A self-interpretable module for deep image classification on small data

Deep neural networks are the driving force of the recent explosion of machine learning applications in everyday life. However, they usually require a lot of training data to work well, and they act as black-boxes, making predictions without any explanation about them. This paper presents Memory Wrap, a module (i.e, a set of layers) that can be added to deep learning models to improve their performance and interpretability in settings where few data are available. Memory Wrap adopts a sparse content-attention mechanism between the input and some memories of past training samples. We show that adding Memory Wrap to standard deep neural networks improves their performance when they learn from a limited set of data, and allows them to reach comparable performance when they learn from the full dataset. We discuss how the analysis of its structure and content-attention weights helps to get insights about its decision process and makes their predictions more interpretable, compared to the same networks without Memory Wrap. We test our approach on image classification tasks using several networks on three different datasets, namely CIFAR10, SVHN, and CINIC10.

Skin Lesion Area Segmentation Using Attention Squeeze U-Net for Embedded Devices

Melanoma is the deadliest form of skin cancer. Early diagnosis of malignant lesions is crucial for reducing mortality. The use of deep learning techniques on dermoscopic images can help in keeping track of the change over time in the appearance of the lesion, which is an important factor for detecting malignant lesions. In this paper, we present a deep learning architecture called Attention Squeeze U-Net for skin lesion area segmentation specifically designed for embedded devices. The main goal is to increase the patient empowerment through the adoption of deep learning algorithms that can run locally on smartphones or low cost embedded devices. This can be the basis to (1) create a history of the lesion, (2) reduce patient visits to the hospital, and (3) protect the privacy of the users. Quantitative results on publicly available data demonstrate that it is possible to achieve good segmentation results even with a compact model.

A Methodology to Design and Evaluate HRI Teaming Tasks in Robotic Competitions

As social robots become more prominent in our lives, their interaction with humans takes an increasing role, and new collaborative scenarios emerge. This development brings the need to realize robust test methods enabling the design and evaluation of HRI teaming tasks to prove functionality and promote adoption. In this paper, we present a general purpose and repeatable methodology for conducting studies in collaborative HRI in the range of robotic competitions. The methodology includes a step-by-step approach to design HRI teaming tasks tailored to be enacted in a robotic competition and to evaluate the performance of social robots to execute the designed tasks, exploring the relationship between robots’ performance and user perceptions based on the feedback of the users participating to such tasks. We assess the feasibility of the methodology to design and evaluate a HRI teaming task in the context of SciRoc (“Smart CIties RObotics Challenges”) competition, which targets at investigating the impact of social of robots in smart cities.

The Psychological Implications of Companion Robots: A Theoretical Framework and an Experimental Setup

In this paper we present a theoretical framework to understand the underlying psychological mechanism involved in human-Companion Robot interactions. At first, we take the case of Sexual Robotics, where the psychological dynamics are more evident, to thereafter extend the discussion to Companion Robotics in general. First, we discuss the differences between a sex-toy and a Sexual Robots, concluding that the latter may establish a collusive and confirmative dynamics with the user. We claim that the collusiveness leads to two main consequences, such as the fixation on a specific and atypical type of sexual interaction, called paraphilic, and to the infantilization of the user, which we explain through the theoretical framework of "object-relation theory". We argue that these dynamics may degrade to an infantile stage the relational abilities of users, extending this argument to Companion Robots in general. Then, we enquire if and how the relational dynamics enacted in HRI may shift to human relations: we discuss the analogy with virtual reality concluding that, under certain condition, a symbolic shift might happen. In the last part of this work, we propose an experimental setup to verify if a collusive and confirmative interaction with a Companion Robot can, over time, impact on the user's ability to manage relational frustration.

Age-associated decline in septum neuronal activation during spatial learning in homing pigeons (Columba livia)

The relationship between hippocampal aging and spatial-cognitive decline in birds has recently been investigated. However, like its mammalian counterpart, the avian hippocampus does not work in isolation and its relationship to the septum is of particular interest. The current study aimed to investigate the effects of age on septum (medial and lateral) and associated nucleus of the diagonal band (NDB) neuronal activation (as indicated by c-Fos expression) during learning of a spatial, delayed non-match-to-sample task conducted in a modified radial arm maze. The results indicated significantly reduced septum, but not NDB, activation during spatial learning in older pigeons. We also preliminarily investigated the effect of age on the number of cholinergic septum and NDB neurons (as indicated by expression of choline acetyltransferase; ChAT). Although underpowered to reveal a statistical effect, the data suggest that older pigeons have substantially fewer ChAT-expressing cells in the septum compared to younger pigeons. The data support the hypothesis that reduced activation of the septum contributes to the age-related, spatial cognitive impairment in pigeons.

Spatial reorientation with a geometric array of auditory cues

A visuocentric bias has dominated the literature on spatial navigation and reorientation. Studies on visually accessed environments indicate that, during reorientation, human and non-human animals encode the geometric shape of the environment, even if this information is unnecessary and insufficient for the task. In an attempt to extend our limited knowledge on the similarities and differences between visual and non-visual navigation, here we examined whether the same phenomenon would be observed during auditory-guided reorientation. Provided with a rectangular array of four distinct auditory landmarks, blindfolded, sighted participants had to learn the location of a target object situated on a panel of an octagonal arena. Subsequent test trials were administered to understand how the task was acquired. Crucially, in a condition in which the auditory cues were indistinguishable (same sound sample), participants could still identify the correct target location, suggesting that the rectangular array of auditory landmarks was encoded as a geometric configuration. This is the first evidence of incidental encoding of geometric information with auditory cues and, consistent with the theory of functional equivalence, it supports the generalisation of mechanisms of spatial learning across encoding modalities.

Local geometric properties do not support reorientation in hippocampus-engaged homing pigeons

It is generally accepted that the geometry of an environment is a reliable source of information for spatial navigation used by most vertebrate species. However, there is a continuing debate on which geometrical properties of space are the ones that matter for reorientation. In this study, pigeons were trained to find a food reward hidden in 2 opposite corners in a rectangular arena. The animals were then tested in a kite-shaped environment similar to Pearce, Good, Jones, and McGregor (2004). We found that pigeons, unlike rats, were not able to identify the correct corner in the kite arena even though elements clearly preserved the correct long wall-short wall geometric configuration and the local aspect of the trained goal. This behavioral study was followed by a c-Fos, IEG analysis of brain activation that contrasted pigeons exposed to the trained, familiar rectangular environment with pigeons that were exposed to an unfamiliar, trapezoid arena. The hippocampal formation (HF) displayed greater c-Fos expression in the animals exposed to the familiar, training arena, which further supports the conclusion that pigeons do not substantially rely on local geometric features for reorientation. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

No environmental context-dependent effect, but interference, of physical activity on object location memory

Research on context-dependent memory has addressed many external and internal types of contexts. However, whether the physical activity engaged in at the time of encoding and recall can act as an environmental context cue has been systematically investigated only in one study. The purpose of the present study was to replicate this; furthermore, given the effect of physical activity/effort on the way space is represented, we sought to extend the findings to object location memory. Using a 1-list paradigm (Experiment 1) and a 2-list paradigm (Experiment 2), participants had to learn the locations of objects on a grid and then recall them, while standing or walking on a health walker. No evidence of activity context effects was found. However, an interference effect of the motor task on location memory was detected, such that participants' performance was worse when walking, compared to standing, at encoding (Experiment 2) or recall (Experiment 1). Results are discussed based on the outshining hypothesis and the possible link between motor task and object location memory.

Spatial reorientation with non-visual cues: Failure to spontaneously use auditory information

Among the environmental stimuli that can guide navigation in space, most attention has been dedicated to visual information. The process of determining where you are and which direction you are facing (called reorientation) has been extensively examined by providing the navigator with two sources of information—typically the shape of the environment and its features—with an interest in the extent to which they are used. Similar questions with non-visual cues are lacking. Here, blindfolded sighted participants had to learn the location of a target in a real-world, circular search space. In Experiment 1, two ecologically relevant non-visual cues were provided: the slope of the floor and an array of two identical auditory landmarks. Slope successfully guided behaviour, suggesting that proprioceptive/kinesthetic access is sufficient to navigate on a slanted environment. However, despite the fact that participants could localise the auditory sources, this information was not encoded. In Experiment 2, the auditory cue was made more useful for the task because it had greater predictive value and there were no competing spatial cues. Nonetheless, again, the auditory landmark was not encoded. Finally, in Experiment 3, after being prompted, participants were able to reorient by using the auditory landmark. Overall, participants failed to spontaneously rely on the auditory cue, regardless of how informative it was.

Sex differences and errors in the use of terrain slope for navigation

Unlike most of the spatial cues that have received attention, a sloping terrain can be perceived by multimodal sensory inputs (vision, balance, and kinesthesia), making it potentially very salient for navigation. Furthermore, a homogeneous slope can be used like a compass to identify directions (e.g., uphill, downhill, and sideways), but not to determine distances. We briefly review recent evidence on navigation with slope, emphasizing two main findings. On the one hand, we focus on the conspicuous sex difference found in the ability to localize a target in a square, tilted enclosure; this has emerged in human adults and children, and we suggest that it is related to lower awareness of the slope for females. On the other hand, we describe the general pattern of errors that arises when localizing the target during the task; these errors indicate the use of a bi-coordinate representation of the slope. Limitations and ideas for future studies are proposed.

RoboCup Soccer Leagues

RoboCup was created in 1996 by a group of Japanese, American, and European artificial intelligence and robotics researchers with a formidable, visionary long-term challenge: By 2050 a team of robot soccer players will beat the human World Cup champion team. In this article, we focus on RoboCup robot soccer, and present its five current leagues, which address complementary scientific challenges through different robot and physical setups. Full details on the status of the RoboCup soccer leagues, including league history and past results, upcoming competitions, and detailed rules and specifications are available from the league homepages and wikis.

Where is uphill? Exploring sex differences when reorienting on a sloped environment presented through 2-D images

One of the spatial abilities that has recently revealed a remarkable variability in performance is that of using terrain slope to reorient. Previous studies have shown a very large disadvantage for females when the slope of the floor is the only information useful for encoding a goal location. However, the source of this sex difference is still unclear. The slope of the environment provides a directional source of information that is perceived through dissociable visual and kinesthetic sensory modalities. Here we focused on the visual information, and examined whether there are sex differences in the perception of a slope presented through 2-D images with a desktop computer connected to an eye-tracking device. Participants had to identify and point to the uphill direction by looking at different orientations of two virtual, slanted environments (one indoor and one outdoor). Men were quicker and more accurate than women, indicating that the female difficulty with slope emerges at an early, unisensory, perceptual level. However, the eye-tracking data revealed no sex differences in the slope cues used, providing no support to the hypothesis of sex-specific, visual-processing strategies. Interestingly, performance correlated with a test of mental rotation, and we speculate that the disadvantage in mental rotation ability might be an important factor responsible for females' difficulty using slope.

Up by upwest: Is slope like north?

Terrain slope can be used to encode the location of a goal. However, this directional information may be encoded using a conceptual north (i.e., invariantly with respect to the environment), or in an observer-relative fashion (i.e., varying depending on the direction one faces when learning the goal). This study examines which representation is used, whether the sensory modality in which slope is encoded (visual, kinaesthetic, or both) influences representations, and whether use of slope varies for men and women. In a square room, with a sloped floor explicitly pointed out as the only useful cue, participants encoded the corner in which a goal was hidden. Without direct sensory access to slope cues, participants used a dial to point to the goal. For each trial, the goal was hidden uphill or downhill, and the participants were informed whether they faced uphill or downhill when pointing. In support of observer-relative representations, participants pointed more accurately and quickly when facing concordantly with the hiding position. There was no effect of sensory modality, providing support for functional equivalence. Sex did not interact with the findings on modality or reference frame, but spatial measures correlated with success on the slope task differently for each sex.

Hippocampal size predicts rapid learning of a cognitive map in humans

The idea that humans use flexible map-like representations of their environment to guide spatial navigation has a long and controversial history. One reason for this enduring controversy might be that individuals vary considerably in their ability to form and utilize cognitive maps. Here we investigate the behavioral and neuroanatomical signatures of these individual differences. Participants learned an unfamiliar campus environment over a period of three weeks. In their first visit, they learned the position of different buildings along two routes in separate areas of the campus. During the following weeks, they learned these routes for a second and third time, along with two paths that connected both areas of the campus. Behavioral assessments after each learning session indicated that subjects formed a coherent representation of the spatial structure of the entire campus after learning a single connecting path. Volumetric analyses of structural MRI data and voxel-based morphometry (VBM) indicated that the size of the right posterior hippocampus predicted the ability to use this spatial knowledge to make inferences about the relative positions of different buildings on the campus. An inverse relationship between gray matter volume and performance was observed in the caudate. These results suggest that (i) humans can rapidly acquire cognitive maps of large-scale environments and (ii) individual differences in hippocampal anatomy may provide the neuroanatomical substrate for individual differences in the ability to learn and flexibly use these cognitive maps. © 2013 Wiley Periodicals, Inc.

Use of slope and feature cues in pigeon (Columba livia) goal-searching behavior

Terrain slope provides a directional frame of reference for reorientation and navigation, similar to cardinal directions. Previous studies have shown that, in a goal location task, slope is a very salient cue and that pigeons tend to rely on it even if it is not the most informative cue. Such a strong dependence on one type of information, when there are more effective predictors of reward, is a key premise for a modular view of information processing. Here we tested the provocative hypothesis of a "slope module" for reorientation in slanted environments. Pigeons had to solve a goal location task using slope or another, theoretically salient cue: a beacon feature. Overall, searching behavior was controlled almost equally by the two cues. The fact that, for the first time, slope failed to capture most of the associative strength allows us to reject a strong modularity view and suggests instead that there is competition between cues based on salience. As an interesting additional finding, the reliance on slope and the feature was affected by training location (uphill vs. downhill), suggesting the possibility of a modulatory role of effort on the cue-weighting mechanism of reorientation.

Pigeon (Columba livia) encoding of a goal location: the relative importance of shape geometry and slope information

The ability to use the geometric shape of an environment as an orienting cue for goal location has been shown in many vertebrate groups. Experimentally, however, geometric spatial tasks are typically carried out on horizontal surfaces. The present study explored how learning a geometry task is affected by training on a surface extending in the vertical dimension-a slope. In a reference memory task, pigeons (Columba livia) were trained to locate a goal in an isosceles trapezoid arena. Learning on a slope proceeded more rapidly or with fewer errors than on a flat surface, presumably because of kinesthetic, vestibular, and visual information extractable from an inclined surface. Experiment 1 showed that, although the geometric shape of the arena was encoded, pigeons trained on a slope were guided by a goal representation based on the vertical and orthogonal axes of the slope to solve the task. Experiment 2 revealed that geometric learning was neither overshadowed nor facilitated by training on a slope. The data highlight a potentially important role for slope as an allocentric cue for goal location.

Asymmetrical participation of the left and right hippocampus for representing environmental geometry in homing pigeons

Control, right and left HF lesioned homing pigeons (Columba livia) were trained to locate a goal in one corner of a rectangular enclosure with a distinctive feature cue. Probe tests revealed that all groups were able to encode in parallel geometric (enclosure shape) and feature information, and in the absence of one of them, they could us the other to locate the goal. However, left HF lesioned pigeons learned the task at a faster rate, and when the geometric and feature information were set in conflict, they relied more on the feature cue compared to control and right HF lesioned pigeons. It was also found that pigeons, independent of group, trained to a goal adjacent to the feature cue learned the task in fewer sessions and relied more on feature information compared to pigeons trained to a goal opposite the feature cue. The latter group relied more on geometric information. The results support the hypothesis that the left HF plays a more important role in the representation of a goal location with respect to environmental shape/geometry. We further propose that the observed functional asymmetry can be explained by the lateralized properties of the pigeon tectofugal visual system.

A lateralized avian hippocampus: preferential role of the left hippocampal formation in homing pigeon sun compass-based spatial learning

The hippocampal formation (HF) plays a crucial role in amniote spatial cognition. There are also indications of functional lateralization in the contribution of the left and right HF in processes that enable birds to navigate space. The experiments described in this study were designed to examine left and right HF differences in a task of sun compass-based spatial learning in homing pigeons (Columba livia). Control, left (HFL) and right (HFR) HF lesioned pigeons were trained in an outdoor arena to locate a food reward using their sun compass in the presence or absence of alternative feature cues. Subsequent to training, the pigeons were subjected to test sessions to determine if they learned to represent the goal location with their sun compass and the relative importance of the sun compass vs. feature cues. Under all test conditions, the control pigeons demonstrated preferential use of the sun compass in locating the goal. By contrast, the HFL pigeons demonstrated no ability to locate the goal by the sun compass but an ability to use the feature cues. The behaviour of the HFR pigeons demonstrated that an intact left HF is sufficient to support sun compass-based learning, but in conflict situations and in contrast to controls, they often relied on feature cues. In conclusion, only the left HF is capable of supporting sun compass-based learning. However, preferential use of the sun compass for learning requires an intact right HF. The data support the hypothesis that the left and right HF make different but complementary contributions toward avian spatial cognition.

Multirobot Systems: A Classification Focused on Coordination

Multirobot systems (MRS) are, nowadays, an important research area within robotics and artificial intelligence and a growing number of systems have recently been presented in the literature. Since application domains and tasks that are faced by MRS are of increasing complexity, the ability of the robots to cooperate can be regarded as a fundamental feature. In this paper, we present a survey of the recent work in the area by specifically examining the forms of cooperation and coordination realized in the MRS. In particular, we propose a new taxonomy for classification of the approaches to coordination in MRS and we describe some systems, which we consider representative in our taxonomy. We finally discuss the outcomes of our analysis and try to highlight future trends of the research on MRS.