Ramachandran and Hirstein offer a number of interesting ideas about aesthetic preference. In this commentary I shall focus mainly on their ideas concerning peak shift and prototypicality. The authors give the example of a rat rewarded for responding to a rectangle and not rewarded for responding to a longer triangle . They argue that the rat will respond even more to a more elongated rectangle. In fact, two phenomena are involved here. Peak shift refers to the fact that the rat (...) responds more to a supernormal stimulus . Behavioural contrast refers to the fact that the rat responds more to the S++ than it did to the S+ before discrimination training took place. It does not detract from the authors’ main point, but we would be unlikely to find peak shift in the hypothetical experiment described. Peak shift only occurs when the S+ and the S- are very similar to each other . Let us say that we train a rat to respond to a 1000 Hz tone and not to respond to a 500 Hz tone. No peak shift will be found. We will find peak shift if the S+ is 1000 Hz and the S- is 950 Hz. The conventional explanation for this is that the animal shows generalization. A rat trained simply to respond to a 1000 Hz tone will also respond a bit less to one of 990 Hz, even less to one of 980 Hz, and so on. In discrimination training, our rat would have a strong tendency to respond to a 1000 Hz tone but, because of generalization from the 950 Hz tone, it would also have some tendency not to respond to it. If we subtract the inhibitory gradient around 950 Hz from the excitatory gradient around 1000 Hz, we will end up with maximal responding around, say, 1010 Hz . Intuitively, we would expect the rat to respond more to the S++ but also less vigorously than it had to the S+ before discrimination training was begun. This is because the inhibitory gradient has been subtracted from the excitatory gradient. Surprisingly, we find behavioural contrast; the rat responds more vigorously to the S++ than it had to the S+. Grossberg argues that this is because there is a tendency to keep amount of activation on a layer of cortex constant. (shrink)

In this paper, I distinguish between two types of normative accounts of discrimination – general and special – and argue for the former and against the latter. General accounts consider the moral status of discrimination in light of all of the reasons that apply to discrimination, and hold that these reasons are not unique to discrimination (for example, the reasons to bring about the greater benefit or prevent the greater burden, to give priority for people who (...) are worse off, and to give people what they morally deserve). In contrast, special accounts argue that discrimination is objectionable due to factors that are especially salient in the context of discrimination (examples are the "deliberative" freedom of people to decide how to live without considering facts such as their race or sex, the message conveyed by discriminatory actions in terms of the moral status of people, and the mental states that accompany discriminatory actions). I argue that general accounts are more plausible than special ones, as foundational accounts of discrimination that determine its overall moral status (both types of accounts appear to suggest such accounts). One argument is that special accounts that suggest conclusions regarding the overall moral status of discrimination without considering all of the pertinent factors, especially general ones, are implausible. Another argument is that special accounts that claim that the factors that they highlight are basic are misguided since these factors are morally significant only if, and to the degree to which, they are derived from general ones. (shrink)

Anxiety patients over-generalize fear, possibly because of an incapacity to discriminate threat and safety signals. Discrimination trainings are promising approaches for reducing such fear over-generalization. Here we investigated the efficacy of a fear-relevant vs. a fear-irrelevant discrimination training on fear generalization and whether the effects are increased with feedback during training. Eighty participants underwent two fear acquisition blocks, during which one face, but not another face, was associated with a female scream. During two generalization blocks, (...) both CSs plus four morphs were presented. Between these generalization blocks, half of the participants underwent a fear-relevant discrimination training with or without feedback and the other half a fear-irrelevant discrimination training with or without feedback. US expectancy, arousal, valence ratings, and skin conductance responses indicated successful fear acquisition. Importantly, fear-relevant vs. fear-irrelevant discrimination trainings and feedback vs. no feedback reduced generalization as reflected in US expectancy ratings independently from one another. No effects of training condition were found for arousal and valence ratings or SCR. In summary, this is a first indication that fear-relevant discrimination training and feedback can improve the discrimination between threat and safety signals in healthy individuals, at least for learning-related evaluations, but not evaluations of valence or arousal. (shrink)

In assigning the benefits and burdens of society, we sometimes discriminate using a broad category (age, gender, race, etc.) we think correlates well with the possession of some other skill, qualification, or character trait. In this essay, I explore one rationale for this type of discrimination. I suggest a method for determining when this rationale provides a moral justification for the discrimination, and when it does not. I defend the method against some potential criticisms, and point out some (...) exceptions to its application. (shrink)

Zero-shot learning is dedicated to solving the classification problem of unseen categories, while generalized zero-shot learning aims to classify the samples selected from both seen classes and unseen classes, in which “seen” and “unseen” classes indicate whether they can be used in the training process, and if so, they indicate seen classes, and vice versa. Nowadays, with the promotion of deep learning technology, the performance of zero-shot learning has been greatly improved. Generalized zero-shot learning is a challenging topic that has (...) promising prospects in many realistic scenarios. Although the zero-shot learning task has made gratifying progress, there is still a strong deviation between seen classes and unseen classes in the existing methods. Recent methods focus on learning a unified semantic-aligned visual representation to transfer knowledge between two domains, while ignoring the intrinsic characteristics of visual features which are discriminative enough to be classified by itself. To solve the above problems, we propose a novel model that uses the discriminative information of visual features to optimize the generative module, in which the generative module is a dual generation network framework composed of conditional VAE and improved WGAN. Specifically, the model uses the discrimination information of visual features, according to the relevant semantic embedding, synthesizes the visual features of unseen categories by using the learned generator, and then trains the final softmax classifier by using the generated visual features, thus realizing the recognition of unseen categories. In addition, this paper also analyzes the effect of the additional classifiers with different structures on the transmission of discriminative information. We have conducted a lot of experiments on six commonly used benchmark datasets. The experimental results show that our model outperforms several state-of-the-art methods for both traditional as well as generalized zero-shot learning. (shrink)