Generalization gradients in human predictive learning: Effects of discrimination training and within-subjects testing

Generalization gradients have been investigated widely in animal conditioning experiments, but much less so in human predictive learning tasks. Here, we apply the experimental design of a recent study on conditioned fear generalization in humans (Lissek et al., 2008) to a predictive learning task, and examine the effects of a number of relevant procedural parameters drawn from the generalization literature in animal conditioning. Experiment 1 shows that prior discrimination learning and steady-state testing procedures sharpen the gradient; Experiment 2 shows that within-subjects testing of the range of generalization stimuli also sharpens the gradient. In addition, Experiment 2 shows that, in case of very flat initial generalization, an orderly gradient can reveal itself through differential rates of extinction learning. Finally, Experiment 2 also evidenced an orderly gradient of generalization-of-extinction. These results suggest that discrimination processes have an important effect on the generalization of predictive learning in humans, and highlight behavioral analogies between animal conditioning and human predictive learning.     

人类颜色视觉的计算理论

该文有机地结合了计算视觉理论和生态学视觉理论,指出颜色信息处理的根本任务是检测环境中的光不变量。在此基础上,作者提出了颜色视觉的计算理论以及计算理论本身的生物学标准。初级视觉计算是典型的不适定问题,动物的视觉系统则利用视环境中存在的条件将该不适定问题转化为定解问题。本文引入颜色视觉计算的免要条件,客观性约束,以及颜色认知的神经表象,证明了上述约束下颜色算法的存在性。本文给出了构造颜色知觉的基本假设。同时,该文还讨论了与上述问题密切相关的几个基本问题:神经表象的完备性,主观色觉的客观性,明度知觉和颜色知觉的统一,人类主观色觉的实现方式。     

Design and evaluation of a touchscreen apparatus for operant research with pigeons

We developed a touchscreen apparatus for pigeons and conducted a series of experiments that assessed its utility for free-operant procedures. The apparatus incorporated an on-board Windows computer, an electromechanical interface, an amplified speaker, and the touchscreen. We found that merely projecting a virtual key on the screen was insufficient; too many pecks missed the key. Adding a visual target in the center of the key and providing visual feedback for on-key pecks both failed to improve response accuracy. Accuracy was improved by imposing a timeout after off-key pecks or providing a physical boundary around the key. With the physical boundary, response accuracy was comparable to that obtained with conventional plastic keys, and response acquisition via autoshaping also was comparable. Mixing the color elements of the screen's pixels produced color stimuli, but the colors did not function as pure wavelengths of light in tests of stimulus generalization. Both colors and geometric shapes functioned as discriminative stimuli in multiple schedules with variable-interval and extinction components or rich and lean fixed-ratio components. In general, our touchscreen apparatus is a viable alternative to conventional pigeon chambers and increases the experimenter's options for visual stimuli, auditory stimuli, and the number and location of response keys.     

Temporal specificity of perceptual learning in an auditory discrimination task

Although temporal processing is used in a wide range of sensory and motor tasks, there is little evidence as to whether a single centralized clock or a distributed system underlies timing in the range of tens to hundreds of milliseconds. We investigated this question by studying whether learning on an auditory interval discrimination task generalizes across stimulus types, intervals, and frequencies. The degree to which improvements in timing carry over to different stimulus features constrains the neural mechanisms underlying timing. Human subjects trained on a 100- or 200-msec interval discrimination task showed an improvement in temporal resolution. This learning generalized to a perceptually distinct duration stimulus, as well as to the trained interval presented with tones at untrained spectral frequencies. The improvement in performance did not generalize to untrained intervals. To determine if spectral generalization was dependent on the importance of frequency information in the task, subjects were simultaneously trained on two different intervals identified by frequency. As a whole, our results indicate that the brain uses circuits that are dedicated to specific time spans, and that each circuit processes stimuli across nontemporal stimulus features. The patterns of generalization additionally indicate that temporal learning does not rely on changes in early, subcortical processing, because the nontemporal features are encoded by different channels at early stages.     

情绪学习促进无意识信息进入意识

采用突破连续闪烁抑制范式(b-CFS), 考察了经过情绪学习的刺激在无意识中是否存在加工优势。实验1使用不同朝向的光栅刺激, 结果发现情绪学习条件比对照条件的刺激在CFS中突破抑制时间更短。实验2考察了更复杂的朝向和颜色整合的光栅刺激, 发现情绪优势效应具有特异性, 仅局限于情绪学习的整合刺激。研究表明经过情绪学习后的刺激在无意识加工中存在优势。     

Words can slow down category learning

Words have been shown to influence many cognitive tasks, including category learning. Most demonstrations of these effects have focused on instances in which words facilitate performance. One possibility is that words augment representations, predicting an across the-board benefit of words during category learning. We propose that words shift attention to dimensions that have been historically predictive in similar contexts. Under this account, there should be cases in which words are detrimental to performance. The results from two experiments show that words impair learning of object categories under some conditions. Experiment 1 shows that words hurt performance when learning to categorize by texture. Experiment 2 shows that words also hurt when learning to categorize by brightness, leading to selectively attending to shape when both shape and hue could be used to correctly categorize stimuli. We suggest that both the positive and negative effects of words have developmental origins in the history of word usage while learning categories. [corrected]     

The use of faces as stimuli in neuroimaging and psychological experiments: A procedure to standardize stimulus features

In psychological experiments involving facial stimuli, it is of great importance that the basic perceptual or psychological characteristics that are investigated are not confounded by factors such as brightness and contrast, head size, hair cut and color, skin color, and the presence of glasses and earrings. Standardization of facial stimulus materials reduces the effect of these confounding factors. We therefore employed a set of basic image processing techniques to deal with this issue. The processed images depict the faces in gray-scale, all at the same size, brightness, and contrast, and confined to an oval mask revealing only the basic features such as the eyes, nose, and mouth. The standardization was successfully applied to four different face databases, consisting of male and female faces and including neutral as well as happy facial expressions. An important advantage of the proposed standardization is that featural as well as configurational information is retained. We also consider the procedure to be a major contribution to the development of a de facto standard for the use of facial stimuli in psychological experiments. Such methodological standardization would allow a better comparison of the results of these studies.     

Perceiving, imaging, and preferring physiognomic stimuli

Physiognomic color responses in perception, imagery, and affect were investigated. Maluma and taketa, nonsense stimuli defined by many investigators as physiognomic, were utilized as prototypical physiognomic stimuli, along with eight other stimuli of various sorts. In Experiment 1, 22 subjects matched the colors of the stimuli; in Experiment 2, 27 subjects reported their imagery to the stimuli; and in Experiment 3, 16 subjects gave their color preferences for the stimuli. The Munsell sets of colors were employed throughout. Significant differences between the physiognomic and other stimuli were found on the brightness and saturation of color matches, images, and preferences. Other differences (e.g., the latency of color images) were also present. Distinctions were also noted between the two physiognomic stimuli. These results support the priority of innate and perceptual processes in physiognomy over those of learning and memory, although some ambiguities still remain.     

Automaticity and preattentive processing.

The characteristics of automatized performance resemble those of preattentive processing in some respects. In the context of visual search tasks, these include spatially parallel processing, involuntary calling of attention, learning without awareness, and time-sharing with other tasks. However, this article reports some evidence suggesting that extended practice produces its effects through different mechanisms from those that underlie preattentive processing. The dramatic changes in search rate seem to depend not on the formation of new preattentive detectors for the task-relevant stimuli, nor on learned abstracted procedures for responding quickly and efficiently, but rather on changes that are very specific both to the particular stimuli and to the particular task used in practice. We suggest that the improved performance may depend on the accumulation of separate memory traces for each individual experience of a display (see Logan, 1988), and we show that the traces differ for conjunction search in which stimuli must be individuated and for feature search where a global response to the display is sufficient.     

The cognitive implications of asymmetric color generalization in honeybees

Generalization occurs when a conditioned response formed to one stimulus is also elicited by other stimuli which have not been used in the course of conditioning. Here, we studied color generalization in honeybees Apis mellifera trained to two rewarded colors, S₁+ and S₂+. After training, bees were tested with non-rewarded novel stimuli, which lay between the trained stimuli in a honeybee color space (Int) or outside the range defined by the trained stimuli (E ₁ and E ₂). We analyzed whether bees interpolated their choice to Int and/or extrapolated it to E ₁ and E ₂. We compared the performances of the group trained with S₁+ and S₂+ to those of control groups trained only with S₁+ or S₂+. Bees trained with S₁+ and S₂+ responded similarly and highly to all test stimuli. These results do not allow discerning between generalization models based on the presence of interpolation and/or extrapolation. Nevertheless, bee’s performance was consistent with a linear summation of the two generalization gradients generated by S₁+ and S₂+, respectively. These gradients were asymmetric because control bees responded to the test stimuli as if these belonged to different similarity classes in spite of the fact that they had similar perceptual distances separating them. Stimuli treated as similar were located in the same half of the color spaces, whereas stimuli treated as different were located in opposite halves. Our results suggest that color categories could exist in honeybees and may underlie the performance of the control groups. Under this assumption, color categories would be also present in simpler nervous systems, and would not require factors such as language to be expressed.     

Kindling and its consequences on learning in rats.

To study the learning performance of pentylenetetrazol- and amygdala-kindled Wistar rats we used the following learning tests: short-term memory was tested in the response-to-change model, brightness discrimination was tested in a Y-chamber, and two-way active avoidance learning was tested in a shuttle-box. Short-term memory was not impaired by both kindling procedures. Considering two-way active avoidance learning the performance of pentylenetetrazol (PTZ)-kindled rats was significantly diminished. This effect persists over a period of 4 weeks. However, amygdala (AMY)-kindled rats acquired this task like the controls. In brightness discrimination reaction (BDR) the learning performance of PTZ-kindled animals was not influenced. Although the acquisition of BDR was nearly identical, the 24-h retention was remarkably diminished in AMY-kindled rats. It was hypothesized that the different kindling procedures interfere in different ways and extent with neuronal circuits resulting in different functional impairments.     

On the generality and limits of abstraction in rats and humans

In this review, we address the question, central to cognition, of whether nonhuman animals such as rats are capable of extracting and extending information from a given learning situation to a new learning situation without generalizing through a physical dimension of the stimuli. This capacity underlies abstraction, which is a hallmark of human cognition and necessary for complex information processing such as language acquisition. We selectively review recent experiments with rats in which systematic changes in information processing of new stimuli are observed after training with different stimuli. These results strongly suggest that this capacity is present in rats. We also review two articles in which clear limitations to this capacity are detected. We conclude that, within specified limits, rats are capable of using prior experience when faced with a learning situation that involves new stimuli. We interpret this ability as a rudimentary form of abstraction. In the face of these provocative results, new theories of learning should be designed to account for these findings.     

Secondary generalization and categorization in pigeons

In Experiment 1, one group of pigeons learned to classify a set of stimuli into the human language classes cat, flower, car, and chair (categorization); another group learned to classify the same set into arbitrary classes (pseudocategorization). Then, both groups were trained on a new categorization task and their performance compared to that of a control group that had no initial classification training. Hull's (1943) notion of secondary generalization (generalization that is not based on physical similarity but on mediating associations) predicts that categorization experience will facilitate the learning of a new categorization task, whereas pseudocategorization experience will impair it. However, in Experiment 1, performance on the new categorization task was not differently affected by prior experience. In Experiment 2, pigeons initially trained to classify a set of 48 stimuli (original training) were later trained to classify a subset of four of these stimuli using new responses (reassignment training). Then, they were tested on the 44 remaining stimuli. Performance better accorded with original than with reassignment training, indicating that categorization training did not lead to the formation of equivalence classes of stimuli, in which the equivalence relationship is mediated by secondary generalization. The lack of evidence of secondary generalization implies that our pigeons failed to meet Lea's (1984) criterion for conceptual behavior.     

Brightness contrast: a reinterpretation of compound cue and combined cue experiments with pigeons.

A group of three pigeons was trained on a 4-ply multiple schedule: a green color and a vertical line superimposed upon an achromatic background as positive stimuli, and a red color and a horizontal line on an achromatic background as negative stimuli. The pigeons were tested with the vertical line superimposed upon different achromatic background intensities, then with the vertical line superimposed upon different green background intensities, and finally with the vertical line and its training achromatic backgfound attenuated (and unattenuated) by a neutral density filter. The gradients peaked at the luminance of the achromatic background used during training and at the equivalent luminance for the green background when it was substituted for the achromatic background. The brightness contrast, not the background luminance, was the critical variable as the neutral density filter attenuated both the line and the background equally, leaving brightness contrast unchanged; there was no response decrement to this attenuated stimulus. Two other groups of three pigeons showed that they attended to line orientation as well as to brightness contrast. The brightness contrast hypothesis was extended to explain results of attention experiments and combined cue experiments which have used line stimuli in combinations with different backgrounds.     

A neural model of hippocampal–striatal interactions in associative learning and transfer generalization in various neurological and psychiatric patients

Building on our previous neurocomputational models of basal ganglia and hippocampal region function (and their modulation by dopamine and acetylcholine, respectively), we show here how an integration of these models can inform our understanding of the interaction between the basal ganglia and hippocampal region in associative learning and transfer generalization across various patient populations. As a common test bed for exploring interactions between these brain regions and neuromodulators, we focus on the acquired equivalence task, an associative learning paradigm in which stimuli that have been associated with the same outcome acquire a functional similarity such that subsequent generalization between these stimuli increases. This task has been used to test cognitive dysfunction in various patient populations with damages to the hippocampal region and basal ganglia, including studies of patients with Parkinson’s disease (PD), schizophrenia, basal forebrain amnesia, and hippocampal atrophy. Simulation results show that damage to the hippocampal region—as in patients with hippocampal atrophy (HA), hypoxia, mild Alzheimer’s (AD), or schizophrenia—leads to intact associative learning but impaired transfer generalization performance. Moreover, the model demonstrates how PD and anterior communicating artery (ACoA) aneurysm—two very different brain disorders that affect different neural mechanisms—can have similar effects on acquired equivalence performance. In particular, the model shows that simulating a loss of dopamine function in the basal ganglia module (as in PD) leads to slow acquisition learning but intact transfer generalization. Similarly, the model shows that simulating the loss of acetylcholine in the hippocampal region (as in ACoA aneurysm) also results in slower acquisition learning. We argue from this that changes in associative learning of stimulus–action pathways (in the basal ganglia) or changes in the learning of stimulus representations (in the hippocampal region) can have similar functional effects.     

Differential brain responses when applying criterion attribute versus family resemblance rule learning

Subsystems of category learning have been identified on the basis of general domains of content (e.g., tools, faces). The present study examined categories from the standpoint of internal structure and determined brain topography associated with expressing two fundamentally different category rule structures (criterion attribute, CA, and family resemblance, FR). CA category learning involves processing stimuli by isolated features and classifying by properties held by all members. FR learning involves processing stimuli by integral wholes and classifying on overall similarity among members without sharing identical features. fMRI BOLD response to CA and FR categorization was measured with pseudowords as stimuli. Category knowledge for both tasks was mastered prior to brain imaging. Areas of activation emerged unique to the structure of each category and followed from the nature of the rule abstraction procedure. CA categorization was implemented by strong target monitoring and expectation (medial parietal), rule maintenance in working memory, feature selection processes (inferior frontal), and a sensitivity to high frequency components of the stimulus such as isolated features (anterior temporal). FR categorization, consistent with its multi-featural nature, involved word-level processing (left extrastriate) that evoked articulatory rehearsal (medial cerebellar). The data suggest category structure is an important determinant of brain response during categorization. For instance, anterior temporal structures may help attune visual processing systems to high frequency components to support the learning of criterial, highly predictive rules.     

Generalization peak shift for autoshaped and operant key pecks.

Pigeons acquired discriminated key pecking between 528- and 540-nm stimuli by either a response-reinforcer (operant group) or a stimulus-reinforcer (autoshaped group) contingency, with other training-schedule parameters comparable over groups. For the birds in the operant group, key pecks intermittently produced grain in the presence of one hue on the key (positive stimulus) but not in the other (negative stimulus). For the birds in the autoshaped group, pecking emerged when grain was intermittently presented independently of key pecking during one key color but was not presented during the other key color. Two independent contingency assays, peck-location comparisons and elimination of differences in reinforcement rate, confirmed the effectiveness of the two training procedures in establishing operant or respondent control of key pecking. After reaching a 10:1, or better, discrimination ratio between key pecks during the two key colors, the birds received a wavelength generalization test. Criterion baseline key-peck rates were comparable for operant and autoshaped groups prior to testing. On the generalization test, performed in extinction, all birds pecked most at a stimulus removed from the positive training stimulus in the direction away from the negative stimulus. In testing, autoshaped "peak" rates (24.5 to 64.9 pecks per minute) were from 33% to 80% higher than rates in the presence of the training stimuli. Respondent peak shift rarely has been reported heretofore, and never this consistently and robustly. These results further confirm the similarity of perceptual processing in classical and operant learning. They are discussed in terms of Spence's gradient-interaction theory and Weiss' (1978) two-process model of stimulus control.     

DOES CONDITIONAL DISCRIMINATION LEARNING BY PIGEONS NECESSARILY INVOLVE HIERARCHICAL RELATIONSHIPS?

Four experiments demonstrate that when putative conditional and discriminative cues are presented simultaneously in the single reversal procedure, it is not possible to ascribe a uniquely conditional or uniquely discriminative function to either of the cues. In Experiment 1, pigeons were trained to respond to a blue key and not to a red key while the houselight was on; then in a different session they learned the reversal of this discrimination with the houselight off (single reversal). Separate groups were tested for color generalization with houselight conditions alternating in blocks of trials or for houselight intensity generalization with blue and red key colors alternating in blocks of trials. Both test procedures revealed a conditional relationship between houselight and key color conditions. Experiment 2 produced the same result following training in which the key colors were held constant across training sessions while the houselight and no houselight conditions varied within sessions. In Experiment 3, separate groups were trained with the two procedures but were tested with randomly ordered combinations of key colors and houselight intensities. The two groups yielded indistinguishable bidimensional generalization gradients with peaks at both previously reinforced stimulus combinations. In Experiment 4 the subjects were switched from one of these training procedures to the other with no decrement in their discriminative performance. We conclude that for successive discriminations between conditional- and discriminative-stimulus combinations, the notion of a hierarchical relation between conditional and discriminative stimuli must be extended to include a symmetrical relationship or the notion should be abandoned altogether.     

Appetitive and aversive olfactory learning induce similar generalization rates in the honey bee

Appetitive and aversive learning drive an animal toward or away from stimuli predicting reinforcement, respectively. The specificity of these memories may vary due to differences in cost–benefit relationships associated with appetitive and aversive contexts. As a consequence, generalization performances may differ after appetitive and aversive training. Here, we determined whether honey bees show different rates of olfactory generalization following appetitive olfactory conditioning of the proboscis extension response, or aversive olfactory conditioning of the sting extension response. In both cases, we performed differential conditioning, which improves discrimination learning between a reinforced odor (CS+) and a non-reinforced odor (CS?) and evaluated generalization to two novel odors whose similarity to the CS+ and the CS? was different. We show, given the same level of discriminatory performance, that rates of generalization are similar between the two conditioning protocols and discuss the possible causes for this phenomenon.     

Learning efficient visual search for stimuli containing diagnostic spatial configurations and color-shape conjunctions

Visual search is often slow and difficult for complex stimuli such as feature conjunctions. Search efficiency, however, can improve with training. Search for stimuli that can be identified by the spatial configuration of two elements (e.g., the relative position of two colored shapes) improves dramatically within a few hundred trials of practice. Several recent imaging studies have identified neural correlates of this learning, but it remains unclear what stimulus properties participants learn to use to search efficiently. Influential models, such as reverse hierarchy theory, propose two major possibilities: learning to use information contained in low-level image statistics (e.g., single features at particular retinotopic locations) or in high-level characteristics (e.g., feature conjunctions) of the task-relevant stimuli. In a series of experiments, we tested these two hypotheses, which make different predictions about the effect of various stimulus manipulations after training. We find relatively small effects of manipulating low-level properties of the stimuli (e.g., changing their retinotopic location) and some conjunctive properties (e.g., color-position), whereas the effects of manipulating other conjunctive properties (e.g., color-shape) are larger. Overall, the findings suggest conjunction learning involving such stimuli might be an emergent phenomenon that reflects multiple different learning processes, each of which capitalizes on different types of information contained in the stimuli. We also show that both targets and distractors are learned, and that reversing learned target and distractor identities impairs performance. This suggests that participants do not merely learn to discriminate target and distractor stimuli, they also learn stimulus identity mappings that contribute to performance improvements.