marginalize over possible tokenizations?
N. Chirkova, G. Kruszewski, J. Rozen, M. Dymetman
2023
disco: a toolkit for Distributional Control of Generative Models
[CODE]
G. Kruszewski, J. Rozen, M. Dymetman
In Proceedings of ACL 2023 Demo Track [outstanding paper]
2023
Aligning language models with preferences through f-divergence minimization
D. Go, T. Korbak, G. Kruszewski, J. Rozen, N. Ryu, M. Dymetman
2023
On reinforcement learning and distribution matching for fine-tuning language models with no catastrophic forgetting
T. Korbak, H. Elsahar, G. Kruszewski, M. Dymetman
2022
On reinforcement learning and distribution matching for fine-tuning language models with no catastrophic forgetting
T. Korbak, H. Elsahar, G. Kruszewski, M. Dymetman
2022
Emergence of Self-Reproducing Metabolisms as Recursive Algorithms in an Artificial Chemistry
[EXTRAS] [CODE]
G. Kruszewski, T. Mikolov
2023
An approximate sampler for energy-based models with divergence diagnostics
B. Eikema, G. Kruszewski, H. Elsahar and M. Dymetman
2021
Controlling Conditional Language Models with Distributional Policy Gradients
T. Korbak, H. Elsahar, G. Kruszewski and M. Dymetman
2021
Energy-Based Models for Code Generation under Compilability Constraints
T. Korbak, H. Elsahar, M. Dymetman and G. Kruszewski
2021
Combinatory Chemistry: Towards a Simple Model of Emergent Evolution
[EXTRAS]
G. Kruszewski, T. Mikolov
In Proceedings of ALIFE 2020
2020
Resilient Life: An Exploration of Perturbed Autopoietic Patterns in Conway’s Game of Life [CODE]
A. Cika, E. Cohen, G. Kruszewski, L. Seet, P. Steinmann, W. Yin
In Proceedings of ALIFE 2020
2020
Class-Agnostic Continual Learning of Alternating Languages and Domains [CODE]
G. Kruszewski, I. Sorodoc, T. Mikolov
2020
The emergence of number and syntax units in LSTM language models [CODE]
Y. Lakretz, G. Kruszewski, T. Desbordes, D. Hupkes, S. Dehaene, M. Baroni
Proceedings of NAACL 2019
2019
Cooperative learning of disjoint syntax and semantics [CODE]
S. Havrylov, G. Kruszewski, A. Joulin
Proceedings of NAACL 2019
2019
The Fast and the Flexible: training neural networks to learn to follow instructions from small data. [CODE]
R. Leonandya, D. Hupkes, E. Bruni, G. Kruszewski
Proceedings of IWCS 2019
2019
Learning compositionally through attentive guidance
D. Hupkes, A. Singh, K. Korrel, G. Kruszewski, E. Bruni
Proceedings of CICLing 2019
2019
What you can cram into a single $&!#* vector: Probing sentence embeddings for linguistic properties. [CODE]
A. Conneau, G. Kruszewski, G. Lample, M. Baroni, L. Barrault
Proceedings of ACL 2018
2018
Memorize or generalize? Searching for a compositional RNN in a haystack
A. Liska, G. Kruszewski, M. Baroni
Pre-Proceedings of AEGAP 2018 Workshop.
2018
CommAI: Evaluating the first steps towards a useful general AI [CODE]
M. Baroni, A. Joulin, A. Jabri, G. Kruszewski, A. Lazaridou, K. Simonic, T. Mikolov
Proceedings of ICLR 2017 Workshop Track.
2017
Convolutional Neural Network Language Models
NQ. Pham, G. Kruszewski, G. Boleda
Proceedings of EMNLP 2016.
2016
There is no logical negation here, but there are alternatives: modeling conversational negation with distributional semantics
G. Kruszewski, D. Paperno, R. Bernardi and M. Baroni
Computational Linguistics.
2016
How cosmopolitan are emojis?
F. Barbieri, G. Kruszewski, D. Paperno, F. Ronzato and H. Saggion
ACM Multimedia 2016.
2016
The LAMBADA dataset: Word prediction requiring a broad discourse context [DATA]
D. Paperno, G. Kruszewski, A. Lazaridou, Q Pham, R. Bernardi, S. Pezzelle, M. Baroni, G. Boleda and R. Fernandez.
Proceedings of ACL 2016
2016
Deriving Boolean Structures from distributional vectors [CODE]
G. Kruszewski, D. Paperno and M. Baroni
Transactions of the Association for Computational Linguistics 3: 375-388
2015
Jointly optimizing word representations for lexical and sentential tasks with the C-PHRASE model.
N. Pham, G. Kruszewski, A. Lazaridou and M. Baroni
Proceedings of ACL 2015 (53rd Annual Meeting of the Association for Computational Linguistics), East Stroudsburg PA: ACL, 971-981.
2015
So similar and yet incompatible: Toward automated identification of semantically compatible words [DATA] [CODE]
G. Kruszewski and M. Baroni
Proceedings of NAACL HLT 2015 (2015 Conference of the North American Chapter of the Association for Computational Linguistics - Human Language Technologies), East Stroudsburg PA: ACL, 964-969.
2015
Dead parrots make bad pets: Exploring modifier effects in noun phrases [DATA]
G. Kruszewski and M. Baroni
Proceedings of *SEM 2014 (Third Joint Conference on Lexical and Computational Semantics), East Stroudsburg PA: ACL, 171-181.
2014
Don't count, predict! A systematic comparison of context-counting vs. context-predicting semantic vectors.
M. Baroni, G. Dinu and G. Kruszewski
Proceedings of ACL 2014 (52nd Annual Meeting of the Association for Computational Linguistics), East Stroudsburg PA: ACL, 238-247.
2014
Generation for Grammar Engineering.
C. Gardent and G. Kruszewski
Proceedings of INLG 2012 (Seventh International Natural Language Generation Conference), East Stroudsburg PA: ACL, 31-39.
2012
Generating grammar exercises.
L. Perez-Beltrachini, C. Gardent and G. Kruszewski
Proceedings of the Seventh Workshop on Building Educational Applications Using NLP (2012), East Stroudsburg PA: ACL, 147-156.
2012
Volunteer Work
En La Tecla computer literacy group
2007 - 2011
This was a project sponsored by the Exact and Natural Sciences Faculty of the University of Buenos Aires, having the double scope of teaching computer basics to low income people while also bridging the gap between the University and people living in poor regions of the city.
Free Time
(A rarely seen species)
Photography
I enjoy taking pictures. Here are some that I have taken over the years:
