Sendai Tokyu Hotel, Sendai, Japan, October 6 - 8, 1997.
The 8th International Workshop on Algorithmic Learning Theory (ALT) was held in Sendai, Japan. This is the eighth in a series of annual conferences started in 1990 and it was supported jointly by the Japanese Society for Artificial Intelligence (JSAI) and the Tohoku University. The IFIP TC 1.4 sponsored a scholarship over 500 USD which was awarded to the student author N. V. Vinodchandran of India.*
For the technical sessions, 64 authors contributed 4 invited and 26 (out of 42 submitted) selected talks to the program. The contributions ranged over the subjects the design and analysis of learning algorithms, the theory of machine learning, computational logic of and for machine discovery, inductive inference, learning via queries, artificial and biological neural networks, pattern recognition, learning by analogy, Bayesian, MDL and MML estimation, statistical learning, inductive logic programming, application of learning to databases and biological sequence analysis. The four invited talks where presented by distinguished speakers, who gave nice and challenging overviews on their research fields and working areas:
(1) Manuel Blum (UC Berkeley and City University, Hong Kong) on ``Program error detection/correction: Turning PAC learning into PERFECT learning.'' Since the output of a PAC learner may still be incorrect on some data-items, these non-correct parts should be repaired in a second step by a special algorithm. The speaker presented a way how such an algorithm can be realized and encouraged the integration of error-correction into learning theory.
(2) Lenny Pitt (University of Illinois at Urbana) on ``Exploiting knowledge and concept use in learning theory.'' Many formal models of concept learning from examples showed that learnability is quite restricted and succeeds for easy concepts only. So scientists analyzed how human beings deal with this problem. Humans also fail to solve this problem totally, but among their approaches the use of background information is promising. The speaker outlined that the task of modelling background knowledge adaquately is an important challenge for learning theory.
(3) Masahiko Sato (Kyoto University) on ``Classical Brouwer-Heyting-Kolmogorov interpretation.'' The speaker follows a new model which gives a classical logical interpretation to the proof-theoretic operators considered by Brouwer, Heyting and Kolmogorov. The speaker deepens this idea and develops abstract mathematical entities where the truth of a proposition becomes equivalent ot the existence of proofs for this proposition. This approach is studied for the classical Peano Arithmetic and also for a non-deterministic ``system of catch/throw calculus.''
(4) Wolfgang Maass (Technische Universität Graz) on ``The relevance of time in neural computation and learning.'' The speaker studies realistic models for biological neurons. Therefore the focus of his attention is the model of the spiking neuron, which - roughly spoken - collects signals from the input and fires a spike at the output whenever a threshold is reached. The speaker shows that spiking neurons can not be simulated by traditional neural models provided that the algorithm, which decides when the neuron fires, has some minimum complexity, that is, the algorithm neither depends only on the least recent input nor only on the firing rates of the neurons at the input. This justifies a deeper study of the theory of spiking neurons which may now also be physically realized by the new generation of ``pulsed'' VLSI-chips.
The conference was chaired by Professor Akira Maruoka; Professor Ming Li served as Program Committee** Chair. The Proceedings of ALT'97 have been published as ``Algorithmic Learning Theory, 8th International Workshop, ALT'97, Sendai, Japan, October 1997'' within the series Lecture Notes in Artificial Intelligence number 1316, Springer-Verlag, Heidelberg, ISBN 3-540-63577-7, edited by Ming Li and Akira Maruoka.
Besides the scientific program, the participants where invited to a culinarily comprehensive banquet and enjoyed a marvelous post conference tour to the local attractions Matsushima and Zuiganji. This tour provided the view on more than 260 isles in the bay, whose unique beauty is due to the erosive actions carried out by wind and waves. After presenting the most well-known of them from a ferry, the tour turned to the more man-made attractions: a delicious shojin-ryori lunch and a visit to Zuiganji temple. This most famous Zen temple in Tohoku District was built in the year 828 and contains much outstanding artwork.
Finally, let me conclude this report by announcing that the next ALT will be held in Otzenhausen, Germany. It is organized by Michael Richter and Rolf Wiehangen from the University of Kaiserslautern, a place well-known for its successful soccer team. More information can be found on ALT'98's homepage.
Frank Stephan, firstname.lastname@example.org
Mathematisches Institut, Universität Heidelberg
The committee likes to thank the referees Hiroki Arimura, Shai Ben-David, Michael Ben-Or, Bernd Borchert, Xiaotie Deng, Carlos Domingo, Ricard Gavalda, Peter Gruenwald, Koichi Hirata, Ker-I Ko, Satoshi Kobayashi, Takeshi Koshiba, Tetsuro Nishino, Eli Shamir, Takeshi Shinohara, Noriko Sugimoto, Sei-ichi Tani, Noriyuki Tanida, John Tromp, Lidror Troyansky, Reyuhei Uehara, Yuri Ventsov, Mitsuo Wakatsuki, Kenji Yamanishi and Thomas Zeugmann for cooperating in the reviewing process.
to the ALT'97 Proceedings Page