Search

In recent years, a considerable research effort has shown the energy benefits of implementing neural networks with memristors or other emerging memory technologies. However, for extreme-edge applications with high uncertainty, access to reduced amounts of data, and where explainable decisions are required, neural networks may not provide an acceptable form of intelligence. Bayesian reasoning can solve these concerns, but it is computationally expensive and, unlike neural networks, does not translate naturally to memristor-based architectures. In this work, we introduce, demonstrate experimentally on a fully fabricated hybrid CMOS-memristor system, and analyze a Bayesian machine designed for highly-energy efficient Bayesian reasoning. The architecture of the machine is obtained by writing Bayes' law in a way making its implementation natural by the principles of distributed memory and stochastic computing, allowing the circuit to function using solely local memory and minimal data movement. Measurements on a fabricated small-scale Bayesian machine featuring 2,048 memristors and 30,080 transistors show the viability of this approach and the possibility of overcoming the challenges associated with its design: the inherent imperfections of memristors, as well as the need to distribute very locally higher-than-nominal supply voltages. The design of a scaled-up version of the machine shows its outstanding energy efficiency on a real-life gesture recognition task: a gesture can be recognized using 5,000 times less energy than using a microcontroller unit. The Bayesian machine also features several desirable features, e.g., instant on/off operation, compatibility with low supply voltages, and resilience to single-event upsets. These results open the road for Bayesian reasoning as an attractive way for energy-efficient, robust, and explainable intelligence at the edge.

International audience; Visual object tracking is a fundamental function of computer vision that has been the object of numerous studies. The diversity of the proposed approaches leads to the idea of trying to fuse them and take advantage of their individual strengths while controlling the noise they may introduce in some circumstances. The work presented here describes a generic framework for combining and/or selecting on-line the different components of the processing chain of a set of trackers, and examines the impact of various fusion strategies. The results are assessed from a repertoire of 9 state-of-the-art trackers evaluated over 46 fusion strategies on the VOT 2013, VOT 2015 and OTB-100 datasets. A complementarity measure able to predict the overall performance of a given set of trackers is also proposed.; Le suivi d'objets visuels est une fonction fondamentale de la vision par ordinateur qui a fait l'objet de nombreuses études. La diversité des approches proposées conduit à l'idée d'essayer de les fusionner et de tirer parti de leurs forces individuelles tout en contrôlant le bruit qu'elles peuvent introduire dans certaines circonstances. Le travail présenté ici décrit un cadre générique pour combiner et / ou sélectionner en ligne les différentes composantes de la chaîne de traitement d'un ensemble d'algorithmes de pistage et examine l'impact de différentes stratégies de fusion. Les résultats sont évalués à partir d'un répertoire de 9 algorithmes de l'état de l'art évalués sur 46 stratégies de fusion sur les jeux de données VOT 2013, VOT 2015 et OTB-100. Une mesure de complémentarité capable de prédire la performance globale d'un ensemble donné d'algorithles de pistage est également proposée.

Living organisms survive and multiply even though they have uncertain and incomplete information about their environment and imperfect models to predict the consequences of their actions. Bayesian models have been proposed to face this challenge. Indeed, Bayesian inference is a way to do optimal reasoning when only uncertain and incomplete information is available. Various perceptive, sensory-motor, and cognitive functions have been successfully modeled this way. However, the biological mechanisms allowing animals and humans to represent and to compute probability distributions are not known. It has been proposed that neurons and assemblies of neurons could be the appropriate scale to search for clues to probabilistic reasoning. In contrast, in this paper, we propose that interacting populations of macromolecules and diffusible messengers can perform probabilistic computation. This suggests that probabilistic reasoning, based on cellular signaling pathways, is a fundamental skill of living organisms available to the simplest unicellular organisms as well as the most complex brains. Cell signaling implements computation of Rational Function with Non-negative coefficients.Subjective probabilistic inference requires computation of Rational Function with Non-negative coefficients.Cell signaling implements computation of subjective probabilistic inference.

We propose a time-domain multiple sound source localization (SSL) method based on Bayesian inference. This method is specifically designed to run on the stochastic machines (SM) that we are currently developing to perform efficient low-level sensor signal processing with ultra-low power consumption. The proposed SSL method is divided into two main parts. First, a probabilistic model is run on 50 very short time frames (3. 75ms each) of multichannel recorded signals. Second, the results obtained on the different frames are fused to obtain a final localization map. Using the system in a supervised way allows to extract estimated source locations by selecting as many maxima as there are sources in the room. We explain how this method is implemented on a SM. Experiments are presented to illustrate the performance and robustness of the resulting system.

International audience; Compared to conventional processors, stochastic computing architectures have strong potential to speed up computation time and to reduce power consumption. We present such an architecture, called Bayesian Machine (BM), dedicated to solving Bayesian inference problems. Given a set of noisy signals provided by low-level sensors, a BM estimates the posterior probability distribution of an unknown target information. In the present study, a BM is used to solve a sound source localization (SSL) problem: the BM computes the probability distribution of the position of a sound source given acoustic signals captured by a set of microphones. Assuming free field wave propagation (no reverberations), we express the SSL problem as the maximization of a likelihood function fed with audio features provided by the time-frequency (TF) analysis of the captured audio waves. The proposed BM uses bitwise parallel sampling to fuse the resulting multi-channel information. As the number of channels to fuse is large, the standard BM architecture encounters the so-called " time dilution problem " (long delays are necessary to obtain valid samples). We tackle this problem by using max-normalization of the distributions combined with a periodic re-sampling of the bit streams after processing a reasonably small subset of evidences. Finally, we compare the localization performance of the proposed machine with the results obtained using a standard version of the machine. The re-sampling leads to an impressive acceleration factor of 10³ in the computation.

International audience; Advancements in autonomous robotic systems have been impeded by the lack of a specialized computational hardware that makes real-time decisions based on sensory inputs. We have developed a novel circuit structure that efficiently approximates naïve Bayesian inference with simple Muller C-elements. Using a stochastic computing paradigm, this system enablesreal-time approximate decision-making with an area-energy-delay product nearly one billiontimes smaller than a conventional general-purpose computer. In this paper, we propose severaltechniques to improve the approximation of Bayesian inference by reducing stochastic bitstream autocorrelation. We also evaluate the effectiveness of these techniques for various naïve inference tasks and discuss hardware considerations, concluding that these circuits enable approximate Bayesian inferences while retaining orders-of-magnitude hardware advantagescompared to conventional general-purpose computers.

International audience; —As the physical limits of Moore's law are being reached, a research effort is launched to achieve further performance improvements by exploring computation paradigms departing from standard approaches. The BAMBI project (Bottom-up Approaches to Machines dedicated to Bayesian Inference) aims at developing hardware dedicated to probabilistic computation , which extends logic computation realised by boolean gates in current computer chips. Such probabilistic computing devices would allow to solve faster and at a lower energy cost a wide range of Artificial Intelligence applications, especially when decisions need to be taken from incomplete data in an uncertain environment. This paper describes an architecture where very simple operators compute on a time coding of probability values as stochastic signals. Simulation tests and a reconfigurable logic hardware implementation demonstrated the feasibility and performances of the proposed inference machine. Hardware results show this architecture can quickly solve Bayesian sensor fusion problems and is very efficient in terms of energy consumption.

Reconstruction of the tridimensional geometry of a visual scene using the binocular disparity information is an important issue in computer vision and mobile robotics, which can be formulated as a Bayesian inference problem. However, computation of the full disparity distribution with an advanced Bayesian model is usually an intractable problem, and proves computationally challenging even with a simple model. In this paper, we show how probabilistic hardware using distributed memory and alternate representation of data as stochastic bitstreams can solve that problem with high performance and energy efficiency. We put forward a way to express discrete probability distributions using stochastic data representations and perform Bayesian fusion using those representations, and show how that approach can be applied to diparity computation. We evaluate the system using a simulated stochastic implementation and discuss possible hardware implementations of such architectures and their potential for sensorimotor processing and robotics., Preprint of article submitted for publication in International Journal of Approximate Reasoning and accepted pending minor revisions

We present a new method for solving the problem of camera pose and calibration from a limited number of correspondences between noisy 2D and 3D features. We show that the probabilistic estimation problem can be expressed as a partially linear problem, where point and line correspondences are mixed using a common formulation. Our Sampling-Solving algorithm enables to robustly estimate the parameters and evaluate the probability distribution of the estimated parameters. It solves the problem of pose estimation with unknown focal length using a minimum of only four correspondences (five if the principal point is also unknown). To our knowledge, this is the first calibration method using so few correspondences of both points and lines. Experimental results on minimal data sets show that the algorithm is very robust to Gaussian noise. Experimental comparisons show that our method is much more stable than existing camera calibration methods for small data sets. Finally, some tests show the potential of global uncertainty estimates on real data sets.

The display of space filling data is still a challenge for the community of visualization. Direct volume rendering (DVR) is one of the most important techniques developed to achieve direct perception of such volumetric data. It is based on semitransparent representations, where the data are accumulated in a depth-dependent order. However, it produces images that may be difficult to understand, and thus several techniques have been proposed so as to improve its effectiveness, using for instance lighting models or simpler representations (e.g., maximum intensity projection). In this article, we present three perceptual studies that examine how DVR meets its goals, in either static or dynamic context. We show that a static representation is highly ambiguous, even in simple cases, but this can be counterbalanced by use of dynamic cues (i.e., motion parallax) provided that the rendering parameters are correctly tuned. In addition, perspective projections are demonstrated to provide relevant information to disambiguate depth perception in dynamic displays.

In contrast to Constructivist Views, which construe perceptual cognition as an essentially reconstructive process, this article recommends the Deictic View, which grounds perception in perceptual-demonstrative reference and the use of deictic tracking strategies for acquiring and updating knowledge about individuals. The view raises the problem of how sensory-motor tracking connects to epistemic and integrated forms of tracking. To study the strategies used to solve this problem, we report a study of the ability to track distal individuals when only their directions can be perceived and not their locations. We introduce a new experimental paradigm named the ‘Modified Traveling Salesman Problem’ (MTSP), which requires subjects to visit n invisible targets in a 2D display once each. Surprisingly, subjects are competent at this task for up to 10 targets. We consider two types of tracking strategies that subjects might use: ‘location-based’ strategies and ‘deictic direction-based’ strategies. A number of obse...

International audience; We present an architecture and a compilation toolchain for stochastic machines dedicated to Bayesian inferences. These machines are not Von Neumann and code information with stochastic bitstreams instead of using floating point representations. They only rely on stochastic arithmetic and on Gibbs sampling to perform approximate inferences. They use banks of binary random generators which capture the prior knowledge on which the inference is built. The output of the machine is devised to continuously sample the joint probability distribution of interest. While the method is explained on a simple example, we show that our machine computes a good approximation of the solution to a problem intractable in exact inference.

International audience; Bayesian inference is a powerful approach for integrating independent conflicting information for decision-making. Though an important component of robotic, biological, and other sensory-motors systems, general-purpose computers perform Bayesian inference with limited efficiency. Here we show that Bayesian inference can be efficiently performed with stochastic signals, which are particularly adapted to novel low power nano-devices that exhibit faults and device variations. A simple Muller C-element directly implements Bayes' rule. Complex inferences are performed by C-element trees, which compute the probability of an event based on multiple independent sources of evidence. A naïve Bayesian spam filter circuit is demonstrated as a pedagogical application, and design techniques for improving circuit functionality are described. Limitations of this structure are discussed in terms of signal autocorrelation. The stochastic inference structure is exceptionally robust to faults, an essential feature of decision circuits, and can therefore leverage the increased efficiency of emerging nanodevices. This hardware implementation of Bayesian inference is extremely area and power efficient, with an area-energy-delay product several orders of magnitude less than the conventional floating point implementation. These results open a pathway for a direct stochastic hardware implementation of Bayesian inference, enabling a new class of embedded decision circuits for robotics and medical applications.

When driving in curves, how do drivers use the force appearing on the steering wheel? As it carries information related to lateral acceleration, this force could be necessary for drivers to tune their internal model of vehicle dynamics; alternatively, being opposed to the drivers' efforts, it could just help them stabilize the steering wheel position. To assess these two hypotheses, we designed an experiment on a motion-based driving simulator. The steering characteristics of the vehicle were modified in the course of driving, unknown to drivers. Results obtained with standard drivers showed a surprisingly wide range of adaptation, except for exaggerated modifications of the steering force feedback. A two-level driver model, combining a preview of vehicle dynamics and a neuromuscular steering control, reproduces these experimental results qualitatively and indicates that adaptation occurs at the haptic level rather than in the internal model of vehicle dynamics. This effect is related to other theories on...

Visual object tracking is a standard function of computer vision that has been the source of numerous propositions. This diversity of approaches leads to the idea of trying to fuse them and take advantage of the strengths of each of them while controlling the noise they may introduce in some circumstances. The work presented here describes a generic framework for the combination of trackers, where fusion may occur at different levels of the processing chain. The fusion process is governed by the online detection of abnormal behavior either from specific features provided by each tracker, or from out of consensus detection. The fusion of three trackers exploiting complementary designs and features is evaluated on 72 fusion schemes. Thorough experiments on 12 standard video sequences and on a new set of 13 videos addressing typical difficulties faced by vision systems used in the demanding context of driving assistance, show that using fusion improves greatly the performance of each individual tracker, and reduces by a factor two the probability of drifting.

We studied the reliability of different measures of the visual contribution to postural steadiness by recording the postural sway during standing with eyes open (EO) or eyes closed (EC). The COP trajectory was recorded in 21 subjects aged 42-61 standing on a firm or foam support. The improvement of postural steadiness due to vision was measured with a higher reliability (i.e. lower intra- and inter-subject variabilities) with the sway velocity V, than with the position RMS. Due to the increase of the variability of V and RMS with their own mean values, we quantified the visual contribution to posture by the stabilization ratio (SR), based on a logarithm transform of V or RMS. As compared to the Romberg quotient (EC/EO), SR improved the reliability of the measurement of the visual contribution to posture within individuals, across subjects, and even across different studies in the literature. Our method led to decrease the inter-subject coefficient of variation of this measurement to about 25%, using a foam support. It leads to a similar accuracy in binocular and monocular vision, and it also applies to the quantification of other non-visual sensory contributions to posture.

Background: This study evaluated the visual contribution to postural steadiness in primary open angle glaucoma (POAG), in correlation with the mean deviation (MD) measured through conventional perimetry, and with the Advanced Glaucoma Intervention Study (AGIS) score, which quantifies the extent of losses in the visual field. Methods: In 35 POAG patients and 21 age-matched normal subjects, the sway of the centre of pressure of the feet, on a firm or foam support, was recorded. The subjects stood on a force-plate with eyes closed, or with one or two eyes open. Results: For all subjects, the sway velocity was lower with vision than without vision, indicating the existence of a visual contribution to posture at all stages of glaucoma. This contribution was significantly lower for POAG patients than for normals in monocular and binocular vision, and decreased with the MD, or as the AGIS score increased. Among the maximum, minimum and average values of the two monocular MD, the MD of the worse eye presented the most significant negative correlation with the visual contribution to posture. The somatosensory contribution to postural steadiness was larger in POAG patients, as compared to normals, in monocular or binocular vision. Conclusion: Primary open angle glaucoma induces a deficit in the visual contribution to postural steadiness, which should be taken into account for the prevention of falls.

International audience; Few studies have explored how humans memorize landmarks in complex multifloored buildings. They have observed that participants memorize an environment either by floors or by vertical columns, influenced by the learning path. However, the influence of the building's actual structure is not yet known. In order to investigate this influence, we conducted an experiment using an object-in-place protocol in a cylindrical building to contrast with previous experiments which used rectilinear environments. Two groups of 15 participants were taken on a tour with a first person perspective through a virtual cylindrical three-floored building. They followed either a route discovering floors one at a time, or a route discovering columns (by simulated lifts across floors). They then underwent a series of trials, in which they viewed a camera movement reproducing either a segment of the learning path (familiar trials), or performing a shortcut relative to the learning trajectory (novel trials). We observed that regardless of the learning path, participants better memorized the building by floors, and only participants who had discovered the building by columns also memorized it by columns. This expands on previous results obtained in a rectilinear building, where the learning path favoured the memory of its horizontal and vertical layout. Taken together, these results suggest that both learning mode and an environment's structure influence the spatial memory of complex multifloored buildings.

This article describes a computational model for the sensory perception of self-motion, considered as a compromise between sensory information and physical coherence constraints. This compromise is realized by a dynamic optimization process minimizing a set of cost functions. Measure constraints are expressed as quadratic errors between motion estimates and corresponding sensory signals, using internal models of sensor transfer functions. Coherence constraints are expressed as quadratic errors between motion estimates, and their prediction is based on internal models of the physical laws governing the corresponding physical stimuli. This general scheme leads to a straightforward representation of fundamental sensory interactions (fusion of visual and canal rotational inputs, identification of the gravity component from the otolithic input, otolithic contribution to the perception of rotations, and influence of vection on the subjective vertical). The model is tuned and assessed using a range of well-known psychophysical results, including off-vertical axis rotations and centrifuge experiments. The ability of the model to predict and help analyze new situations is illustrated by a study of the vestibular contributions to self-motion perception during automobile driving and during acceleration cueing in driving simulators. The extendable structure of the model allows for further developments and applications, by using other cost functions representing additional sensory interactions.

We investigated the ability of monocular human observer to scale absolute distance during sagittal head motion in the presence of pure optic flow information. Subjects were presented at eye-level computer-generated spheres (covered with randomly distributed dots) placed at several distances. We compared the condition of self-motion (SM) versus object-motion (OM) using equivalent optic flow field. When the amplitude of head movement was relatively constant, subjects estimated absolute distance rather accurately in both the SM and OM conditions. However, when the amplitude changed on a trial-to-trial basis, subjects' performance deteriorated only in the OM condition. We found that distance judgment in OM condition correlated strongly with optic flow divergence, and that non-visual cues served as important factors for scaling distances in SM condition. Absolute distance also seemed to be better scaled with sagittal head movement when compared with lateral head translation.

The versatility of nanodevices dynamics may allow original architectures for computation but care should be taken to handle fluctuations arising at this scale. In the network of oscillatory units which we propose, bistability with down-quiescent and up-oscillatory states enable tolerance to noise in storage and retrieval dynamics for patterns or sequences. We illustrate this by simulations of the stochastic differential equations for a network with connectivity corresponding to stored patterns.

International audience; We propose a new approach for solving a class of discrete decision making problems under uncertainty with positive cost. This issue concerns multiple and diverse fields such as engineering, economics, artificial intelligence, cognitive science and many others. Basically, an agent has to choose a single or series of actions from a set of options, without knowing for sure their consequences. Schematically, two main approaches have been followed: either the agent learns which option is the correct one to choose in a given situation by trial and error, or the agent already has some knowledge on the possible consequences of his decisions; this knowledge being generally expressed as a conditional probability distribution. In the latter case, several optimal or suboptimal methods have been proposed to exploit this uncertain knowledge in various contexts. In this work, we propose following a different approach, based on the geometric intuition of distance. More precisely, we define a goal independent quasimetric structure on the state space, taking into account both cost function and transition probability. We then compare precision and computation time with classical approaches.

International audience; Although a number of studies have been devoted to 2-D navigation, relatively little is known about how the brain encodes and recalls navigation in complex multifloored environments. Previous studies have proposed that humans preferentially memorize buildings by a set of horizontal 2-D representations. Yet this might stem from the fact that environments were also explored by floors. Here, we have investigated the effect of spatial learning on memory of a virtual multi-floored building. Two groups of 28 participants watched a computer movie that showed either a route along floors one at a time or travel between floors by simulated lifts, consisting in both cases of a 2-D trajectory in the vertical plane. To test recognition, the participants viewed a camera movement that either replicated a segment of the learning route (familiar segment) or did not (novel segment—i.e., shortcuts). Overall, floor recognition was not reliably superior to column recognition , but learning along a floor route produced a better spatial memory performance than did learning along a column route. Moreover, the participants processed familiar segments more accurately than novel ones, not only after floor learning, but crucially, also after column learning, suggesting a key role of the observation mode on the exploitation of spatial memory.

To evaluate the influence of egomotion on the three-dimensional visual processing of structure-frommotion (SFM), we compared the visual discrimination between planar and spherical surfaces during subject-translation, object-translation, or rotation of the object in depth. Performance was the best for object-rotation, intermediate for subject-translation, and the poorest for object-translation—and thus increased with the quality of retinal image stabilization achieved in the different conditions. This suggests that the major role of self-motion information was to stabilize retinal images. In view of previous results, we propose that the interactions between self-motion information and SFM are reduced to functional complementarity, in the sense that self-motion can lift visual ambiguities but does not improve the sensitivity of SFM processes.

Session: Visualization; International audience; We present a new method to visualize uncertain scalar data fields by combining color scale visualization techniques with animated, perceptually adapted Perlin noise. The parameters of the Perlin noise are controlled by the uncertainty information to produce animated patterns showing local data value and quality. In order to precisely control the perception of the noise patterns, we perform a psychophysical evaluation of contrast sensitivity thresholds for a set of Perlin noise stimuli. We validate and extend this evaluation using an existing computational model. This allows us to predict the perception of the uncertainty noise patterns for arbitrary choices of parameters. We demonstrate and discuss the efficiency and the benefits of our method with various settings, color maps and data sets.

The vestibular system detects head movement in space and maintains visual and postural stability. The semicircular canal system is responsible for registering head rotation. How it responds to head rotation is determined by the rotational axis and the angular acceleration of the head, as well as the sensitivity and orientation of each semicircular canal. The morphological parameters of the semicircular canals are supposed to allow an optimal detection of head rotations induced by some behaviours, especially locomotor. We propose a new method of semicircular canal analysis, based on the computation of central streamlines of virtually reconstructed labyrinths. This method allows us to ascertain the functional structure of the semicircular canal system and to infer its capacity to detect particular head rotations, induced by particular behaviours. In addition, this method is well-suited for datasets provided by any kind of serial sectioning methods, from MRI to μCT scanning and even mechanical serial sectioning, of extant and extinct taxa.

We evaluated the influence of long-term practice on the performance of a mental rotation task in which subjects judged whether two 3-D objects presented in different orientations were identical. Stimuli and experimental conditions were analogous to those used by Shepard and Metzler. Sixteen subjects were selected, to test the influence of aptitude for mental imagery on this learning process. Subjects participated in 12 to 15 sessions over 6 weeks. Two catalogues of different stimuli were alternatively used during three (or six) consecutive sessions to determine the influence of complexity and familiarity of figures. For all subjects, the inverse of the velocity of mental rotation along the sessions was adequately fitted by a decreasing exponential curve. However, evidence for mental rotation did not disappear, even after 15 sessions. Asymptotic variations can be attributed to differences in stimuli as well as imaging skills of subjects. Our results lead to a new interpretation of the mental rotation process.

This paper explores the possibility of using the binocular optic flow as an input for the correspondence process between stereoscopic images. The main advantage of the stereocorrespondence from optic flow (SCOF) is that it does not require the use of any a priori hypothesis concerning the 3D object under analysis. In order to determine its performance relative to noisy data, we applied an algorithm of SCOF on different rigid surfaces undertaking various 3D motions. We found that when SCOF is possible it is rather robust to noise. Moreover, the study of its domain of optimal efficiency shows that SCOF is likely to cooperate well with static stereopsis or structure from motion algorithms, thereby strengthening the processing of dynamic stereo images. As far as human vision is concerned, our psychophysical results indicate that a SCOF process does not seem to be used in the perception of 3D structure. This could be accounted for by the poor contribution of convergence signals to the perception of absolute depth in human vision, which seems incompatible with the precise knowledge of the geometry of the viewing system required by the SCOF.

Multiple views (successive or simultaneous) of the same object provide the visual system with depth information through motion or stereopsis. The retinal velocity field and binocular disparities are coded in the visual system, and may be used in an intermediate step of the visual processing of 3-D structure. Here we address the problem of recovering an intrinsic characteristic of surfaces — curvature — from moving images. We provide a theory about visual detection of surface curvatures and show how our model can be extended to stereopsis. Experimental validation of this model is brought about by our own experimental work as well as by that of others. We also discuss how spatial constancy can emerge from the visual processing of multiple views.

The question of whether neural synchrony may be preserved in adult mammalian visual cortex despite abnormal postnatal visual experience was investigated by combining anatomical and computational approaches. Single callosal axons in visual cortex of early monocularly deprived (MD) adult cats were labeled anterogradely with biocytin in vivo and reconstructed in 3D. Spike propagation was then orthodromically simulated within each of these axons with NEURON software. Data were systematically compared to those previously obtained in normally reared (NR) adult cats with comparable approaches. The architecture of the callosal axons in MD animals differed significantly from the NR group, with longer branches and first nodes located deeper below the cortex. But, surprisingly, simulation of spike propagation demonstrated that transmission latencies of most spikes remained inferior to 2 ms, like the NR group. These results indicate that synchrony of neural activity may be preserved in adult visual cortex despite abnormal postnatal visual experience. According to the temporal binding hypothesis, this also indicates that the necessary timing for visual perception is present despite anatomical abnormalities in visual cortex.

We present a new method for solving the problem of camera pose and calibration from a limited number of correspondences between noisy 2D and 3D features. We show that the probabilistic estimation problem can be expressed as a partially linear problem, where point and line correspondences are mixed using a common formulation. Our Sampling-Solving algorithm enables to robustly estimate the parameters and evaluate the probability distribution of the estimated parameters. It solves the problem of pose estimation with unknown focal length using a minimum of only four correspondences (five if the principal point is also unknown). To our knowledge, this is the first calibration method using so few correspondences of both points and lines. Experimental results show that the algorithm is very robust to Gaussian noise, even for minimal data sets. Finally, some tests show the potential of global uncertainty estimates on real data sets.

International audience; This study aims at segregating the neural substrate for the 3D-form and 3D-motion attributes in structure-from-motion perception, and at disentangling the stimulus-driven and endogenous-attention-driven processing of these attributes. Attention and stimulus were manipulated independently: participants had to detect the transitions of one attribute –form, 3D motion or colour– while the visual stimulus underwent successive transitions of all attributes. We compared the BOLD activity related to form and 3D motion in three conditions: stimulus-driven processing (unattended transitions), endogenous attentional selection (task) or both stimulus-driven processing and attentional selection (attended transitions). In all conditions, the form versus 3D-motion contrasts revealed a clear dorsal/ventral segregation. However, while the form-related activity is consistent with previously described shape-selective areas, the activity related to 3D motion does not encompass the usual “visual motion” areas, but rather corresponds to a high-level motion system, including IPL and STS areas. Second, we found a dissociation between the neural processing of unattended attributes and that involved in endogenous attentional selection. Areas selective for 3D-motion and form showed either increased activity at transitions of these respective attributes or decreased activity when subjects’ attention was directed to a competing attribute. We propose that both facilitatory and suppressive mechanisms of attribute selection are involved depending on the conditions driving this selection. Therefore, attentional selection is not limited to an increased activity in areas processing stimulus properties, and may unveil different functional localization from stimulus modulation.

In addition to the five senses usually described, vertebrate species possess a sensory organ that detects motion of the head. This organ is the vestibular system, located in the inner ear. Motion information collected by the vestibular system is crucial for equilibrium. It also contributes to stabilizing the gaze in space during head movements. Motion information provided by the vestibular system generates compensatory eye movement, a phenomenon called the Vestibulo-Ocular Reflex (VOR). The importance of this function is illustrated by the following example (from Guedry (1974)): you can look at the lines on your hand and shake your head at the same time. The VOR provides efficient gaze stabilization in this condition. In contrast, if you shake your hand, looking at the lines becomes impossible.

We use multiple sensory modalities to perceive our environment. One of these is optic flow, the displacement and deformation of the image on the retina. It is generally caused by a relative motion between an observer and the objects in the visual scene. As optic flow depends largely on three-dimensional (3D ) shapes and motions, it can be used to extract structure from motion (the sfm problem). Motion parallax and the kinetic depth effect are special cases of this phenomenon, noticed by Von Helmholtz (1867), and experimentally quantified by Wallach and O’Connell (1953).

International audience; This study investigated how the visual system detects a surface deviation from planar, induced by, crossed or uncrossed, horizontal disparities continuously increasing with eccentricity. Binocular disparities increased linearly and concentrically, between two given eccentricities. The thresholds of deformation detection were gathered using a method in which observers halted a dynamic stimulus. The thresholds were substantially higher than those measured by the control experiment using a method of constant stimuli. Results, using the adjustment method, highlighted lower discrimination thresholds for uncrossed disparities than for crossed disparities. For all the directions of disparity, thresholds varied similarly as a function of eccentricity, however two observations could be pointed out: thresholds of peripheral start depended on disparity gradient and starting eccentricity; fovea start thresholds did not depend on disparity gradient alone. Data suggests that, in peripheral field, the visual system was more sensitive to uncrossed disparities than crossed disparities, relative to the frontoparallel plane. According to a verbal report from observers, the reference used for the perceptive judgment appears not to be the screen plane but rather the peripheral stimulus. Moreover, in the deformation detection of planar surfaces, horizontal disparities processing depends on the eccentric location of the disparities. It could be global for the peripheral locations and could be based more on depth contrast for the central locations.

Human observers can perceive the three- dimensional (3-D) structure of their environment using various cues, an important one of which is optic flow. The motion of any point’s projection on the retina depends both on the point’s movement in space and on its distance from the eye. Therefore, retinal motion can be used to extract the 3-D structure of the environment and the shape of objects, in a process known as structure-from-motion (SFM). However, because many combinations of 3-D structure and motion can lead to the same optic flow, SFM is an ill-posed inverse problem. The rigidity hypothesis is a constraint supposed to formally solve the SFM problem and to account for human performance. Recently, however, a number of psychophysical results, with both moving and stationary human observers, have shown that the rigidity hypothesis alone cannot account for human performance in SFM tasks, but no model is known to account for the new results. Here, we construct a Bayesian model of SFM based mainly on one new hypothesis, that of stationarity, coupled with the rigidity hypothesis. The predictions of the model, calculated using a new and powerful methodology called Bayesian programming, account for a wide variety of experimental findings.

Human observers can perceive the threedimensional (3-D) structure of their environment using various cues, an important one of which is motion parallax. The motion of any points projection on the retina depends both on the points movement in space and on its distance from the eye. Therefore, retinal motion can be used to extract the 3-D structure of the environment and the shape of objects, in a process known as structurefrom- motion (sfm). However, because many combinations of 3-D structure and motion can lead to the same optic flow, sfm is an ill-posed inverse problem. The rigidity assumption is a constraint supposed to formally solve the sfm problem and to account for human performance. Recently, however, a number of psychophysical results, in both moving and stationary human observers, have shown that the rigidity assumption alone cannot account for sfm, but no model is known to account for the new results. Here, we construct a Bayesian model of sfm based on only one new assumption, that of stationarity, coupled with the assumption of rigidity. The predictions of the model, calculated using a new and powerful methodology called Bayesian programming, account for a wide variety of experimental findings.

Subjects indicated the tilt of dotted planes rotating in depth, in monocular viewing, under perspective projection. The responses depended on the FOV (field of view) and on the angle W between the tilt and frontal translation (orthogonal to the rotation axis). Response accuracy increased with the FOV, and decreased with W. Our results support the processing of the second-order optic flow in all cases, but indicate that this flow is quantitatively small in small-field, leading to tilt ambiguities. We examine computational models based on the affine components of the optic flow to interpret our results.

In this study, we evaluated the capacity of human beings to perceive and reproduce forces applied to the hand. We tested for perceptive distortions and/or privileged directions in the performance of these two tasks. Subjects resisted a reference force applied by a joystick in a given direction, with instructions to keep the hand at a constant position. In a perception task, subjects subsequently resisted a second such force, the direction of which they could adjust with a potentiometer; the task was to reorient the second force to be in the same perceived direction as the reference. In a reproduction task, subjects were instructed to push against the now elastically constrained joystick with the same force that was required to resist the initially applied reference force. Twenty-four reference force directions in the horizontal plane were tested twice each. We observed systematic distortions in the reproduction of force direction that were not present in the perception task. We further observed that the distortions could be predicted by anisotropy of limb stiffness and could be affected by manipulating the mechanical impedance of the hand-joystick interaction. We conclude that human subjects specify and store forces to be applied by the hand not in terms of a perceived force vector, but rather in terms of the motor activity required to resist or produce the force—i.e., subjects possess a multi-dimensional “sense of effort.”

This study investigates a new experimental paradigm called the Modified Traveling Salesman Problem (Bullot & Droulez, submitted). This task requires subjects to visit once and only once n invisible targets in a 2D display, using a virtual vehicle controlled by the subject. Subjects can only see the directions of the targets from the current location of the vehicle, displayed by a set of oriented segments that can be viewed inside a circular window surrounding the vehicle. Two conditions were compared. In the “allocentric” condition, subjects see the vehicle move across the screen and change orientation under their command. The “egocentric” condition is similar except for how the information is provided: the position and orientation of the vehicle icon remains fixed at the center of the screen and only target directions, as indicated by the oriented segments, change as the subject “moves” the vehicle. The unexpected finding was that this task can be performed, in either condition, for up to 10 targets. We consider two possible strategies that might be used, a location-based strategy and a segment strategy. The location-based strategy relies on spatial memory and attempts to infer the locations of all the targets. The segment strategy is more local and focuses on the directional segments themselves, keeping track of the ones that represent already-visited targets. A number of observations suggest that the segment strategy was used, at least for larger numbers of targets. According to our hypothesis, keeping track of the segments requires one to use indexical reference for associating the segments with their status in the task - given by current status predicates Visited(x) or Not-visited(x) -, perhaps using visual indexes (Pylyshyn, 2001), deictic pointers (Ballard et al., 1997), or object files (Kahneman et al, 1992).

Postures and chimereae, or philosophy in a bathroom. A plea for " representations " in neurosciences exemplified by an analysis of postural control and based on the specificity of neurofunctional structures as produced by the brain independently of represented physical variables., Illustré par l'analyse du contrôle postural dans la vie quotidienne, un plaidoyer pour «les représentations » en neurosciences fondé sur la spécificité des architectures neuro-fonctionnelles, en tant qu'élaborations du cerveau autonomes par rapport aux grandeurs physiques représentées., Vidal Pierre Paul, Droulez Jacques. 2.3. Postures et chimères, ou la philosophie dans la salle de bain. In: Intellectica. Revue de l'Association pour la Recherche Cognitive, n°36-37, 2003/1-2. Repenser le corps l'action et la cognition avec les neurosciences, sous la direction de Jean-Luc Petit. pp. 161-180.

We investigated the role of extraretinal information in the perception of absolute distance. In a computersimulated environment, monocular observers judged the distance of objects positioned at different locations in depth while performing frontoparallel movements of the head. The objects were spheres covered with random dots subtending three different visual angles. Observers viewed the objects at eye level, either in isolation or superimposed on a ground floor. The distance and size of the spheres were covaried to suppress relative size information. Hence, the main cues to distance were the motion parallax and the extraretinal signals. In three experiments, we found evidence that (1) perceived distance is correlated with simulated distance in terms of precision and accuracy, (2) the accuracy in the distance estimate is slightly improved by the presence of a ground-floor surface, (3) the perceived distance is not altered significantly when the visual field size increases, and (4) the absolute distance is estimated correctly during self-motion. Conversely, stationary subjects failed to report absolute distance when they passively observed a moving object producing the same retinal stimulation, unless they could rely on knowledge of the three-dimensional movements.

Experimental studies show that automobile drivers adjust their speed in curves so that maximum vehicle lateral accelerations decrease at high speeds. This pattern of lateral accelerations is described by a new driver model, assuming drivers control a variable safety margin of perceived lateral acceleration according to their anticipated steering deviations. Compared with a minimum time-to-lane-crossing (H. Godthelp, 1986) speed modulation strategy, this model, based on nonvisual cues, predicts that extreme values of lateral acceleration in curves decrease quadratically with speed, in accordance with experimental data obtained in a vehicle driven on a test track and in a motion-based driving simulator. Variations of model parameters can characterize "normal" or "fast" driving styles on the test track. On the simulator, it was found that the upper limits of lateral acceleration decreased less steeply when the motion cuing system was deactivated, although drivers maintained a consistent driving style. This is interpreted per the model as an underestimation of curvilinear speed due to the lack of inertial stimuli. Actual or potential applications of this research include a method to assess driving simulators as well as to identify driving styles for on-board driver aid systems