Peter W. Frey

Letter Recognition Using Holland-Style Adaptive Classifiers

Machine rule induction was examined on a difficult categorization problem by applying a Holland-style classifier system to a complex letter recognition task. A set of 20,000 unique letter images was generated by randomly distorting pixel images of the 26 uppercase letters from 20 different commercial fonts. The parent fonts represented a full range of character types including script, italic, serif, and Gothic. The features of each of the 20,000 characters were summarized in terms of 16 primitive numerical attributes. Our research focused on machine induction techniques for generating IF-THEN classifiers in which the IF part was a list of values for each of the 16 attributes and the THEN part was the correct category, i.e., one of the 26 letters of the alphabet. We examined the effects of different procedures for encoding attributes, deriving new rules, and apportioning credit among the rules. Binary and Gray-code attribute encodings that required exact matches for rule activation were compared with integer representations that employed fuzzy matching for rule activation. Random and genetic methods for rule creation were compared with instance-based generalization. The strength/specificity method for credit apportionment was compared with a procedure we call “accuracy/utility.”

Sensitivity and responsivity measures for discrimination learning

Abstract The present report considers alternative measures of sensitivity and response bias for the discrimination learning paradigm. The classical signal detection measures, d′ and β, were compared with their nonparametric equivalents, A′ and B″, with theoretical measures derived from threshold theory and with empirical measures derived from the ROC graph. Differential rabbit eyelid conditioning data from three experiments were analyzed with these measures, and the results of these analyses were used along with other information to determine which measures of sensitivity and response bias are most useful for the analysis of discrimination learning data.

Student Ratings of Instruction: Validation Research

Correlations between instructional ratings and exam performance at three universities indicated that three rating factors (student accomplishment, presentation clarity, and organization-planning) correlated highly with educational achievement. These rating factors were derived by a factor analysis of a 21-item questionnaire. Separate analyses of the relationship between instructional ratings and student characteristics indicated that the student’s grade point average and math aptitude score (SAT) did not systematically vary with his ratings. However, more senior students (number of terms in college) rated instructors more favorably than their less experienced classmates.

Recall memory for visually presented chess positions

A series of three experiments replicated and extended earlier research reported by Chase and Simon (1973), de Groot (1965), and Charness (Note 1). The first experiment demonstrated that the relationship between memory for chess positions and chess skill varies directly with the amount of chess-specific information in the stimulus display. The second experiment employed tachistoscopic displays to incrementally “build” tournament chess positions by meaningful or nonmeaningful chunks and demonstrated that meaningful piece groupings during presentation markedly enhance subsequent recall performance. The third experiment tested memory for one of two positions presented in immediate sequence and demonstrated that explanations based on a limited-capacity short-term memory (Chase & Simon, 1973) are not adequate for explaining performance on this memory task.

A two-dimensional analysis of student ratings of instruction.

Student ratings of instruction were analyzed in terms of two global factors. One factor, which includes items on advanced planning, presentation clarity, and increased student knowledge, was named “pedagogical skill.” The other factor taps information about class discussion, grading, and the availability of help and was named “rapport.” Ratings on the “skill” factor did not covary with class size or the leniency of the instructors grading but did correlate with a reasonable external criterion of student learning. Ratings of “rapport” correlated inversely with class size and directly with average class grade and showed only a weak relationship to the external criterion of student learning. The “skill” factor showed more interclass stability than the “rapport” factor. Previous research studies which have examined the reliability and validity of instructional ratings and their relationship to student grades and class size have reported inconsistent findings. These inconsistencies appear to result from an inappropriate unidimensional analysis of ratings which should be examined in terms of two of more separate attitude dimensions.

Extinction after aversive conditioning: An associative or nonassociative process?

Abstract A series of three eyelid-conditioning experiments with rabbits examined the traditional extinction procedure in which the CS is presented in the absence of US presentations. The first study employed a savings test to demonstrate that the rate of reacquisition is faster after a conventional extinction procedure than after a procedure in which the CS and US have been unpaired. The second experiment indicated that this readiness to respond in reacquisition was related to a carry-over effect from the initial acquisition trials. The third study demonstrated that acquisition and reacquisition are similar when the US is maintained in extinction but not when it is omitted. These results imply that procedures which retain the US during extinction are more effective than the conventional CS-alone extinction procedure in eliminating or inhibiting the associative connection between the CS and the US.

The eyeblink as a time-locked response: Implications for serial and second-order conditioning.

A series of experiments were conducted to determine what factors control responding to the first element of a two-element serial compound in rabbit eyelid conditioning. An examination of response topography indicated that the eye-blink CR is rigidly timed to occur when the US is expected. This response-system characteristic prevents the occurrence of a CR during the first element of a serial compound or during the second-order CS in second-order conditioning. The comparison of a serial gap procedure with conventional serial and trace conditioning procedures suggested that the associative strength of the first element of a serial CS is not strongly influenced by either a second-order conditioning process or by the variable-reinforcement principle.

Blocking in eyelid conditioning: Effect of changing the CS-US interval and introducing an intertriai stimulus

Kamin’s three-stage blocking paradigm was investigated in rabbit eyelid conditioning, Two manipulations were examined. A change in the CS-US interval from Stage 1 to Stage 2 did not attenuate blocking. The introduction of a salient stimulus during the intertriai interval in Stage 2 also failed to attenuate blocking. The first result is not consistent with Kamin’s interpretation of the blocking effect in terms of US surprisingness. The second resuit is inconsistent with a prediction based on the Rescorla-Wagner model.

An introduction to computer chess

Chess has been an intriguing problem for individuals interested in machine intelligence for many years. Claude Shannon, the American mathematician, first proposed a plan for computer chess in 1949 [81]. The literature on mechanical chess-playing prior to this time reveals that the early automatons were merely facades which concealed a skilled human player [50]. Shannon believed that chess was an ideal problem for experimentation with machine intelligence since the game is clearly defined in terms of allowed operations (the legal moves) and in the ultimate goal (mate). At the same time, chess is neither so simple as to be trivial nor so complex as to be impossible. Shannon felt that the development of a credible chess program would demonstrate that “mechanized thinking” was feasible.

Second-order conditioning: The importance of stimulus overlap on second-order trials

Second-order conditioning was examined using the rabbit eyeblink paradigm and the gerbil CER paradigm. Pavlov’s hypothesis that stimulus overlap on second-order trials produces conditioned inhibition and that nonoverlap leads to second-order conditioning was not confirmed. Our results also revealed that the manner in which first-order and second-order trials are intermixed has an important influence on the properties of the second-order CS. A within-session mixture of first- and second-order trials tended to produce second-order conditioning, and a between-session mixture tended to produce conditioned inhibition. Second-order conditioning was more prominent with the gerbil fear response than with the rabbit eyelid response.