TAYSIR logo

Transformers+RNN: Algorithms to Yield Simple and Interpretable Representations

News

February 1st: 

Our first call for participation is now available. Check it out!!

What it is

Transformers+RNN: Algorithms to Yield Simple and Interpretable Representations (TAYSIR, the arab word for 'to simplify') competition is an on-line challenge on extracting simpler models from already trained neural networks. These neural nets are trained on sequential categorial (=symbolic) data. Some of these data are artificial, some come from real world problems (NLP, Bio-informatics, software Engineering, etc.)

The quality of the extracted models are evaluated following two directions:
► Approximation of the original model
► Simplicity of the extracted model

Two tracks are proposed:
► Neural nets trained for binary classification, that is, a language in the Formal Language Theory sense
► Neural nets trained for regression, that is, a function assigning a real number to any finite sequence of symbols (e.g the density estimation of Language modelling RNN) Each track consists of roughtly 15 trained models.

The trained models are in pytorch but available also in a MLFlow format for compatibility with other framework.

When

The Competition is scheduled to take place during Winter 2023, between mid-February and April.

Half a day will be dedicated to the competition results during the 16th International Conference on Grammatical Inference to be held in Morocco in July 2023.

Participants in TAYSIR will be encouraged to attend ICGI 2023 and to submit an extended abstract presenting their work (2 to 4 pages, including appendices) which will be appended to the proceedings of ICGI (Publisher: PMLR) in a track dedicated to the competition. These abstracts will be peer-reviewed primarily for clarity of presentation.

How to participate

A dedicated website will be available soon. There you will need to register in order to download the already trained model.

After running your extraction algorithms, we will expect participant to upload on our website their extracted model as MLFlow model.

Evaluation

We are going to use 2 types of metrics: one to evaluate the quality of the approximation, the other to evaluate the simplicity of the submitted model.

More details to be given soon.

Thanks

This competition is financialy supported by the ANR TAUDoS and the firm EURA NOVA.

Main organizers

  • Team Member

    Chihiro Shibata

    Hosei University, Japan

    Team Member

    Dakotah Lambert

    Université Jean Monnet, Saint-Etienne, France

    Team Member

    Jeffrey Heinz

    Stony Brook University, New York, USA

    Team Member

    Rémi Eyraud

    Université Jean Monnet, Saint-Etienne, France

Data Scientists

  • Aidar Gaffarov

    MLDM Mater program

    Badr Tahri Joutei

    MLDM Master program

    Mathias Cabanne

    EURA NOVA

Scientific Committee

    Ariadna Quattoni

    Universitat Politècnica de Catalunya

  • Bob Frank

    Yale University, USA

    Borja Balle

    DeepMind

    François Coste

    INRIA Rennes

    Jean-Christophe Janodet

    Université Paris-Saclay

    Matthias Gallé

    Naver Lab

    Sicco Verwer

    TU Delft