Aline Villavicencio is the Director of the Institute of Data Science and Artificial Intelligence, University of Exeter, affiliated to the  Department of Computer Science, University of Sheffield (UK), and has a Fellowship at the Alan Turing Institute. Before these, she held academic positions in the Institute of Informatics, Federal University of Rio Grande do Sul, Brazil (between 2005 and 2021) and in the School of Computer Science and Electronic Engineering, University of Essex, UK. 

Titulo: Testing the Language Limits of Large Language Models: The strange case of the idiomatic eager beaver in cloud nine

Large language models have been successfully used for capturing distinct (and very specific) word usages, and therefore could provide an attractive alternative for accurately determining meaning in language. However, these models still face a serious challenge when dealing with non-literal language, like that involved in Multiword Expressions (MWEs) such as idioms (make ends meet), light verb constructions (give a sigh), verb particle constructions (shake up) and noun compounds (loan shark). MWEs are an integral part of the mental lexicon of native speakers often used to express complex ideas in a simple and conventionalised way accepted by a given linguistic community. Although they may display a wealth of idiosyncrasies, from lexical, syntactic and semantic to statistical, that represents a real challenge for current NLP techniques, their accurate integration has the potential for improving the precision, naturalness and fluency of downstream tasks like machine translation. In this talk, I will present an overview of how advances in word representations have made an impact for the identification and modelling of idiomaticity and MWEs. I will concentrate on what models seem to incorporate of idiomaticity, as idiomatic interpretation may require knowledge that goes beyond what can be gathered from the individual words of an expression (e.g. “dark horse” as an unknown candidate who unexpectedly succeeds). 

Head of AI Research Group of the School of Engineering and Informatics of the University of Sussex (UK)

Training spiking neural networks with exact gradients

The energy consumption of modern AI is increasing at an alarming pace. Spiking neural networks are widely seen as possible more energy efficient alternatives, especially when run on emerging neuromorphic hardware.

The recently developed Eventprop algorithm allows us to calculate exact gradients in spiking neural networks and hence to train them with gradient descent.

In this talk I will give a brief introduction into this area and then present recent work in which we scale up gradient-based learning in SSNs to challenging speech recognition benchmarks. I will finish with a general discussion of strengths and weaknesses of exact gradient-based learning in SSNs and how the latter may be overcome.

PUC-RS

UFSC

UFRJ

Research Fellow at Université Catholique de Louvain (Belgium)

UFRJ

UFSCAR

PORTO DIGITAL e UFPE