Homecoming: language, learning, complexity and crowds

Marta Arias, Ramon Ferrer and Núria Pelechano joined the LSI faculty at the beginning of this term. The first two of them got their engineering degrees from the Barcelona School of Informatics, UPC (Facultat d'Informàtica de Barcelona) and Núria was a student in València Technical University Computing School (Facultat d'Informàtica de València). They are back home after their personal international and tran disciplinary journeys.

Marta Arias has joined the LARCA research group, led by José L. Balcázar. She got her degree in Computer Engineering at UPC in 1998. Her final career project was focused on textual corpus. This helped her in getting her first professional job at INCYTA, a company now called Translendium that specializes in products for linguistic processing. After sometime there, she decided to expand her knowledge and widen her horizons. She started a PhD at Edinburgh University under the supervision of Roni Khardon. In 2000 Marta followed him to his new academic post at Tufts University in Boston. There, she finished her PhD Thesis in Computational Learning, "Exact Learning of First-Order Expressions from Queries", in May 2004. Leslie Valiant, one of the most relevant scientists in Computational Learning Theory, had been Marta's advisor's advisor. So she is an "academic granddaughter" of Valiant.

After that, Marta moved to Columbia University, New York. There, her strong theoretical background was put to practical test in the area of power network alarm detection and prediction. For two years, her group was involved in an industrial contract with the power distribution company for the City of New York. It was no small task to sift in real time through all the incoming unreliable data and detect alarms and emergencies. It was her task to ensure that the system was able to discriminate whether data corresponded or not to a real alarm or when the predictive model was no longer valid and, then, to discard it. This work resulted in an award from AAAI.

In LARCA she intends to return to more theoretical research. However, her choice of possible lines of research shows some influence of her previous practical experience. She will work with Ricard Gavaldà on learning methods that are able to adapt to "concept drift". These systems take data and calculate how far the real results are from that data coherent with the underlying model. From this, the system is able to automatically propose another model that explains or predicts better the incoming data. Marta is working on extending different variations of the "weighted majority" family of algorithms that assign a different value to each model so that it is possible to decide which one to discard or accept. She will also work with José L. Balcázar on the learning complexity of an important class of logic equations, Horn clauses, and on the connections of this problem with Data Mining.

During these few years Marta has got the following awards:

- Award for Deployed Application in the Conference on Innovative Applications of Artificial Intelligence (IAAI 2006)

- Machine Learning Journal Best Student Paper Award, 13th International Conference on Inductive Logic Programming (ILP 2003)

- Outstanding Graduate Researcher in Engineering Award, School of Arts and Sciences and the School of Engineering of Tufts University (2003)

- Marta Arias's personal page: http://www.lsi.upc.edu/~marias/

- LARCA's research group page: http://www.lsi.upc.edu/%7Ebalqui/larca/larca0.htmlº
 

Ramon Ferrer is a scientist with an rich multidisciplinary background. He is back at the place where he was "born" as a Computer Engineer, the UPC. He got his degree from the Informatics School here. After this, he worked for a short time on a research project about neuronal nets and language acquisition with Professor Josep Maria Sopena (University of Barcelona), who now is a visiting at the LSI department. Still fascinated by language, he decided to start his PhD on this topic. This effort resulted in his thesis "Language: universals, principles and origins" (supervised by Ricard Solé who worked at the UPC and at Pompeu Fabra Universities and Santa Fe Centre for Complexity).
 

During this period Ramon Ferrer received the UPC's Doctoral Dissertation Extraordinary Award (2003). Also, his research was awarded the Barcelona City Research Prize, 2003. To top it off, that same year he was awarded the Joan Lluís Vives Prize, 2003, to scientific communication for an article about the characteristics of language. No wonder that he remembers this period as one of the best times of his life, although he mentions the research atmosphere in his group as the best of it all. Once he finished his PhD, he spent two years in Rome at La Sapienza University. He came back to Barcelona in 2006 to work with Albert Díaz in the Complex Systems group at the Fundamental Physics Department at University of Barcelona. He finally joined UPC last September as a lecturer in our department.

Here he has joined the Natural Language Processing Research Group. Ramon sees his arrival in this group as a very promising start. He wants to continue his studies on statistical regularities of languages. He is also keen on introducing some engineering flavour to his research. In the GPLN group he found several kinds of researchers. Some of them work on similar areas to the ones Ramon is interested in. Other ones are more focused towards as other types of technological questions related to language: automatically language learning, word analysis...etc. Ramon is open to cooperate with whoever has common interests with him.

Ramon can talk about language for hours. We found out with him, for example, that some established myths about human language fall apart under his findings. For example, the belief that human language is unique, having no features in common with the rest of beings. For one, some of the long term regularities he found in the language of dolphins seem to point to a very different conclusion. As he says, maybe a paramecium doesn't communicate as we humans do, but once you reach a certain level of complexity the differences between human and non-human languages start to become blurred.

Ramon is acting as the Local Coordinator of the 7th International Congress about Language Evolution. that will take place soon in Barcelona. He defines this as a triple-event. There is a congress, a workshop concerning communication in big apes and a museum evening at CosmoCaixa with speakers both from the congress and the workshop like, for example: Gary Marcus, Friedemann Pulvermüller and David Premack.

- Ramon Ferrer's personal page: http://www.lsi.upc.edu/~rferrericancho/

- Natural Language Processing page: http://www.lsi.upc.edu/%7Enlp/web/

- International Evolution of Language Conference 2008 page:   

    http://stel.ub.edu/~evolang2008/

Núria Pelechano studied Computer Engineering at Universitat de València. Once she finished her degree, she took a master in Vision, Imaging and Virtual Environments at the University College London. It was during her time in London that she decided to follow on to doctoral studies under the supervision of different professors from Valencia and London. His advisor in London was Melvyn Slater (now at LSI too) and he was the one who encouraged her to apply for a Fulbright scholarship. With it, in 2003, she started a PhD at University of Pennsylvania (Philadelphia) under the supervision of Dr. Norman I. Badler. She received her PhD in 2006.

Now, she is back in the Old Continent. She is going to work again with her first advisor, Melvyn Slater, with the MOVING group at LSI department. She will take part in a big CICYT - funded research project. Núria will be doing research on crowd simulations and on coordinated multi-agent simulation.

The area of Multi Agents Systems (MAS) is a subfield of Artificial intelligence (AI) that aims to develop mechanisms for the coordination of the behaviours of independent agents. MAS requires agents to act efficiently both autonomously and as a part of a team of agents. Núria works with agents acting in real time collaborative environments in which there are also real humans. Núria admits that in her real life, sometimes she can't help looking at people with her scientist eyes studying the way they behave.

The problem on present crowd simulations is how to evaluate their results, that is, how to decide that a simulation is better than another one. The new CICYT project focuses on research this issue. Melvyn Slater will be participating in this project too, and the idea is to mix both his knowledge on people behaviour in virtual environments and Núria's to come up with an evaluation method.

We can find multiple real life applications for the research Núria is involved in. Animation in movies, may be the best known one. However, present systems used by the movie industry, like MASSIVE, and in contrast to Núria's are geared towards short simulations, typically a few seconds, compared to the over 15 minutes simulations that Núria gets with her virtual and more realistic agents. They are more realistic because they don't react in the way that most simulators for crowds do. Instead of treating each agent as a particle interacting with other particles, Núria's agents have their own goals and necessities, so that reactions and interactions are more human-like.

In Architecture this type of simulations are used to analyze human flows on rush hours on specific buildings while they are still being designed. Núria began working in this field when she started her post-doctoral research at the TC Chan Centre for Building Simulation and Energy Studies with Dr. Ali Malkawi and the Centre for Human Modelling and Simulation with Dr. Norman I. Badler in the USA.

Núria was awarded the 2005 Teaching Practicum Award 05-06. Computer and Information Science Department. University of Pennsylvania.