CMU Portugal alum and research institute

André Martins, a Portuguese researcher from the Instituto de Telecomunicações (He. She) and the CMU Portugal alum, won the euroConsolidated Bean Research Council (ERC) grant CoG) from 2 m € To study artificial neural networks applied to natural language processing (NLP).

ERC Consolidator funding will be applied to DECOLLAGE (Deep Learning for Ignition for LANguage Generation) projectFocuses on finding solutions to some of the fundamental problems in NLP, using an innovative multidisciplinary methodology that combines tools from artificial intelligence, sparse modeling, neuroscience, and cognitive science.

Existing deep learning models of natural language processing, while impressive at times, are often unreliable and even misleading: they do not generalize well to new domains, they do not exploit contextual information, they are not well calibrated, and their memory is untraceable. These limitations stem from their homogeneous structures, which are suitable for cognition but not suitable for tasks that require a higher level of cognition.

“This project is a further step in overcoming the limitations of current NLP technologies, making it possible for humans and machines to communicate effectively in natural language and work collaboratively to solve increasingly difficult problems,” Andre Martins

With the funding available under this new grant, the research will focus on three main steps. First, using uncertainty and quality estimates as a guiding principle for controlled generation, combining this control mechanism with efficient encoding of contextual information and incorporation of multiple modalities. second, to develop sparse and structured models of long-term memory, with attention to descriptive representations, and third, to devise new mathematical models of sparse communication (bridging the gap between discrete and continuous representations), supporting end-to-end differentiation and enabling a co-working space where units and agents can communicate. These innovations will be applied to very challenging generation tasks, including machine translation, dialog, and open generation.

This ERC award follows on the heels of an early start grant won by the researcher in 2017, of €1.4 million, applied to DeepSPIN Project This allowed him to advance his research in deep learning and structured prediction to solve challenging tasks in natural language processing, including machine translation, quality estimation, and syntactic analysis.

André Martins was the first Portuguese MD from CMU. Graduate in Language Technologies, having completed his PhD in 2012 awarded by Técnico and Carnegie Mellon University (CMU). Since then, he has been heavily involved in the program’s initiatives, and has supervised two Ph.D.s. candidates and lead two large-scale projects CMU Portugal (MAIA And GoLocal) in strict collaboration with CMU teams. He is currently Vice President of Artificial Intelligence Research at Unbabel and Associate Professor at the Instituto Superior Técnico (Technico). He co-founded and co-organized the Lisbon Machine Learning School (LxMLS), already in its twelfth edition, which is supported by CMU Portugal, and is a Fellow of the ELLIS Association, the Pan-European Artificial Intelligence Network. With over 100 papers published, +6,000 citations, and a h-index of 37 at high-level conferences and journals, André stands among the top 2% of the most cited scientists in his scientific field by Elsevier.