The goal of INTERACT is to develop interactive machine learning algorithms motivated by applications in natural language understanding (NLU).
Currently, the assumptions behind supervised approaches are unrealistic because most NLU applications have unique information needs,
and large collections of annotated data are necessary to achieve good performance.
We attack this fundamental limitation by following a collaborative training paradigm that breaks the distinction between annotation and training.
Because we are modelling human language which is rich, complex and compositional in nature, our departing point are compositional latent-state models.
Our main premise is that if the learning algorithm could ask the right questions only little human feedback would be needed. In other words, we must eliminate annotation redundancy.
To achieve this we combine: (1) An optimal human feedback strategy, with (2) inducing a latent structure of parts in the compositional domain.
Annotation effort will be minimized because the method will only request representative feedback from each latent class.
INTERACT marries representation learning (i.e. of parts) and active learning for compositional latent state models.
We empower the learner with the ability to generate samples and ask labels for any complete or partial structure in the domain.
We work under the framework of spectral learning of weighted automata and grammars (which are proper sub-classes or recurrent neural networks) and experiment with NLU tasks of increasing complexity,
from sequence and tree classification to parsing problems where the outputs are trees.
News: We are hiring
Phd-Student
Job location: UPC, Barcelona.
Position: Full Time PhD Student.
Expected start: 2020 (Flexible).
Duration: 42 months.
Qualification Requirements: Candidates should have completed a masters in computer science
or related field. Ideally they would have previous experience/knowledge/interest in at least on of the following areas:
1) machine learning, 2) natural language processing, 3) human/computer interaction.
Post-Doctoral Researcher
Job location: UPC, Barcelona.
Position: Post-Doctoral Researcher.
Expected start: 2020 (Flexible).
Duration: 24 months.
Qualification Requirements: Candidates should have completed a Ph.D in computer science
or related field and have a good publication record in machine learning and/or natural language processing.
Contact
If you are interested in any of these positions send and email to: ariadna.quattoni@gmail.com
