Matteo Gagliolo - CoMo Home Page

Ciao! I am a postdoc researcher at the CoMo lab (Vrije Universiteit Brussel).

I am currently working at applying Reinforcement Learning in a single and multi-agent setting, in the context of the LeCoPro project, funded by the IWT.

I was awarded a PhD by the Faculty of Informatics of the University of Lugano, Switzerland, for research performed at IDSIA, supervised by Jürgen Schmidhuber. My thesis, titled "Online Dynamic Algorithm Portfolios", was defended in March 2010.

During my PhD, I also visited IRIDIA (Université Libre de Bruxelles), and cooperated with the Statistics Institute of Université Catholique de Louvain, supported by a Swiss SNF grant for prospective researchers.

Before my PhD, I was trained as an electronic engineer at the Università degli Studi di Genova (Italy).

Matteo Gagliolo

News

• CoMo is organizing a special session at IEEE WCCI 2012, on Real World Applications of Reinforcement Learning.
• I attended the 15th International Conference on System Theory, Control and Computing, ICSTCC 2011.
• I joined the Program Committee of the NIPS Workshop on Bayesian Optimization, Experimental Design and Bandits, Sierra Nevada, Spain, December 2011.
• I attended the 9th European Workshop on Reinforcement Learning, EWRL 2011.
• I attended the Santa Fe Institute's 2011 Complex Systems Summer School.

Research interests

• Single and multi-agent reinforcement learning, multi-armed bandit problems, online learning.
• Social learning, evolutionary game theory, complex dynamic networks.
• Recurrent artificial neural networks, neuroevolution and evolutionary computation in general.
• Static and dynamic algorithm portfolios, restart strategies, meta-learning, algorithm selection.
• Survival analysis and multilevel analysis methods for algorithm performance modeling.

Publications   

• Gagliolo, M., Van Vaerenbergh, K., Rodriguez, A., Nowé, A., Goossens, S., Pinte, G., and Symens, W. (2011b). Policy search reinforcement learning for automatic wet clutch engagement. In 15th International Conference on System Theory, Control and Computing - ICSTCC 2011, pages 250-255. IEEE. [ bib | event ]
• Pinte, G., Stoev, J., Symens, W., Dutta, A., Zhong, Y., Wyns, B., De Keyser, R., Depraetere, B., Swevers, J., Gagliolo, M., and Nowé, A. (2011). Learning strategies for wet clutch control. In 15th International Conference on System Theory, Control and Computing - ICSTCC 2011, pages 467-474. IEEE. [ bib | event ]
• Rodriguez, A., Gagliolo, M., Vrancx, P., Grau, R., and Nowé, A. (2011). Improving the performance of continuous action reinforcement learning automata. 9th European Workshop on Reinforcement Learning, EWRL 2011. [ bib | event | .pdf ]
• Gagliolo, M., Van Vaerenbergh, K., Rodriguez, A., Nowé, A., Goossens, S., Pinte, G., and Symens, W. (2011a). Policy gradient methods for controlling systems with discrete sensor information. 20th Annual Belgian-Dutch Conference on Machine Learning - BeNeLearn, online proceedings, p. 115-116. [ bib | event | .pdf ]
• Gagliolo, M. and Schmidhuber, J. (2011). Algorithm portfolio selection as a bandit problem with unbounded losses. Annals of Mathematics and Artificial Intelligence, 61(2):49-86. [ bib | DOI | .pdf | Abstract ]
• Goossens, S., Pinte, G., Symens, W., Gagliolo, M., Rodriguez, A., and Nowé, A. (2011). Reinforcement learning for repetitive systems with discrete sensors. In 30th Benelux Meeting on Systems and Control - Book of Abstracts, page 149, Gent, Belgium. Universiteit Gent. [ bib | event | slides | .pdf ]
• Gagliolo, M. and Legrand, C. (2010). Algorithm survival analysis. In Bartz-Beielstein, T., Chiarandini, M., Paquete, L., and Preuss, M., editors, Experimental Methods for the Analysis of Optimization Algorithms, pages 161-184. Springer, Berlin, Heidelberg. [ bib | DOI | .pdf | Abstract ]
• Gagliolo, M. and Schmidhuber, J. (2010). Algorithm selection as a bandit problem with unbounded losses. In Blum, C. and Battiti, R., editors, Learning and Intelligent Optimization. 4th International Conference, LION 4, Venice, Italy, January 18-22, 2010. Selected Papers, volume 6073 of LNCS, pages 82-96, Berlin, Heidelberg. Springer. [ bib | DOI | arXiv | event | slides | .pdf | Abstract ]
• Gagliolo, M. (2010). Online Dynamic Algorithm Portfolios. PhD thesis, IDSIA/University of Lugano, Lugano, Switzerland. [ bib | slides | .pdf | Abstract ]
• Gagliolo, M., Legrand, C., and Birattari, M. (2009). Mixed-effects modeling of optimisation algorithm performance. In Stützle, T., Birattari, M., and Hoos, H. H., editors, Engineering Stochastic Local Search Algorithms. Designing, Implementing and Analyzing Effective Heuristics, Second International Workshop, SLS 2009, Brussels, Belgium, September 3-4, 2009. Proceedings, volume 5752 of LNCS, pages 150-154. Springer. [ bib | DOI | event | poster | .pdf | Abstract ]
• Gagliolo, M. and Schmidhuber, J. (2009). Towards distributed algorithm portfolios. In Corchado, J. M., Rodríguez, S., Llinas, J., and Molina, J. M., editors, International Symposium on Distributed Computing and Artificial Intelligence 2008 (DCAI 2008), volume 50 of Advances in Soft Computing, pages 634-643. Springer. [ bib | DOI | event | slides | .pdf | Abstract ]
• Gagliolo, M. and Schmidhuber, J. (2008). Algorithm selection as a bandit problem with unbounded losses. Technical Report IDSIA - 07 - 08, IDSIA. Published in LION4 Proceedings, Springer LNCS, 2010. [ bib | arXiv | Abstract ]
• Schmidhuber, J., Wierstra, D., Gagliolo, M., and Gomez, F. (2007). Training recurrent networks by evolino. Neural computation, 19(3):757-779. [ bib | DOI | .pdf | Abstract ]
• Gagliolo, M. (2007). Universal search. Scholarpedia, 2(11):2575. [ bib | DOI | http ]
• Gagliolo, M. and Schmidhuber, J. (2007). Learning restart strategies. In Veloso, M. M., editor, IJCAI 2007 - Twentieth International Joint Conference on Artificial Intelligence, vol. 1, pages 792-797. AAAI Press. [ bib | event | slides | .pdf | Abstract ]
• Gagliolo, M. and Schmidhuber, J. (2006d). Learning dynamic algorithm portfolios. Annals of Mathematics and Artificial Intelligence, 47(3):295-328. AI&MATH 2006 Special Issue. [ bib | DOI | slides | techrep | .pdf | Abstract ]
• Gagliolo, M. and Schmidhuber, J. (2006b). Gambling in a computationally expensive casino: Algorithm selection as a bandit problem. Online Trading of Exploration and Exploitation - NIPS 2006 Workshop. [ bib | event | .pdf ]
• Gagliolo, M. and Schmidhuber, J. (2006c). Impact of censored sampling on the performance of restart strategies. In Benhamou, F., editor, Principle and Practice of Constraint Programming - CP 2006, 12th International Conference, CP 2006, Nantes, France, September 25-29, 2006, Proceedings, volume 4204 of LNCS, pages 167-181. Springer. [ bib | DOI | event | slides | .pdf | Abstract ]
• Gagliolo, M. (2006). Dynamic meta learning. 50th Anniversary Summit of Artificial Intelligence, Monte Veritá, Switzerland. [ bib | event | poster | .pdf ]
• Schmidhuber, J., Gagliolo, M., Wierstra, D., and Gomez, F. (2006). Evolino for recurrent support vector machines. In Verleysen, M., editor, ESANN '06 - 14 th European Symposium on Artificial Neural Networks, pages 593-598, Evere, Belgium. d-side. [ bib | arXiv | event | techrep ]
• Gagliolo, M. and Schmidhuber, J. (2006a). Dynamic algorithm portfolios. AI&MATH '06 - Ninth International Symposium on Artificial Intelligence and Mathematics. [ bib | event | slides | .pdf ]
• Schmidhuber, J., Gagliolo, M., Wierstra, D., and Gomez, F. (2005). Evolino for recurrent support vector machines. Technical Report IDSIA - 19 - 05, IDSIA. Shorter version published in ESANN 2006 Proceedings. [ bib | arXiv | .pdf | Abstract ]
• Gagliolo, M. and Schmidhuber, J. (2005b). Towards life-long meta learning. Inductive Transfer : 10 Years Later - NIPS 2005 workshop. [ bib | event | .pdf ]
• Gagliolo, M. and Schmidhuber, J. (2005a). A neural network model for inter-problem adaptive online time allocation. In Duch, W., Kacprzyk, J., Oja, E., and Zadrożny, S., editors, Artificial Neural Networks: Formal Models and Their Applications - ICANN 2005, 15th International Conference, Warsaw, Poland, September 11-15, 2005, Proceedings, Part 2, volume 3697 of LNCS, pages 7-12. Springer. [ bib | DOI | event | slides | .pdf | Abstract ]
• Gagliolo, M., Zhumatiy, V., and Schmidhuber, J. (2004). Adaptive online time allocation to search algorithms. In Boulicaut, J., F.Esposito, Giannotti, F., and Pedreschi, D., editors, Machine Learning: ECML 2004. Proceedings of the 15th European Conference on Machine Learning, Pisa, Italy, September 20-24, 2004, volume 3201 of LNCS, pages 134-143. Springer. [ bib | DOI | event | slides | techrep | Abstract ]
• Schmidhuber, J., Zhumatiy, V., and Gagliolo, M. (2004). Bias-optimal incremental learning of control sequences for virtual robots. In et al., F. G., editor, Proc. 8th Conference on Intelligent Autonomous Systems, IAS-8, pages 658-665, Amsterdam, NL. IOS Press. [ bib ]
• Anguita, D. and Gagliolo, M. (2002). Mdl based model selection for relevance vector regression. In Dorronsoro, J., editor, Artificial Neural Networks - ICANN 2002, volume 2415 of LNCS, pages 468-473. Springer. [ bib | DOI | .pdf | Abstract ]