Computational Intelligence 

Abstract: In 1978, Patrick Hayes promulgated the Naive Physics Manifesto. (It finally appeared as an “official” publication in Hobbs and Moore 1985.) In this paper, he proposed that an allout effort be mounted to formalize commonsense knowledge, using first-order logic as a notation. This effort had its roots in earlier research, especially the work of John McCarthy, but the scope of Hayes’s proposal was new and ambitious. He suggested that the use of Tarskian seniantics could allow us to study a large volume of knowledge-representation problems free from the confines of computer programs. The suggestion inspired a small community of people to actually try to write down all (or most) of commonsense knowledge in predictate calculus. He launched the effort with his own paper on “Liquids” (also in Hobbs and Moore 1985), a fascinating attempt to fix ontology and notation for a realistic domain. Since then several papers in this vein have appeared (Allen 1984; Hobbs 1986; Shoham 1985). I myself have been an enthusiastic advocate of the movement, having written general boosting papers (1978) as well as attempts to actually get on with the work. (1982, 1985). I even coauthored a textbook oriented around Hayes’s idea (Charniak and McDermott 1985). It is therefore with special pain that I produce this report, which draws mostly negative conclusions about progress on Hayes’s project so far, and the progress we can expect. In a nutshell, I will argue that the skimpy progress observed so far is no accident, that in fact it is going to be very difficult to do much better in the future. The reason is that the unspoken premise in Hayes’s arguments, that a lot of reasoning can be analyzed as deductive or approximately deductive, is erroneous. I don’t want what I say in this paper to be taken as a criticism of Pat Hayes, for the simple reason that he is not solely to blame for the position I am criticizing. I will therefore refer to it as the “logicist” position in what follows. It is really the joint work of several people, including John McCarthy, Robert Moore, James Allen, Jerry Hobbs, Patrick Hayes, and me, of whom Hayes is simply the most eloquent.

Abstract: In this paper, we describe a model of causal reasoning that accounts for knowledge concerning cause-and-effect relationships and knowledge concerning the tendency for propositions to persist or not as a function of time passing. Our model has a natural encoding in the form of a network representation for probabilistic models. We explore the computational properties of our model by considering recent advances in computing the consequences of models encoded in this network representation.

Abstract: Resampling methods are commonly used for dealing with the class-imbalance problem. Their advantage over other methods is that they are external and thus, easily transportable. Although such approaches can be very simple to implement, tuning them most effectively is not an easy task. In particular, it is unclear whether oversampling is more effective than undersampling and which oversampling or undersampling rate should be used. This paper presents an experimental study of these questions and concludes that combining different expressions of the resampling approach is an effective solution to the tuning problem. The proposed combination scheme is evaluated on imbalanced subsets of the Reuters-21578 text collection and is shown to be quite effective for these problems.

Abstract: A new approach for learning Bayesian belief networks from raw data is presented. The approach is based on Rissanen's minimal description length (MDL) principle, which is particularly well suited for this task. Our approach does not require any prior assumptions about the distribution being learned. In particular, our method can learn unrestricted multiply-connected belief networks. Furthermore, unlike other approaches our method allows us to trade off accuracy and complexity in the learned model. This is important since if the learned model is very complex (highly connected) it can be conceptually and computationally intractable. In such a case it would be preferable to use a simpler model even if it is less accurate. The MDL principle offers a reasoned method for making this trade-off. We also show that our method generalizes previous approaches based on Kullback cross-entropy. Experiments have been conducted to demonstrate the feasibility of the approach.