http://popups.lib.uliege.be/1373-5411/index.php?id=1660 Publications of Auteurs Hiroyuki Shioya fr 0 http://popups.lib.uliege.be/1373-5411/index.php?id=3623 It becomes more important to construct an information retrieval system that can give flexible responses to variable user's requests in accordance with the evolution of the internet. We proposed a document retrieval system based on the maximum entropy principle on a Bayesian network (Saito, et al. 1998). We can claim two advantages for this document retrieval system. The first is that this system can produce several candidates of keywords which will help a user in retrieving a useful document, even if he can not recall a group of adequate keywords. The second is that this system can be customized corresponding to an individual user by tuning the parameters of a Bayesian network. But there is a problem for this system that it does not produce a retrieval result for some combinations of keywords. We have solved this problem, in this paper, by introducing fuzzy requests. Thu, 26 Sep 2024 10:26:27 +0200 Tue, 08 Oct 2024 17:14:56 +0200 http://popups.lib.uliege.be/1373-5411/index.php?id=3623 http://popups.lib.uliege.be/1373-5411/index.php?id=2063 Several information measures have been used as the criteria in information theory, statistics and various fields of engineering. Especially an information divergence has been well used as the measure of the difference between two probability distributions. In this paper, we propose the pseudo information divergence, which functions as usual information divergence, if two measured probability distributions are in some family of specific distributions. We introduce an example of the pseudo information divergence, and apply it to the problem of training multi-layer perceptrons from the data with the gross error noise. Mon, 29 Jul 2024 10:47:59 +0200 Mon, 29 Jul 2024 10:48:09 +0200 http://popups.lib.uliege.be/1373-5411/index.php?id=2063 http://popups.lib.uliege.be/1373-5411/index.php?id=1653 A reinforcement learning (RL) is known as one of the machine learning methods, and has been applied to multi-agent problems. In this paper, we propose a new RL method using a Bayesian network (BN), which is a stochastic model and plays a role of the supervised learning procedure. An agent learns how to move under certain circumstances by an original RL method, and then the strategy is improved by using BN. We verify the effectiveness of our method by carrying out simulations for a certain multi-agent problem, and show that an agent learns its appropriate strategy for complicated tasks more effectively by using our method. Mon, 15 Jul 2024 16:19:04 +0200 Mon, 15 Jul 2024 16:19:11 +0200 http://popups.lib.uliege.be/1373-5411/index.php?id=1653