Recently, infrared human action recognition has attracted increasing attention for it has many advantages over visible light, that is, being robust to illumination change and shadows. However, the infrared action data is limited until now, which degrades the performance of infrared action recognition. Motivated by the idea of transfer learning, an infrared human action recognition framework using auxiliary data from visible light is proposed to solve the problem of limited infrared action data. In the proposed framework, we first construct a (...) novel Cross-Dataset Feature Alignment and Generalization framework to map the infrared data and visible light data into a common feature space, where Kernel Manifold Alignment and a dual aligned-to-generalized encoders model are employed to represent the feature. Then, a support vector machine is trained, using both the infrared data and visible light data, and can classify the features derived from infrared data. The proposed method is evaluated on InfAR, which is a publicly available infrared human action dataset. To build up auxiliary data, we set up a novel visible light action dataset XD145. Experimental results show that the proposed method can achieve state-of-the-art performance compared with several transfer learning and domain adaptation methods. (shrink)

Whilst the topic of representations is one of the key topics in philosophy of mind, it has only occasionally been noted that representations and representational features may be gradual. Apart from vague allusions, little has been said on what representational gradation amounts to and why it could be explanatorily useful. The aim of this paper is to provide a novel take on gradation of representational features within the neuroscientific framework of predictive processing. More specifically, we provide a gradual account of (...) two features of structural representations: structural similarity and decoupling. We argue that structural similarity can be analysed in terms of two dimensions: number of preserved relations and state space granularity. Both dimensions can take on different values and hence render structural similarity gradual. We further argue that decoupling is gradual in two ways. First, we show that different brain areas are involved in decoupled cognitive processes to a greater or lesser degree depending on the cause (internal or external) of their activity. Second, and more importantly, we show that the degree of decoupling can be further regulated in some brain areas through precision weighting of prediction error. We lastly argue that gradation of decoupling (via precision weighting) and gradation of structural similarity (via state space granularity) are conducive to behavioural success. (shrink)

Acoustic speech output results from coordinated articulation of dozens of muscles, bones and cartilages of the vocal mechanism. While we commonly take the fluency and speed of our speech productions for granted, the neural mechanisms facilitating the requisite muscular control are not completely understood. Previous neuroimaging and electrophysiology studies of speech sensorimotor control has typically concentrated on speech sounds (i.e., phonemes, syllables and words) in isolation; sentence-length investigations have largely been used to inform coincident linguistic processing. In this study, we (...) examined the neural representations of segmental features (place and manner of articulation, and voicing status) in the context of fluent, continuous speech production. We used recordings from the cortical surface (electrocorticography (ECoG)) to simultaneously evaluate the spatial topography and temporal dynamics of the neural correlates of speech articulation that may mediate the generation of hypothesized gestural or articulatory scores. We found that the representation of place of articulation involved broad networks of brain regions during all phases of speech production: preparation, execution and monitoring. In contrast, manner of articulation and voicing status were dominated by auditory cortical responses after speech had been initiated. These results provide a new insight into the articulatory and auditory processes underlying speech production in terms of their motor requirements and acoustic correlates. (shrink)

Embodied cognition theory posits that concept representations inherently rely on sensorimotor experiences that accompany their acquisitions. This is well established through concrete concepts. However, it is debatable whether representations of abstract concepts are based on sensorimotor representations. This study investigated the causal role of associated motor experiences that accompany concept acquisition in the involvement of the motor system in the abstract verb processing. Through two experiments, we examined the action–sentence compatibility effect, in the test phase after an increase in motor (...) features during the learning phase for abstract verbs with low motor features or novel words with no conceptual features at all. After associated motor experiences were added in the word learning phase, action–sentence compatibility effect was found in the semantic processing tasks during the test phase for abstract verbs and novel words. This was lacking in the word font color judgment task requiring no semantic processing. Coupled with our previous study, these findings suggest that motor features formed during word learning could causally affect embodiment in the motor system for abstract verbs, and reactivation of motor experiences in abstract verb processing depends on a given task’s demands. Our study supports the view that conceptual representations, even abstract concepts, can be grounded in sensorimotor experiences. (shrink)

Episodic memories are shaped by the representational format of their contents. These formats are not only determined by medial temporal lobe areas, but essentially also by the neocortical regions which these areas control. The representational formats of medial temporal lobe and neocortex are sufficient to determine both, memory contents and subjective memory qualities, without the further need for an attribution system.

Many studies have observed modulation of the amplitude of the neural index mismatch negativity (MMN) related to which member of a phoneme contrast [phoneme A, phoneme B] serves as the frequent (standard) and which serves as the infrequent (deviant) stimulus (i.e., AAAB vs. BBBA) in an oddball paradigm. Explanations for this amplitude modulation range from acoustic to linguistic factors. We tested whether exchanging the role of the mid vowel /ε/ vs. high vowel /ɪ/ of English modulated MMN amplitude and whether (...) the pattern of modulation was compatible with an underspecification account, in which the underspecified height values are [−high] and [−low]. MMN was larger for /ε/ as the deviant, but only when compared across conditions to itself as the standard. For the within-condition comparison, MMN was larger to /ɪ/ deviant minus /ε/ standard than to the reverse. A condition order effect was also observed. MMN amplitude was smaller to the deviant stimulus if it had previously served as the standard. In addition, the amplitudes of late discriminative negativity (LDN) showed similar asymmetry. LDN was larger for deviant /ε/ than deviant /ɪ/ when compared to themselves as the standard. These findings were compatible with an underspecification account, but also with other accounts, such as the Natural Referent Vowel model and a prototype model; we also suggest that non-linguistic factors need to be carefully considered as additional sources of speech processing asymmetries. (shrink)

Abstract Since their inception, distributional models of semantics have been criticized as inadequate cognitive theories of human semantic learning and representation. A principal challenge is that the representations derived by distributional models are purely symbolic and are not grounded in perception and action; this challenge has led many to favor feature-based models of semantic representation. We argue that the amount of perceptual and other semantic information that can be learned from purely distributional statistics has been underappreciated. We (...) compare the representations of three feature-based and nine distributional models using a semantic clustering task. Several distributional models demonstrated semantic clustering comparable with clustering-based on feature-based representations. Furthermore, when trained on child-directed speech, the same distributional models perform as well as sensorimotor-based feature representations of children’s lexical semantic knowledge. These results suggest that, to a large extent, information relevant for extracting semantic categories is redundantly coded in perceptual and linguistic experience. Detailed analyses of the semantic clusters of the feature-based and distributional models also reveal that the models make use of complementary cues to semantic organization from the two data streams. Rather than conceptualizing feature-based and distributional models as competing theories, we argue that future focus should be on understanding the cognitive mechanisms humans use to integrate the two sources. (shrink)

Civil proceedings do not terminate after adoption of a court decision. The enforcement of a material subjective right requires the second stage of legal defence, i.e., compulsory enforcement of a court decision. State courts are authorised to adopt decisions on behalf of the state and to ensure compulsory enforcement of an adopted court decision. Although enforcement proceedings constitute the final stage of civil procedure, it has some specific features, e.g., the representation in enforcement proceedings is distinctive. The article analyzes (...) application of this institute in enforcement proceedings: the subjects eligible to act as representatives in enforcement proceedings, the order of execution of their powers, and the problems of adjudicating remuneration for representation in enforcement proceedings. (shrink)

Structural representations are increasingly popular in philosophy of cognitive science. A key virtue they seemingly boast is that of meeting Ramsey's job description challenge. For this reason, structural representations appear tailored to play a clear representational role within cognitive architectures. Here, however, I claim that structural representations do not meet the job description challenge. This is because even our most demanding account of their functional profile is satisfied by at least some receptors, which paradigmatically fail the job description challenge. Hence, (...) the functional profile typically associated with structural representations does not identify representational posits. After a brief introduction, I present, in the second section of the paper, the job description challenge. I clarify why receptors fail to meet it and highlight why, as a result, they should not be considered representations. In the third section I introduce what I take to be the most demanding account of structural representations at our disposal, namely Gładziejewski's account. Provided the necessary background, I turn from exposition to criticism. In the first half of the fourth section, I equate the functional profile of structural representations and receptors. To do so, I show that some receptors boast, as a matter of fact, all the functional features associated with structural representations. Since receptors function merely as causal mediators, I conclude structural representations are mere causal mediators too. In the second half of the fourth section I make this conclusion intuitive with a toy example. I then conclude the paper, anticipating some objections my argument invites. (shrink)

Arguments for feature creation receive support from studies of young infants forming category representations from perceptual experience. A challenge for Schyns et al. will be to determine how a feature creation system might interface with a perceptual system that appears constrained to follow organizational principles that specify how edge segments should be grouped into functional units of coherent object representations.

Acts of intimate partner violence and abuse of nonhuman animals are common, harmful, and co-occurring phenomena. The aim of the present study was to identify perpetrator subtypes based on variable paths hypothesized to influence physical violence toward both partners and nonhuman animals: callousness and instrumental representations of aggression and rejection-sensitivity and expressive representations of aggression. Strong associations emerged between callousness and instrumental representations and between rejection-sensitivity and expressive representations. For males, callousness directly predicted both IPV and animal abuse. For females, (...) rejection-sensitivity predicted IPV. Instrumental representations mediated the relationship between callousness and animal abuse for females but not for males. Results suggest that IPV and animal abuse functionally interconnect, that perpetration of animal abuse may differ in function across gender, and that identifying distinct pathways to violence may facilitate violence prediction and prevention. (shrink)

Many philosophers and psychologists have attempted to elucidate the nature of mental representation by appealing to notions like isomorphism or abstract structural resemblance. The ‘structural representations’ that these theorists champion are said to count as representations by virtue of functioning as internal models of distal systems. In his 2007 book, Representation Reconsidered, William Ramsey endorses the structural conception of mental representation, but uses it to develop a novel argument against representationalism, the widespread view that cognition essentially involves (...) the manipulation of mental representations. Ramsey argues that although theories within the ‘classical’ tradition of cognitive science once posited structural representations, these theories are being superseded by newer theories, within the tradition of connectionism and cognitive neuroscience, which rarely if ever appeal to structural representations. Instead, these theories seem to be explaining cognition by invoking so-called ‘receptor representations’, which, Ramsey claims, aren’t genuine representations at all—despite being called representations, these mechanisms function more as triggers or causal relays than as genuine stand-ins for distal systems. I argue that when the notions of structural and receptor representation are properly explicated, there turns out to be no distinction between them. There only appears to be a distinction between receptor and structural representations because the latter are tacitly conflated with the ‘mental models’ ostensibly involved in offline cognitive processes such as episodic memory and mental imagery. While structural representations might count as genuine representations, they aren’t distinctively mental representations, for they can be found in all sorts of non-intentional systems such as plants. Thus to explain the kinds of offline cognitive capacities that have motivated talk of mental models, we must develop richer conceptions of mental representation than those provided by the notions of structural and receptor representation. (shrink)

According to a standard representationalist view cognitive capacities depend on internal content-carrying states. Recent alternatives to this view have been met with the reaction that they have, at best, limited scope, because a large range of cognitive phenomena—those involving absent and abstract features—require representational explanations. Here we challenge the idea that the consideration of cognition regarding the absent and the abstract can move the debate about representationalism along. Whether or not cognition involving the absent and the abstract requires the positing (...) of representations depends upon whether more basic forms of cognition require the positing of representations. (shrink)

This paper investigates how "representation" is actually used in some areas in cognitive neuroscience. It is argued that recent philosophy has largely ignored an important kind of representation that differs in interesting ways from the representations that are standardly recognized in philosophy of mind. This overlooked kind of representation does not represent by having intentional contents; rather members of the kind represent by displaying or instantiating features. The investigation is not simply an ethnographic study of the discourse (...) of neuroscientists. If there are indeed two different kinds of representations, and the non-standard ones are the ones referred to in some areas of cognitive neuroscience, then we will have to give up the idea that appealing to inner representations with intentional contents is the defining distinction between cognitive neuroscience and behaviorist psychology (Montgomery, 1995). Further, if the conclusions of this paper are correct, many general accounts of how neural states represent are either false or theoretically ill-motivated. (shrink)

This paper develops a theory of analog representation. We first argue that the mark of the analog is to be found in the nature of a representational system’s interpretation function, rather than in its vehicles or contents alone. We then develop the rulebound structure theory of analog representation, according to which analog systems are those that use interpretive rules to map syntactic structural features onto semantic structural features. The theory involves three degree-theoretic measures that capture three independent ways (...) in which a system can be more or less analog. We explain how our theory improves upon prior accounts of analog representation, provides plausible diagnoses for novel challenge cases, extends to hybrid systems that are partially analog and partially symbolic, and accounts for some of the advantages and disadvantages of representing analogically versus symbolically. (shrink)

Among theories of human language comprehension, cue‐based memory retrieval has proven to be a useful framework for understanding when and how processing difficulty arises in the resolution of long‐distance dependencies. Most previous work in this area has assumed that very general retrieval cues like [+subject] or [+singular] do the work of identifying (and sometimes misidentifying) a retrieval target in order to establish a dependency between words. However, recent work suggests that general, handpicked retrieval cues like these may not be enough (...) to explain illusions of plausibility (Cunnings & Sturt, 2018), which can arise in sentences like The letter next to the porcelain plate shattered. Capturing such retrieval interference effects requires lexically specific features and retrieval cues, but handpicking the features is hard to do in a principled way and greatly increases modeler degrees of freedom. To remedy this, we use well‐established word embedding methods for creating distributed lexical feature representations that encode information relevant for retrieval using distributed retrieval cue vectors. We show that the similarity between the feature and cue vectors (a measure of plausibility) predicts total reading times in Cunnings and Sturt’s eye‐tracking data. The features can easily be plugged into existing parsing models (including cue‐based retrieval and self‐organized parsing), putting very different models on more equal footing and facilitating future quantitative comparisons. (shrink)

This paper contrasts three different schemes of reference relevant to understanding systems of perceptual representation: a location-based system dubbed "feature-placing", a system of "visual indices" referring to things called "proto-objects", and the full sortal-based individuation allowed by a natural language. The first three sections summarize some of the key arguments (in Clark, 2000) to the effect that the early, parallel, and pre-attentive registration of sensory features itself constitutes a simple system of nonconceptual mental representation. In particular, (...) class='Hi'>feature integration--perceiving something as being both F and G, where F and G are sensible properties registered in distinct parallel streams--requires a referential apparatus. Section V. reviews some grounds for thinking that at these earliest levels this apparatus is location-based: that it has a direct and nonconceptual means of picking out places. Feature-placing is contrasted with a somewhat more sophisticated system that can identify and track four or five "perceptual objects" or "proto-objects", independently of their location, for as long as they remain perceptible. Such a system is found in Zenon Pylyshyn's fascinating work on "visual indices", in Dana Ballard's notion of deictic codes, and in Kahneman, Treisman, and Wolfe's accounts of systems of evanescent representations they call "object files". Perceptual representation is a layered affair, and I argue that it probably includes both feature-placing and proto-objects. Finally, both nonconceptual systems are contrasted with the full-blooded individuation allowed in a natural language. (shrink)

In this paper, I take scientific models to be epistemic representations of their target systems. I define an epistemic representation to be a tool for gaining information about its target system and argue that a vehicle’s capacity to provide specific information about its target system—its informativeness—is an essential feature of this kind of representation. I draw an analogy to our ordinary notion of interpretation to show that a user’s aim of faithfully representing the target system is necessary (...) for securing this feature. (shrink)

Many philosophers have understood the representational dimension of affective states along the model of sense-perceptual experiences, even claiming the relevant affective experiences are perceptual experiences. This paper argues affective experiences involve a kind of personal level affective representation disanalogous from the representational character of perceptual experiences. The positive thesis is that affective representation is a non-transparent, non-sensory form of evaluative representation, whereby a felt valenced attitude represents the object of the experience as minimally good or bad, and (...) one experiences that evaluative standing as having the power to causally motivate the relevant attitude. I show this view can make sense of distinctive features of affective experiences, such as their valence and connection to value in a way which moves beyond current evaluativist views of affect. (shrink)

The present study evaluates the cognitive representation of a kicking movement performed by a human and a humanoid robot, and how they are represented in experts and novices of soccer and robotics, respectively. To learn about the expertise-dependent development of memory structures, we compared the representation structures of soccer experts and robot experts concerning a human and humanoid robot kicking movement. We found different cognitive representation structures for both expertise groups under two different motor performance conditions . (...) In general, the expertise relies on the perceptual-motor knowledge of the human motor system. Thus, the soccer experts’ cognitive representation of the humanoid robot movement is dominated by their representation of the corresponding human movement. Additionally, our results suggest that robot experts, in contrast to soccer experts, access functional features of the technical system of the humanoid robot in addition to their perceptual-motor knowledge about the human motor system. Thus, their perceptual-motor and neuro-functional machine representation are integrated into a cognitive representation of the humanoid robot movement. (shrink)

Real-world judgements and decisions often require choosing from an open-ended set of options which cannot be exhaustively considered before a choice is made. Recent work has found that the options people do consider tend to have particular features, such as high historical value. Here, we pursue the idea that option generation during decision making may reflect a more general mechanism for calling things to mind, by which relevant features in a context-appropriate representational space guide what comes to mind. In this (...) paper, we evaluate this proposal primarily based on what comes to mind in different familiar categories. We first introduce an empirical approach for deriving the space of features that people use to represent items in a particular category and for locating the category members that come to mind within that space. We show that in both familiar and ad hoc categories, a category member’s location along certain dimensions of the derived feature space predicts its likelihood of coming to mind. Next, we show that category members from these feature space locations come to mind by default in a way that is somewhat impervious to conscious control. We then demonstrate that the extent to which a given dimension is a predictor of what comes to mind within a category is related to how relevant that feature is for representing the category in question, using a novel measure of general feature relevance. Finally, we illustrate the usefulness of this framework in the context of a decision making task. We close with the proposal that people call category members to mind according to their location in representational space, specifically based on the predicted usefulness of considering category members with particular features. (shrink)

Berkeley’s likeness principle states that only an idea can be like an idea. In this paper, I argue that the principle should be read as a premise only in a metaphysical argument showing that matter cannot instantiate anything like the sensory properties we perceive. It goes against those interpretations that take it to serve also, if not primarily, an epistemological purpose, featuring in Berkeley’s alleged Representation Argument to the effect that we cannot reach beyond the veil of our ideas. (...) First, in section 1, I raise some concerns about the traditional narrative concerning the likeness principle’s role in Berkeley’s argumentation. In section 2, I delineate an alternative narrative, arguing that there is no ‘missing premise’ in his alleged Representation Argument we need to explain simply because he advances no argument like that in the first place. In section 3–4, I provide a close reading of the relevant passages—first from the _Principles_, then the _Dialogues_—and their contexts, supporting textually a purely metaphysical interpretation of the likeness principle arguments. In section 5, I address some possible objections, based on the phrasing of the likeness principle passages and some related texts. (shrink)

'Representation' is a concept which occurs both in cognitive science and philosophy. It has common features in both settings in that it concerns the explanation of behaviour in terms of the way the subject categorizes and systematizes responses to its environment. The prevailing model sees representations as causally structured entities correlated on the one hand with elements in a natural language and on the other with clearly identifiable items in the world. This leads to an analysis of representation (...) and cognition in terms of formal symbols and their relations. But human perception and cognition use multiple informational constraints and deal with unsystematic and messy input in a way best explained by Parallel Distributed Processing models. This undermines the claim that a formal representational theory of mind is 'the only game in town.' In particular it suggests a radically different model of brain function and its relation to epistemology from that found in current representational theories. (shrink)

The issue of abstractness raises two distinct questions. First, is there a format-independent magnitude representation? Second, does analogue magnitude really play a crucial role in the development of human mathematics? We suggest that neither developmental nor cultural studies support this notion. The field needs to redefine the properties of the core number representation as used in human arithmetic.

The usefulness of contextually guided processors is investigated a little further. A more general use for binding V1 cell responses than the one in Phillips & Singer's target article is proposed, which takes into account that strong responses of these cells can mean more than the presence of lines and edges. The possibility of different grouping depending on the activities of neighboring cells is essential to the approach.

Advocates of dynamical systems theory (DST) sometimes employ revolutionary rhetoric. In an attempt to clarify how DST models differ from others in cognitive science, I focus on two issues raised by DST: the role for representations in mental models and the conception of explanation invoked. Two features of representations are their role in standing-in for features external to the system and their format. DST advocates sometimes claim to have repudiated the need for stand-ins in DST models, but I argue that (...) they are mistaken. Nonetheless, DST does offer new ideas as to the format of representations employed in cognitive systems. With respect to explanation, I argue that some DST models are better seen as conforming to the covering-law conception of explanation than to the mechanistic conception of explanation implicit in most cognitive science research. But even here, I argue, DST models are a valuable complement to more mechanistic cognitive explanations. (shrink)

A representation is a remnant of previous experience that allows that experience to affect later behavior. This paper develops a metatheoretical view of representation and applies it to issues concerning representation in animals. To describe a representational system one must specify the following: thedomainor range of situations in the represented world to which the system applies; thecontentor set of features encoded and preserved by the system; thecodeor transformational rules relating features of the representation to the corresponding (...) features of the represented world; themedium, or the representation's physical instantiation; and thedynamics, or how the system changes with time. In part because of the behaviorist assumption that the hypothetical, covert changes occurring in an organism during learning correspond to the overt physical changes that are observed, issues of representation in animal behavior have been largely ignored as irrelevant or misleading. However, it can be inferred that representations, acting as models of environmental regularities, operate at many levels of behavioral functioning, both cognitive and noncognitive. Objections to the use of this concept in explanations of animal behavior, based on the claim that it is indeterminate and on behaviorist considerations of parsimony, can be answered. Animal representations may be specialized in terms of tasks and species. Data from tasks involving spatial memory, delayed matching-to-sample, and sequence learning suggest some foundations for a general theory of animal representations. (shrink)

The main task of this paper is to understand if and how static images like photographs can represent and/or depict temporal extension (duration). In order to do this, a detour will be necessary to understand some features of the nature of photographic representation and depiction in general. This important detour will enable us to see that photographs (can) have a narrative content, and that the skilled photographer can 'tell a story' in a very clear sense, as well as control (...) and guide the attention of the spectator of the photograph. The understanding and defence of this claim is a secondary aim of this paper, and it will then allow us to provide a good treatment of the particular case of photographic representation and depiction of temporal extension. (shrink)

‘The Natural System’ is the abstract notion of the order in living diversity. The richness and complexity of this notion is revealed by the diversity of representations of the Natural System drawn by ornithologists in the Nineteenth Century. These representations varied in overall form from stars, to circles, to maps, to evolutionary trees and cross-sections through trees. They differed in their depiction of affinity, analogy, continuity, directionality, symmetry, reticulation and branching, evolution, and morphological convergence and divergence. Some representations were two-dimensional, (...) and some were three-dimensional; n-dimensional representations were discussed but never illustrated. The study of diagrammatic representations of the Natural System is made difficult by the frequent failure of authors to discuss them in their texts, and by the consequent problem of distinguishing features which carried meaning from arbitrary features and printing conventions which did not. Many of the systematics controversies of the last thirty years have their roots in the conceptual problems which surrounded the Natural System in the late 1800s, problems which were left unresolved when interest in higher-level systematics declined at the turn of this century. (shrink)

The goal of this paper is to present some advantages of the representational and computational theories of mind when compared to other views, especially behaviorism. The idea is that representational and computational theories allow us to conceive propositional attitudes in a way that preserves two essential features we take them to have in common sense psychological explanations: semantic evaluability and causal efficacy. Behaviorism reconceives mental states in a way that doesn’t preserve these essential features. In so doing, it makes a (...) mystery of the success of common sense psychology. I illustrate some of the difficulties that behaviorism faces by considering and criticizing Wittgenstein’s approach to linguistic understanding. The upshot is that representational and computational theories of mind do a better job at vindicating common sense psychology, and so are to be preferred when compared to behaviorism. (shrink)

For unmanned aerial vehicle, object detection at different scales is an important component for the visual recognition. Recent advances in convolutional neural networks have demonstrated that attention mechanism remarkably enhances multiscale representation of CNNs. However, most existing multiscale feature representation methods simply employ several attention blocks in the attention mechanism to adaptively recalibrate the feature response, which overlooks the context information at a multiscale level. To solve this problem, a multiscale feature filtering network is proposed (...) in this paper for image recognition system in the UAV. A novel building block, namely, multiscale feature filtering module, is proposed for ResNet-like backbones and it allows feature-selective learning for multiscale context information across multiparallel branches. These branches employ multiple atrous convolutions at different scales, respectively, and further adaptively generate channel-wise feature responses by emphasizing channel-wise dependencies. Experimental results on CIFAR100 and Tiny ImageNet datasets reflect that the MFFNet achieves very competitive results in comparison with previous baseline models. Further ablation experiments verify that the MFFNet can achieve consistent performance gains in image classification and object detection tasks. (shrink)

Production and comprehension have long been viewed as inseparable components of language. The study of vision, by contrast, has centered almost exclusively on comprehension. Here we investigate drawing—the most basic form of visual production. How do we convey concepts in visual form, and how does refining this skill, in turn, affect recognition? We developed an online platform for collecting large amounts of drawing and recognition data, and applied a deep convolutional neural network model of visual cortex trained only on natural (...) images to explore the hypothesis that drawing recruits the same abstract feature representations that support natural visual object recognition. Consistent with this hypothesis, higher layers of this model captured the abstract features of both drawings and natural images most important for recognition, and people learning to produce more recognizable drawings of objects exhibited enhanced recognition of those objects. These findings could explain why drawing is so effective for communicating visual concepts, they suggest novel approaches for evaluating and refining conceptual knowledge, and they highlight the potential of deep networks for understanding human learning. (shrink)

Many critics of dual-process models have mistaken long lists of descriptive terms in the literature for a full-blown theory of necessarily co-occurring properties. These critiques have distracted attention from the cumulative progress being made in identifying the much smaller set of properties that truly do define Type 1 and Type 2 processing. Our view of the literature is that autonomous processing is the defining feature of Type 1 processing. Even more convincing is the converging evidence that the key (...) class='Hi'>feature of Type 2 processing is the ability to sustain the decoupling of secondary representations. The latter is a foundational cognitive requirement for hypothetical thinking. (shrink)

Production and comprehension have long been viewed as inseparable components of language. The study of vision, by contrast, has centered almost exclusively on comprehension. Here we investigate drawing—the most basic form of visual production. How do we convey concepts in visual form, and how does refining this skill, in turn, affect recognition? We developed an online platform for collecting large amounts of drawing and recognition data, and applied a deep convolutional neural network model of visual cortex trained only on natural (...) images to explore the hypothesis that drawing recruits the same abstract feature representations that support natural visual object recognition. Consistent with this hypothesis, higher layers of this model captured the abstract features of both drawings and natural images most important for recognition, and people learning to produce more recognizable drawings of objects exhibited enhanced recognition of those objects. These findings could explain why drawing is so effective for communicating visual concepts, they suggest novel approaches for evaluating and refining conceptual knowledge, and they highlight the potential of deep networks for understanding human learning. (shrink)

The paper considers the problems involved in getting neural networks to learn about highly structured task domains. A central problem concerns the tendency of networks to learn only a set of shallow (non-generalizable) representations for the task, i.e., to miss the deep organizing features of the domain. Various solutions are examined, including task specific network configuration and incremental learning. The latter strategy is the more attractive, since it holds out the promise of a task-independent solution to the problem. Once we (...) see exactly how the solution works, however, it becomes clear that it is limited to a special class of cases in which (1) statistically driven undersampling is (luckily) equivalent to task decomposition, and (2) the dangers of unlearning are somehow being minimized. The technique is suggestive nonetheless, for a variety of developmental factors may yield the functional equivalent of both statistical AND informed undersampling in early learning. (shrink)