Erica Melis

IJCAI-97 was held August 23 - 29 in Nagoya, Japan. 17 tutorials and 32 workshops from a variety of AI areas were held in conjunction with IJCAI-97 and attracted many participants, among others many Japanese participants.

IJCAI-97 was very interesting in terms of its invited talks, the technical program, the exhibition, and of course RoboCup-97 that magically attracted most of the IJCAI participants. As usual and very useful, the invited talks provided an overview on what is going on in particular areas of AI and they demonstrated the state-of-the-art in these areas. To name just three of the overall twelve invited talks -- which does not include any preference -- Margarete Boden gave a brilliant talk about creativity and Artificial Intelligence, Wolfgang Bibel wanted to convince the AI-community that theorem proving is well suited to tackle problems in many areas of AI, and Rich Sutton surveyed the recently booming reinforcement learning.

Rightly so, the Computers and Thought lecture by Leslie Kaebling "Why robby can't learn: the difficulty of learning in autonomous agents'', the Research Excellence lecture by Aravind Joshi "Relationship between natural language processing and AI'', and the awarded papers presentations of Lin (Applications of the situation calculus to formalizing control and strategic information: the Prolog Cut operator), Carbonell et al.(Translingual information retrieval: a comparative evaluation), and Huang & Russell (Object Identification in a Bayesian Context) got a lot of attention at IJCAI-97.

As for the technical program, I can give my limited personal impression only because of the ususal parallel execution of the program. My bias is towards planning, theorem proving, learning, and analogy and case-based reasoning. Here, the planning and learning sessions were interesting in general, theorem proving had little contributions, and analogy was rather unsatisfying. In a planning session, e.g., Givan &Dean's paper reconstructed STRIPS planning in terms of finite automata and translated model minimization into systematic regression.

Notably, computer-aided education, neural networks, information retrieval, and vision got a new or revived appearance. For the purpose of unifying the field again, this seems to be promising. Many contributions came from natural language processing and distributed AI in this year's IJCAI. Less papers than at former IJCAIs were presented in non-monotonic reasoning, automated reasoning in general, and knowledge representation.

Universitat des Saarlandes,

Fachbereich Informatik

D-66041 Saarbrucken

Germany