Robert K. Lindsay

Literature-based discovery by lexical statistics

We report experiments that use lexical statistics, such as word frequency counts, to discover hidden connections in the medical literature. Hidden connections are those that are unlikely to be found by examination of bibliographic citations or the use of standard indexing methods and yet establish a relationship between topics that might profitably be explored by scientific research. Our experiments were conducted with the MEDLINE medical literature database and follow and extend the work of Swanson.

Toward discovery support systems: a replication, re-examination, and extension of Swanson's work on literature-based discovery of a connection between Raynaud's and fish oil

Don R. Swanson has undertaken a program of research to use the published medical literature as a source of discoveries. We have attempted to replicate his discovery of a connection between Raynauds disease and dietary fish oil, as well as develop computer‐based searching methods that could usefully support literature‐based discoveries. We have been successful in replicating Swansons discovery and have developed a method of discovery support based on the complete text of MEDLINE records. From these, we compute statistics based both on the frequency of tokens within a literature and on the number of records containing various tokens. We discuss the use of these statistics, suggesting that token and record frequencies are good indicators of literatures profitably related to some source literature, and that relative record frequencies are useful in isolating literatures with the potential of containing a discovery.

Images and inference

It is frequently asked whether imagery differs in a fundamental way from other forms of knowledge representation, specifically the predicative forms employed in artificial intelligence programs. Frequently suggested distinctions are pictorial versus descriptional, and analog versus digital. This paper argues that these distinctions are not central in understanding the role of imagery in cognition, and moreover do not correctly capture the difference between visual perception and language. A distinction is proposed between the representation of images, on the one hand, and a calculus-plus-proof-procedure form of knowledge representation on the other. This distinction is not based upon differences in expressive power or form, but rather is based upon a distinction between how these two representations function, specifically how they are used to make inferences. On this view, an important functional role of imagery is to provide a non-proof-procedural method for inference, using a constraint satisfaction mechanism. Images, even the limited class of images here called diagrams, support inference in a way that is distinct from the way predicative representations support inference. This analysis offers an approach to solving the “frame problem” of cognitive science.

Empirical studies of the genetic algorithm with noncoding segments

The genetic algorithm (GA) is a problem-solving method that is modeled after the process of natural selection. We are interested in studying a specific aspect of the GA: the effect of noncoding segments on GA performance. Noncoding segments are segments of bits in an individual that provide no contribution, positive or negative, to the fitness of that individual. Previous research on noncoding segments suggests that including these structures in the GA may improve GA performance. Understanding when and why this improvement occurs will help us to use the GA to its full potential. In this article we discuss our hypotheses on noncoding segments and describe the results of our experiments. The experiments may be separated into two categories: testing our program on problems from previous related studies, and testing new hypotheses on the effect of noncoding segments.

A comparison of the fixed and floating building block representation in the genetic algorithm

This article compares the traditional, fixed problem representation style of a genetic algorithm (GA) with a new floating representation in which the building blocks of a problem are not fixed at specific locations on the individuals of the population. In addition, the effects of noncoding segments on both of these representations is studied. Noncoding segments are a computational model of noncoding deoxyribonucleic acid, and floating building blocks mimic the location independence of genes. The fact that these structures are prevalent in natural genetic systems suggests that they may provide some advantages to the evolutionary process. Our results show that there is a significant difference in how GAs solve a problem in the fixed and floating representations. Genetic algorithms are able to maintain a more diverse population with the floating representation. The combination of noncoding segments and floating building blocks appears to encourage a GA to take advantage of its parallel search and recombination abilities.

A Survey of Intron Research in Genetics

A brief survey of biological research on non-coding DNA is presented here. There has been growing interest in the effects of non-coding segments in evolutionary algorithms (EAs). To better understand and conduct research on non-coding segments and EAs, it is important to understand the biological background of such work. This paper begins with a review of basic genetics and terminology, describes the different types of non-coding DNA, and then surveys recent intron research.

USING DIAGRAMS TO UNDERSTAND GEOMETRY

This paper describes ARCHIMEDES‐STUDENT, a computer program that constructs and modifies its own representations of diagrams from instructions supplied by a human who is demonstrating a theorem of geometry. The programs representation permits it to make inferences from its constructions and to find a justification for the conclusion of the theorem. It is argued that the sort of perceptual reasoning displayed by this program represents one important aspect of understanding because it relates the abstract mathematical theorem to knowledge of spatial relations. For humans this approach grounds abstraction in experience and thus provides a more compelling demonstration than a formal proof. Because ARCHIMEDES‐STUDENT is a well‐defined computer program, it provides a precise suggestion of how this aspect of understanding can be achieved.

Studies on the effect of non-coding segments on the genetic algorithm

We study a specific aspect of the genetic algorithm (GA): the effect of non-coding segments on GA performance. Non-coding segments are segments of bits in an individual that provide no contribution, positive or negative, to the fitness of that individual. Previous research on non-coding segments suggests that including these structures in the GA population may improve GA performance. As a first step in our research, we tested our program on some of the same problems as the previous studies. This paper compares our results with the previous results and discusses the significance of the similarities and differences.<<ETX>>

DISTINGUISHING VISUAL IMAGERY AND SPATIAL REASONING

The imagery debate in its most recent revival inquires into the “psychological reality” of pictorial representations, asking whether images serve any cognitive function in humans, whether knowledge representation is analog or digital, whether images serve their functions in a stand-alone manner or require tacit knowledge, and so forth. However, these questions have not moved far enough away from the original phenomenological puzzle (how and why do I “see” things that are not there) because they are ill-specified and lack an encompassing theoretical context. For these reasons the debate has not been resolved by psychological experiment and never will be without more precise theories to test. Janice Glasgow joins those who have proposed that we address instead questions of how images can be represented and used by computers. The issue then becomes whether specific depictive representations, whatever form they might take, have computational advantages over alternatives such as predicate logic with deduction. The judgments must be based on comparisons of total systems, including the access processes, not merely the notation used. Later the psychologists can make of these results what they will. Of course, there has already been interest in artificial intelligence (AI) on several problems of spatial reasoning

A processor-building system for experimental programming languages

such as navigation. Nonetheless, Glasgow’s emphasis on the computational questions and imagery is a minority one. Classical AI, particularly knowledge representation and problem solving, has been almost exclusively wedded to search-based forms of deduction and descriptive representations. Since I have also been a member of the minority, I applaud Glasgow’s efforts, both her proselytizing and her research, and on the general issues she and I appear to be in substantial agreement. Here I will contrast my own research with hers to provide some perspective on our differences. Psychology and philosophy have placed emphasis on the percept-like nature of imagery, particularly visual imagery. However, the cognitive roles of most sensory features such as color, pitch, and saltiness are relatively limited in inferential richness and structure and are probably amenable to conventional methods of analysis. However, a consortium of sensory-perceptual abilities evolved to provide efficient ways to move about in space and deal with physical objects in real-time, and these abilities have come to include in man the means to make inferences, especially predictions, about non-current events because such abilities have clear adaptive advantages that are central to our success. My approach adopts the position that the cognitive role of imagery is primarily a result of the fact that visual imagery and, in less obvious ways, tactile, proprioceptive, and auditory imagery, involve the representation and use of the structure of space, time and mechanics. Thus, I have chosen to avoid problems of nonspatial sensory features, and instead study computational methods that use spatial information in reasoning about well-defined tasks, e.g., to understand the role of diagrammatic reasoning in addressing mathematical problems (Lindsay 1988; 1989; 1992). Similar problems have long been addressed in A1 by others, from Gelernter (1959) 10 Larkin and Simon (1987) and Novak and Bulko (1990), but without explicitly attempting to capture the structure of space in a uniform representational scheme. We-have clear mathematical descriptions both of space and of the mechanics of rigid bodies acted upon by forces. The most straightforward ways to represent space are based on these descriptions. One way of doing this is illustrated by the work of Kaufman (1991),