With the assistance of sophisticated training methods applied to single labeled datasets, the performance of fully-supervised person re-identification (Person Re-ID) has been improved significantly ...in recent years. However, these models trained on a single dataset usually suffer from considerable performance degradation when applied to videos of a different camera network. To make Person Re-ID systems more practical and scalable, several cross-dataset domain adaptation methods have been proposed, which achieve high performance without the labeled data from the target domain. However, these approaches still require the unlabeled data of the target domain during the training process, making them impractical. A practical Person Re-ID system pre-trained on other datasets should start running immediately after deployment on a new site without having to wait until sufficient images or videos are collected and the pre-trained model is tuned. To serve this purpose, in this paper, we reformulate person re-identification as a multi-dataset domain generalization problem. We propose a multi-dataset feature generalization network (MMFA-AAE), which is capable of learning a universal domain-invariant feature representation from multiple labeled datasets and generalizing it to `unseen' camera systems. The network is based on an adversarial auto-encoder to learn a generalized domain-invariant latent feature representation with the Maximum Mean Discrepancy (MMD) measure to align the distributions across multiple domains. Extensive experiments demonstrate the effectiveness of the proposed method. Our MMFA-AAE approach not only outperforms most of the domain generalization Person Re-ID methods, but also surpasses many state-of-the-art supervised methods and unsupervised domain adaptation methods by a large margin.
Constructivist views of cognitive development often converge on 2 key points: (a) the child's goal is to build large conceptual structures for understanding the world, and (b) the child plays an ...active role in developing these structures. While previous research has demonstrated that young children show a precocious capacity for concept and theory building when they are provided with helpful data within training settings, and that they explore their environment in ways that may promote learning, it remains an open question whether young children are able to build larger conceptual structures using self-generated evidence, a form of active learning. In the current study, we examined whether children can learn high-order generalizations (which form the basis for larger conceptual structures) through free play, and whether they can do so as effectively as when provided with relevant data. Results with 2- and 3-year-old children over 4 experiments indicate robust learning through free play, and generalization performance was comparable between free play and didactic conditions. Therefore, young children's self-directed learning supports the development of higher-order generalizations, laying the foundation for building larger conceptual structures and intuitive theories.
In the field of stimulus generalization, an old yet unresolved discussion pertains to what extent stimulus misidentifications contribute to the pattern of conditioned responding. In this article, we ...perform cluster analysis on six datasets (four published datasets and two unpublished datasets, included N = 950) to examine the relationship between interindividual differences in (a) stimulus identification, (b) patterns of generalized responding, and (c) verbalized generalization rules. The datasets were obtained from online predictive learning tasks where participants learned associations between colored cues and the presence or absence of a hypothetical outcome. In these datasets, stimulus identification and expectancy ratings were assessed in separate phases to a range of colors varying between blue-green. Using cluster analyses on performance during stimulus identification, we identified different subgroups of participants (good vs. bad identifiers). In all six datasets, we found a close relationship between the pattern of stimulus identification and the shape of the expectancy gradient across the test dimension between the identified subgroups. Furthermore, participants classified as good identifiers were more likely to report a similarity generalization rule than a relational or linear rule, suggesting that individual differences in stimulus identification are related to individual differences in generalization rules. These findings suggest that greater consideration should be given to interindividual variability in stimulus identification, inductive rules, and their relationship in explaining patterns of generalized responses.
Successful speech perception requires that listeners map the acoustic signal to linguistic categories. These mappings are not only probabilistic, but change depending on the situation. For example, ...one talker's /p/ might be physically indistinguishable from another talker's /b/ (cf. lack of invariance). We characterize the computational problem posed by such a subjectively nonstationary world and propose that the speech perception system overcomes this challenge by (a) recognizing previously encountered situations, (b) generalizing to other situations based on previous similar experience, and (c) adapting to novel situations. We formalize this proposal in the ideal adapter framework: (a) to (c) can be understood as inference under uncertainty about the appropriate generative model for the current talker, thereby facilitating robust speech perception despite the lack of invariance. We focus on 2 critical aspects of the ideal adapter. First, in situations that clearly deviate from previous experience, listeners need to adapt. We develop a distributional (belief-updating) learning model of incremental adaptation. The model provides a good fit against known and novel phonetic adaptation data, including perceptual recalibration and selective adaptation. Second, robust speech recognition requires that listeners learn to represent the structured component of cross-situation variability in the speech signal. We discuss how these 2 aspects of the ideal adapter provide a unifying explanation for adaptation, talker-specificity, and generalization across talkers and groups of talkers (e.g., accents and dialects). The ideal adapter provides a guiding framework for future investigations into speech perception and adaptation, and more broadly language comprehension.
A major question for the study of learning and memory is how to tailor learning experiences to promote knowledge that generalizes to new situations. In two experiments, we used category learning as a ...representative domain to test two factors thought to influence the acquisition of conceptual knowledge: the number of training examples (set size) and the similarity of training examples to the category average (set coherence). Across participants, size and coherence of category training sets were varied in a fully crossed design. After training, participants demonstrated the breadth of their category knowledge by categorizing novel examples varying in their distance from the category center. Results showed better generalization following more coherent training sets, even when categorizing items furthest from the category center. Training set size had limited effects on performance. We also tested the types of representations underlying categorization decisions by fitting formal prototype and exemplar models. Prototype models posit abstract category representations based on the category's central tendency, whereas exemplar models posit that categories are represented by individual category members. In Experiment 1, low coherence training led to fewer participants relying on prototype representations, except when training length was extended. In Experiment 2, low coherence training led to chance performance and no clear representational strategy for nearly half of the participants. The results indicate that highlighting commonalities among exemplars during training facilitates learning and generalization and may also affect the types of concept representations that individuals form.
Machine learning systems generally assume that the training and testing distributions are the same. To this end, a key requirement is to develop models that can generalize to unseen distributions. ...Domain generalization (DG), i.e., out-of-distribution generalization, has attracted increasing interests in recent years. Domain generalization deals with a challenging setting where one or several different but related domain(s) are given, and the goal is to learn a model that can generalize to an unseen test domain. Great progress has been made in the area of domain generalization for years. This paper presents the first review of recent advances in this area. First, we provide a formal definition of domain generalization and discuss several related fields. We then thoroughly review the theories related to domain generalization and carefully analyze the theory behind generalization. We categorize recent algorithms into three classes: data manipulation, representation learning, and learning strategy, and present several popular algorithms in detail for each category. Third, we introduce the commonly used datasets, applications, and our open-sourced codebase for fair evaluation. Finally, we summarize existing literature and present some potential research topics for the future.
Humans spontaneously organize a continuous experience into discrete events and use the learned structure of these events to generalize and organize memory. We introduce the Structured Event Memory ...(SEM) model of event cognition, which accounts for human abilities in event segmentation, memory, and generalization. SEM is derived from a probabilistic generative model of event dynamics defined over structured symbolic scenes. By embedding symbolic scene representations in a vector space and parametrizing the scene dynamics in this continuous space, SEM combines the advantages of structured and neural network approaches to high-level cognition. Using probabilistic reasoning over this generative model, SEM can infer event boundaries, learn event schemata, and use event knowledge to reconstruct past experience. We show that SEM can scale up to high-dimensional input spaces, producing human-like event segmentation for naturalistic video data, and accounts for a wide array of memory phenomena.
We introduce the Category Abstraction Learning (CAL) model, a cognitive framework formally describing category learning built on similarity-based generalization, dissimilarity-based abstraction, two ...attention learning mechanisms, error-driven knowledge structuring, and stimulus memorization. Our hypotheses draw on an array of empirical and theoretical insights connecting reinforcement and category learning. The key novelty of the model is its explanation of how rules are learned from scratch based on three central assumptions. (a) Category rules emerge from two processes of stimulus generalization (similarity) and its direct inverse (category contrast) on independent dimensions. (b) Two attention mechanisms guide learning by focusing on rules, or on the contexts in which they produce errors. (c) Knowing about these contexts inhibits executing the rule, without correcting it, and consequently leads to applying partial rules in different situations. The model is designed to capture both systematic and individual differences in a broad range of learning paradigms. We illustrate the model's explanatory scope by simulating several benchmarks, including the classic Six Problems, the 5-4 problem, and linear separability. Beyond the common approach of predicting average response probabilities, we also propose explanations for more recently studied phenomena that challenge existing learning accounts, regarding task instructions, individual differences in rule extrapolation in three different tasks, individual attention shifts to stimulus features during learning, and other phenomena. We discuss CAL's relation to different models, and its potential to measure the cognitive processes regarding attention, abstraction, error detection, and memorization from multiple psychological perspectives.
PDF
