Emmanuel Dupoux, new CoML team leader
The CoML team is entering the Paris resarch center and is headed by Emmanuel Dupoux.
What are the main focuses of your research?
The CoML team's aim is to reverse engineer humans' ability to learn, i.e. to construct effective algorithms that can reproduce certain human processing capabilities, to study their mathematical and algorithmic properties, and to test their empirical validity as models of human cognition, measured using behavioural and neurological methods.
The expected results are more adaptable and reliable learning algorithms and quantitative models of cognitive processes, which can be used to predict human developmental and processing data.
What are your main areas of research?
Unsupervised or weakly supervised language learning: we develop algorithms that can find linguistic units in raw sensory data (audio and video recordings of parent/child interactions). We empirically test the predictions of these algorithms, seen as models of infants.
Human/machine benchmarking: we construct batteries of cognitive tests designed to evaluate how an artificial system performs the different components of a complex function and compare the results with human performance of the same task using the same data.
Do you have any industrial or academic partnerships?
The CoML team is a joint CNRS, ENS and EHESS research team. It works with MIT and Carnegie Mellon (USA), the RIKEN institute (Japan), MacQuarie University (Australia) and the Brno University of Technology (Czech Republic). It is part of a group that collects linguistic developmental data (DARCLE) involving laboratories in the US, Canada and the Netherlands.
We work with Facebook AI Research and Microsoft on automatic learning without machines and human comparative analysis and with IBM on speech technologies applied to the healthcare sector.