High-level Properties Research Articles

Scene Image Retrieval Based on Manifold Structures of Canonical Images

Image retrieval methods have been dramatically developed in the last decade. In this paper, we propose a novel method for image retrieval based on manifold structures of canonical images. Firstly, we present the image normalization process to find a set of canonical images that anchors the probabilistic distributions around the real data manifolds to learn the representations that better encode the manifold structures in general high-dimensional image space. In addition, we employ the canonical images as the centers of the conditional multivariate Gaussian distributions. This approach allows to learn more detailed structures of the partial manifolds resulting in improved representation of the high level properties of scene images. Furthermore, we use the probabilistic framework of the extended model to retrieve images based on the similarity measure of reciprocal likelihood of pairs of images and the sum of likelihood of one of two images based on the other’s best distributions. We estimate our method using SUN database. In the experiments on scene image retrieval, the proposed method is efficient, and exhibits superior capabilities compared to other methods, such as GIST.

Conducting Comparable Research in Representative Worlds

Field-based simulation research can be delayed or prevented due to restricted resources and other practical challenges. Although laboratory work is a feasible alternative, it is often criticized for a lack of generalizability. We faced this issue when investigating the impact of workplace interruptions on nurses’ work performance in the Intensive Care Unit (ICU). The potential relationship between interruptions and errors has been widely investigated in healthcare settings; however, much of the evidence is associative. Some evidence outside the healthcare domain points to a causal connection between interruptions and errors, but the studies are mostly laboratory based and interruptions are artificial to the situation. Our eventual aim is to carry out a high-fidelity randomized-controlled trial to test the hypothesis that interruptions cause errors in healthcare, which could have major implications for interventions and policy. However, there are considerable challenges and constraints to overcome when designing such an experiment; for example (a) a limited potential participant pool within the ICU, (b) constant changes to technology and procedures in the ICU, and (c) restricted access to hospital simulation rooms. There are various ways to address these issues but most options compromise the generalizability of the simulation to authentic situations. By adopting principles of Brunswick’s representative design, we designed an initial laboratory study to be used as a formal pilot study in a different, but parallel, context to healthcare. Representative design refers to the “arrangement of conditions of an experiment so that they represent the behavioural setting to which the results are intended to apply” (Araujo, Davids, & Passos, 2007, p. 72). Representing the ‘natural world’ with high fidelity is not crucial in representative design, but rather ensuring that the properties of the conditions to which the researcher wishes to generalize are adequately captured in the laboratory task (Hammond & Stewart, 2001). Accordingly, we have created a laboratory-based simulation on the basis of the properties we aim to capture in a healthcare-based simulation, so that we can generalize findings from the former to the latter. The laboratory component needed to involve a task that embodied the high-level properties of medication preparation and administration, with a specialized population who regularly perform the task. The task of cocktail making fits the above requirements because it has high-level properties similar to medication preparation and administration. Cocktail making is concerned with controlled liquids, it requires perceptual motor skills, it involves multi-step tasks carried out in a busy environment where there is a high demand on working memory, and it is performed by experts (bartenders). These similarities mean that cocktail making can be used as a laboratory analog of certain aspects of medication preparation and administration. First, we mapped the physical environment across the two domains to ensure the cocktail component possessed the same spatial properties as the medication component. Second, we designed the high-level structure of the cocktail scenarios to approximate the high-level structure of the medication component scenarios. Third, we designed the interrupting tasks and added them to the scenarios. Experiment 1 of the cocktail component was a condition with zero interruptions to provide a baseline error rate with which to calculate the required sample size for Experiment 2. Experiment 2 was a between-subjects study in which participants were randomly assigned to receive either 3 or 12 interruptions. All participants had at least one-month cocktail making experience in a licensed venue. Cocktail errors were the analog of clinical errors in healthcare (Westbrook et al., 2010). In Experiment 1, an average of 44% of cocktails made contained at least one error per scenario ( SD = 18%). To calculate the required sample size for Experiment 2, an effect size was calculated by integrating our baseline data with observational data from Westbrook et al. (2010) and a power analysis was performed. Data collection for Experiment 2 is now completed. The findings from Experiment 2 will be used to calculate the required sample size for the medication component and lessons learned from the cocktail component will help finalize the design of the medication component. The cocktail component of our study is not totally analogous to the medication component, but we have shown that principles of representative design can be used to design a simulation from which we can argue that generalizable findings can be gathered. A major advantage of our approach is that we have been able to design and test many formal aspects of the medication component of our study prior to stepping into the hospital simulator. Findings from the medication component will help to shape interventions and policies in the healthcare domain that reduce error rates and increase patient safety. The cocktail component will contribute to the interruptions literature because the interruptions are representative of those that would actually occur in the workplace – something that is rare in laboratory-based interruptions research.

From Emergence Theory to Panpsychism—A Philosophical Evaluation of Nancey Murphy’s Non-reductive Physicalism

In this article, I offer a critical evaluation of non-reductive physicalism as articulated and defended by Nancey Murphy. I argue that (A) the examples given by Murphy do not illustrate robust emergence and the philosophical idea of downward causation. (B) The thesis of multiple realizability is ontologically neutral, and so cannot support the idea of the causal efficacy of higher-level properties. (C) Supervenience is incompatible with strong emergence. I also argue for the fruitful relationship between emergence theory and panpsychism pertaining to the metaphysical issue of the origin and nature of mind.

Adding Shape to Saliency: A Proto-object Saliency Map for Predicting Fixations during Scene Viewing

Traditional saliency models predict fixations during scene viewing by computing local contrast between low-level color, intensity and orientation features; the higher the summed contrast the greater the probability of fixation. Evidence also suggests that high-level properties of objects are predictive of fixation locations. We attempt fixation prediction from proto-objects (POs), a mid-level representation existing between features and objects. Using our previously-reported proto-object model (Yu et. al., 2014, JoV), we segmented 384 images of real-world scenes into proto-objects, fragments of visual space, at multiple resolutions (feature-space bandwidths). We then built from these segmentations a saliency map by computing feature contrast between each proto-object and its local neighbors using intensity, color, orientation, and now, size and shape features. Center-surround size contrast was computed by comparing pixel area between a given proto-object and each "surrounding" neighbor. To compute shape contrast we first normalized a proto-object and a neighbor to have the same area, aligned them based on maximum area-overlap, then counted the number of pixels in the overlapping area (divided by the union of the areas), with a smaller overlap over neighbors coding a higher contrast. Doing this relative to each proto-object, then combining contrast signals across features and resolutions, generates a proto-object saliency map, which we used to predict the fixation behavior of 12 participants freely viewing the same scenes (each for 3 seconds) in anticipation of a memory test. We found that our proto-object saliency map predicted fixations as well or better than an Itti-Koch saliency model, and was nearly as predictive as the upper-limit defined by a Subject model obtained using the leave-one-out method. We conclude that size and shape features, quantified in terms of proto-objects, are used to guide overt visual attention, and that saliency-based models of fixation prediction need to recognize the importance of these mid-level visual features. Meeting abstract presented at VSS 2016

Form and function in gene regulatory networks: the structure of network motifs determines fundamental properties of their dynamical state space.

Network motifs have been studied extensively over the past decade, and certain motifs, such as the feed-forward loop, play an important role in regulatory networks. Recent studies have used Boolean network motifs to explore the link between form and function in gene regulatory networks and have found that the structure of a motif does not strongly determine its function, if this is defined in terms of the gene expression patterns the motif can produce. Here, we offer a different, higher-level definition of the ‘function’ of a motif, in terms of two fundamental properties of its dynamical state space as a Boolean network. One is the basin entropy, which is a complexity measure of the dynamics of Boolean networks. The other is the diversity of cyclic attractor lengths that a given motif can produce. Using these two measures, we examine all 104 topologically distinct three-node motifs and show that the structural properties of a motif, such as the presence of feedback loops and feed-forward loops, predict fundamental characteristics of its dynamical state space, which in turn determine aspects of its functional versatility. We also show that these higher-level properties have a direct bearing on real regulatory networks, as both basin entropy and cycle length diversity show a close correspondence with the prevalence, in neural and genetic regulatory networks, of the 13 connected motifs without self-interactions that have been studied extensively in the literature.

The Role of Visual and Semantic Properties in the Emergence of Category-Specific Patterns of Neural Response in the Human Brain.

Brain-imaging studies have found distinct spatial and temporal patterns of response to different object categories across the brain. However, the extent to which these categorical patterns of response reflect higher-level semantic or lower-level visual properties of the stimulus remains unclear. To address this question, we measured patterns of EEG response to intact and scrambled images in the human brain. Our rationale for using scrambled images is that they have many of the visual properties found in intact images, but do not convey any semantic information. Images from different object categories (bottle, face, house) were briefly presented (400 ms) in an event-related design. A multivariate pattern analysis revealed categorical patterns of response to intact images emerged ∼80–100 ms after stimulus onset and were still evident when the stimulus was no longer present (∼800 ms). Next, we measured the patterns of response to scrambled images. Categorical patterns of response to scrambled images also emerged ∼80–100 ms after stimulus onset. However, in contrast to the intact images, distinct patterns of response to scrambled images were mostly evident while the stimulus was present (∼400 ms). Moreover, scrambled images were able to account only for all the variance in the intact images at early stages of processing. This direct manipulation of visual and semantic content provides new insights into the temporal dynamics of object perception and the extent to which different stages of processing are dependent on lower-level or higher-level properties of the image.

How facial attractiveness affects sustained attention

The present study investigated whether and how facial attractiveness affects sustained attention. We adopted a multiple-identity tracking paradigm, using attractive and unattractive faces as stimuli. Participants were required to track moving target faces amid distractor faces and report the final location of each target. In Experiment 1, the attractive and unattractive faces differed in both the low-level properties (i.e., luminance, contrast, and color saturation) and high-level properties (i.e., physical beauty and age). The results showed that the attractiveness of both the target and distractor faces affected the tracking performance: The attractive target faces were tracked better than the unattractive target faces; when the targets and distractors were both unattractive male faces, the tracking performance was poorer than when they were of different attractiveness. In Experiment 2, the low-level properties of the facial images were equalized. The results showed that the attractive target faces were still tracked better than unattractive targets while the effects related to distractor attractiveness ceased to exist. Taken together, the results indicate that during attentional tracking the high-level properties related to the attractiveness of the target faces can be automatically processed, and then they can facilitate the sustained attention on the attractive targets, either with or without the supplement of low-level properties. On the other hand, only low-level properties of the distractor faces can be processed. When the distractors share similar low-level properties with the targets, they can be grouped together, so that it would be more difficult to sustain attention on the individual targets.

Physicalism, Truthmaking, and Levels of Reality: Prospects and Problems

This paper considers the extent to which the notion of truthmaking can play a substantive role in defining physicalism. While a truthmaking-based approach to physicalism is prima facie attractive, there is some reason to doubt that truthmaking can do much work when it comes to understanding physicalism, and perhaps austere metaphysical frameworks in general. First, despite promising to dispense with higher-level properties and states, truthmaking appears to make little progress on issues concerning higher-level items and how they are related to how things are physically. Second, it seems that truthmaking-based approaches to physicalism will have a difficult time addressing the status of truthmaking itself without, in effect, appealing to the resources of alternative ways of conceptualizing physicalism.

Automated Determining of Manufacturing Properties and Their Evolutionary Changes from Event Traces

Production plants are usually kept in operation for several decades. During this long operational phase operation requirements and other production conditions change frequently. Accordingly, the plants have to be adjusted in behavior and/or structure by adapting software and physics of the plant to avoid degeneration. Unfortunately, in industrial practice, changes, especially smaller ones, are often performed ad-hoc without appropriate adaptation of formal models or documentation. As a consequence, knowledge about the process is only implicitly available and an evaluation of performed changes is often omitted, resulting in sub-optimal production performance. Present research approaches to overcome these deficiencies usually concentrate on (a) manual modelling with manual or automatic analysis on a high level of abstraction; or (b) on automatic model generation from observations without lifting gathered knowledge to easy interpretable indicators. The approach presented in this paper combines both methods (a) and (b) by learning models from observation of input / output signals of the production plant’s control system. Semantics are added by using a priori information modelling which is less tedious compared to modelling the process itself. The learned models are used to automatically detect changes by continuously comparing their behavior with real plant behavior during operation as well as to evaluate performed changes. An analysis of the models results in high-level property values such as key performance indicators or flexibility measures of the production system.

Cogent

We present an approach to writing and formally verifying high-assurance file-system code in a restricted language called Cogent, supported by a certifying compiler that produces C code, high-level specification of Cogent, and translation correctness proofs. The language is strongly typed and guarantees absence of a number of common file system implementation errors. We show how verification effort is drastically reduced for proving higher-level properties of the file system implementation by reasoning about the generated formal specification rather than its low-level C code. We use the framework to write two Linux file systems, and compare their performance with their native C implementations.

Cogent

We present an approach to writing and formally verifying high-assurance file-system code in a restricted language called Cogent, supported by a certifying compiler that produces C code, high-level specification of Cogent, and translation correctness proofs. The language is strongly typed and guarantees absence of a number of common file system implementation errors. We show how verification effort is drastically reduced for proving higher-level properties of the file system implementation by reasoning about the generated formal specification rather than its low-level C code. We use the framework to write two Linux file systems, and compare their performance with their native C implementations.

Defending the liberal-content view of perceptual experience: direct social perception of emotions and person impressions

The debate about direct perception encompasses different topics, one of which concerns the richness of the contents of perceptual experiences. Can we directly perceive only low-level properties, like edges, colors etc. (the sparse-content view), or can we perceive high-level properties and entities as well (the liberal-content view)? The aim of the paper is to defend the claim that the content of our perceptual experience can include emotions and also person impressions. Using these examples, an argument is developed to defend a liberal-content view for core examples of social cognition. This view is developed and contrasted with accounts which claim that in the case of registering another person’s emotion while seeing them, we have to describe the relevant content not as the content of a perceptual experience, but of a perceptual belief. The paper defends the view that perceptual experiences can have a rich content yet remain separable from beliefs formed on the basis of the experience. How liberal and enriched the content of a perceptual experience is will depend upon the expertise a person has developed in the field. This is supported by the argument that perceptual experiences can be systematically enriched by perceiving affordances of objects, by pattern recognition or by top-down processes, as analyzed by processes of cognitive penetration or predictive coding.

Gappiness and the Case for Liberalism about Phenomenal Properties

Conservatives claim that all phenomenal properties are sensory. Liberals countenance non-sensory phenomenal properties such as what it's like to perceive some high-level property, and what it's like to think that p. A hallmark of phenomenal properties is that they present an explanatory gap, so to resolve the dispute we should consider whether experience has non-sensory properties that appear ‘gappy’. The classic tests for ‘gappiness’ are the invertibility test and the zombifiability test. I suggest that these tests yield conflicting results: non-sensory properties lend themselves to zombie scenarios but not to inversion scenarios. Which test should we trust? Against Carruthers & Veillet, I argue that invertibility is not a viable condition of phenomenality. In contrast, being zombifiable is credibly necessary and sufficient for phenomenality. I conclude that there are non-sensory properties of experience that are ‘gappy’ in the right way, and that liberalism is therefore the most plausible position.

Impact of lime treated soils performance on design of earthfill dikes

Nowadays soil treatment with lime in civil engineering is widespread in many countries on all continents, within several construction fields. The interest of the hydraulic works community regarding this technique is currently growing. It has been indeed shown during the last decade that appropriate treatment technologies provide lime treated soils with high level properties such as excellent homogeneity, low permeability, internal and external erosion resistance and mechanical stability. Those have been shown in laboratory and for some properties with full scale experiments. The so conferred soil properties can lead to innovative earthfill dams and dikes designs by addressing some of the typical designer’s problems, such as stability, watertightness, internal erosion, surface protection and flood control. However, lime treated soil external erosion resistance is still to be quantified in the field for proper designing and dimensioning of lime treated soil external erosion protection or spillways. With this purpose, an experimental earthfill dike has been built along the river Vidourle (France) in July 2015, in the frame of the French RD the experimental dike construction and lessons learned; the monitoring program; the dike design perspectives opened by soil treatment.

Cue-invariant shape recognition in rats as tested with second-order contours.

Nonhuman primates are the main animal model to investigate high-level properties of human cortical vision. For one property, transformation-invariant object recognition, recent studies have revealed interesting and unknown capabilities in rats. Here we report on the ability of rats to rely upon second-order cues that are important to structure the incoming visual images into figure and background. Rats performed a visual shape discrimination task in which the shapes were not only defined by first-order luminance information but also by a variety of second-order cues such as a change in texture properties. Once the rats were acquainted with a first set of second-order stimuli, they showed a surprising degree of generalization towards new second-order stimuli. The limits of these capabilities were tested in various ways, and the ability to extract the shapes broke down only in extreme cases where no local cues were available to solve the task. These results demonstrate how rats are able to make choices based on fairly complex strategies when necessary.

Probability and Manipulation: Evolution and Simulation in Applied Population Genetics

I define a concept of causal probability and apply it to questions about the role of probability in evolutionary processes. Causal probability is defined in terms of manipulation of patterns in empirical outcomes by manipulating properties that realize objective probabilities. The concept of causal probability allows us see how probabilities characterized by different interpretations of probability can share a similar causal character, and does so in such way as to allow new inferences about relationships between probabilities realized in different chance setups. I clarify relations between probabilities and properties defined in terms of them, and argue that certain widespread uses of computer simulations in evolutionary biology show that many probabilities relevant to evolutionary outcomes are causal probabilities. This supports the claim that higher-level properties such as biological fitness and processes such as natural selection are causal properties and processes, contrary to what some authors have argued.

The automaticity of perceiving animacy: Goal-directed motion in simple shapes influences visuomotor behavior even when task-irrelevant.

Visual processing recovers not only simple features, such as color and shape, but also seemingly higher-level properties, such as animacy. Indeed, even abstract geometric shapes are readily perceived as intentional agents when they move in certain ways, and such percepts can dramatically influence behavior. In the wolfpack effect, for example, subjects maneuver a disc around a display in order to avoid several randomly moving darts. When the darts point toward the disc, subjects (falsely) perceive that the darts are chasing them, and this impairs several types of visuomotor performance. Are such effects reflexive, automatic features of visual processing? Or might they instead arise only as contingent strategies in tasks in which subjects must interact with (and thus focus on the features of) such objects? We explored these questions in an especially direct way-by embedding such displays into the background of a completely independent "foraging" task. Subjects now moved their disc to collect small "food" dots (which appeared sequentially in random locations) as quickly as possible. The darts were task-irrelevant, and subjects were encouraged to ignore them. Nevertheless, foraging was impaired when the randomly moving darts pointed at the subjects' disc, as compared to control conditions in which they were either oriented orthogonally to the subjects' disc or pointed at another moving shape-thereby controlling for nonsocial factors. The perception of animacy thus influences downstream visuomotor behavior in an automatic manner, such that subjects cannot completely override the influences of seemingly animate shapes even while attempting to ignore them.

Visual experience: rich but impenetrable

According to so-called “thin” views about the content of experience, we can only visually experience low-level features such as colour, shape, texture or motion. According to so-called “rich” views, we can also visually experience some high-level properties, such as being a pine tree or being threatening. One of the standard objections against rich views is that high-level properties can only be represented at the level of judgment. In this paper, I first challenge this objection by relying on some recent studies in social vision. Secondly, I tackle a different but related issue, namely, the idea that, if the content of experience is rich, then perception is cognitively penetrable. Against this thesis, I argue that the very same criteria that help us vindicate the truly sensory nature of our rich experiences speak against their being cognitively penetrable.

Relative sensitivity to low- vs. high-level visual properties in face-sensitive regions of the human ventral occipito-temporal cortex: evidence from intra-cerebral recordings

Neuroimaging studies have identified a network of regions in the human ventral occipito-temporal cortex (VOTC) sensitive to faces. Using intra-cerebral electrophysiological recordings, we compared the relative responsivity of the different regions to face-structure using a periodic visual stimulation paradigm. Recordings were made in 17 epileptic patients implanted with linear-array depth-electrodes (>1000 contacts in the VOTC). In Experiment 1, we presented sequences of phase-scrambled and intact face stimuli (equalised for low-level image properties) alternating at a fixed frequency of 3Hz (6 images/s) while patients completed an orthogonal task. Significant periodic EEG responses (SSVEPs) reflecting high-level face structure processing were found at 3 Hz, while responses reflecting low-level visual processing (e.g. local contrast change) were recorded at 6 Hz. An index of high-level processing (3Hz/(3Hz+6Hz)) showed a posterior-to-anterior gradient, reflecting increased sensitivity to face structure going from the occipital to the temporo-polar cortex, with a right hemisphere lateralisation. In Experiment 2, we further investigated face structure processing with a sweep VEP paradigm (Ales et al., 2012; 4/17 patients). Stimulation was similar to Experiment 1 except the visibility (phase coherence) of the face image was parametrically varied so that a face gradually emerged over the course of a sequence. The patients performed a face detection task during the trials. Responses at 3Hz, objective markers of face detection, were found in all patients on contacts in the lateral fusiform gyrus and inferior occipital gyrus that overlapped with those defined in Experiment 1. All regions showed non-linear voltage vs coherence functions at 3Hz, with thresholds lying around ≈40% coherence. Sites in the lateral fusiform gyrus had a distinct step-like response profile suggestive of categorical face detection. These findings demonstrate that face-sensitive regions are organised with a posterior-to-anterior gradient of increasingly selective high-level visual processing. These regions appear to detect face structure in a non-linear/categorical manner. Meeting abstract presented at VSS 2015.

Computational Aerodynamic Analysis of a Micro-CT Based Bio-Realistic Fruit Fly Wing.

The aerodynamic features of a bio-realistic 3D fruit fly wing in steady state (snapshot) flight conditions were analyzed numerically. The wing geometry was created from high resolution micro-computed tomography (micro-CT) of the fruit fly Drosophila virilis. Computational fluid dynamics (CFD) analyses of the wing were conducted at ultra-low Reynolds numbers ranging from 71 to 200, and at angles of attack ranging from -10° to +30°. It was found that in the 3D bio-realistc model, the corrugations of the wing created localized circulation regions in the flow field, most notably at higher angles of attack near the wing tip. Analyses of a simplified flat wing geometry showed higher lift to drag performance values for any given angle of attack at these Reynolds numbers, though very similar performance is noted at -10°. Results have indicated that the simplified flat wing can successfully be used to approximate high-level properties such as aerodynamic coefficients and overall performance trends as well as large flow-field structures. However, local pressure peaks and near-wing flow features induced by the corrugations are unable to be replicated by the simple wing. We therefore recommend that accurate 3D bio-realistic geometries be used when modelling insect wings where such information is useful.