Mechanisms of attention in touch

New IROS 2014 paper about touch attention mechanisms in robotic systems. Read more here.

Some notes about attention mechanisms in Humans extracted from the work "Mechanisms of attention in touch" by Lloyd DM, Bolanowski SJ Jr, Howard L, McGlone F.

Introduction: A touch to a body surface will trigger an alerting response and subsequent rapid orientation of the eyes and head towards the stimulated site in order to facilitate an appropriate response (Groh and Sparks, 1996a). However, little is known about the attentional mechanisms operating solely within the tactile modality. Recent studies have demonstrated ipsilateral Reaction Time (RT) inhibition (inhibition of return Ð IOR) to somatic non-informative cues, but have explained this in terms of the reduction of motor readiness at the cued location (Tassinari and Berlucchi, 1995; Tassinari and Campara, 1996). The following study proposes an alternative explanation of these effects by exploiting the model developed by Posner (1980), who described two ways in which attention can be oriented to a potential source of perceptual input, i.e., exogenously or endogenously. In the former, attention is automatically oriented by a sudden onset stimulus, whereas in the latter, attention is under the strategic control of the subject. The presence of an equivalent system in the attentional control of somatosensation is as yet unestablished. The present study aimed to investigate whether similar sampling strategies are employed for touch."

Current understanding of tactile attention mechanisms comes mainly from vibrotactile studies in which subjects had to detect small changes in the intensity of the stimulus (Whang et al., 1991; Evans and Craig, 1992). Interference effects, where the presentation of a non-target stimulus to one fingerpad interferes with the identification of a target stimulus presented to a second fingerpad, can also be used to demonstrate a failure of selective attention (Craig and Evans, 1995). Driver and Grossenbacher (1996) also show that interference effects are not just determined by the sensory surface stimulated, but are influenced by higher level frames of reference. When the task is to respond to a target vibration on one finger, distractors on a homologous finger on the other hand have greater effects when the fingers are close together. As the fingers are moved apart the interference effect declines because the information is represented in action space, not by where in the somatosensory cortex each finger projects. Studies examining the effects of attention on the somatosensory system have used somatosensory evoked potentials (SEPs) to demonstrate orienting of attention by an exogenous cue (Garcia-Larrea et al., 1991). In a related study, Bruyant et al. (1993) showed that ª odd ballº stimuli evoked the distinctive P300a response even when presented to the unattended hand, suggesting that unusual stimuli are automatically analysed by the somatosensory systems. Clearly, much perceptual analysis will take place automatically without the subjects’ orienting attention to the stimulus, or even being consciously aware of its presence. However, some forms of perceptual analysis will only be possible when the subject actively orients attention to the source of sensory stimulation. Tactile alerting responses are beyond a subject’s control, i.e., they cannot prevent attention moving to that area (exogenous), or alternatively, subjects can move attention to particular areas of skin in a voluntary controlled manner (endogenous).

In studies of exogenous (automatic) orienting of visual attention, detection of a target in the cued location is facilitated at short intervals, whereas over longer periods, detection of a stimulus at the previously attended location is inhibited. One of the major functions of visual attention is to ensure that environments are searched efficiently for objects that are of behavioural relevance to the organism. A crucial mechanism that enables efficient search is the IOR of attention (Posner and Cohen, 1984). IOR elicited by visual stimuli is associated with the saccade-generating systems of the superior colliculus (Kalesnykas and Sparks, 1996; Groh and Sparks, 1996b).

In studies of visual endogenous orienting facilitation effects are normally obtained, i.e., at short stimulus onset asynchronies (SOAs) either no effect, or a small facilitation is observed. At longer SOAs the facilitation is larger. This increase reflects the time course of endogenous attention, i.e., it takes some finite interval for strategic attentional processes to be moved from one locus to another. Therefore, longer intervals between the cue and the target make it more likely that attention has arrived at the cued location, and, hence, larger facilitation is observed.

Bradshaw et al. (1988) demonstrated that orienting the eyes to a tactually stimulated body site (the hand) facilitated subsequent detection in a simple RT task, but only when the arms were crossed over the body midline requiring visual input in order to locate relative limb position in external space (Pierson et al., 1991). This facilitation occurs even when the subjects cannot see the normally visible body site being stimulated. Driver and Grossenbacher (1996) found that incongruent tactile distractors (i.e., different from the target) impaired tactile discrimination performance in all their experiments, but that this impairment declined when the hands were spatially separated and the subjects were looking at the relevant hand or a neutral midline position. These results demonstrate contributions from upper-limb proprioception to spatially selective tactile attention and suggest a role for head and eye orientation in the spatial selection of even unseen tactile stimuli.

Honore et al. (1989) also demonstrated that directing the eyes towards the same source as the tactile stimulus yields an additional benefit even when sight itself is not involved (i.e., in a darkened room). Most recently, Tipper et al. (1998) demonstrated that the perception of tactile stimuli is facilitated when subjects look towards the stimulated body site, and, indeed, vision of a body site, independent of proprioceptive orienting, could effect somatosensation.

We will also show that attention can be facilitated when subjects are instructed to visually orient to the stimulated site. Graziano and Gross (1996) have identified neurones in frontal cortex and the basal ganglia which have both visual and tactile receptive fields. The visual receptive field associated with a region of the body (such as the hand) moves with that body part, showing how integral visual and somatosensory interaction can be for guiding action. More recently, the flexibility of these bi-modal cells has been demonstrated by Iriki et al. (1996) who showed that when a monkey used a tool to reach for a food reward, the visual receptive field expanded around the full

Discussion: It is evident from the results that the attentional mechanisms operating in the tactile system are, in some ways similar to those observed with the visual system. Both exogenous automatic orienting and endogenous strategic orienting mechanisms exist. Even though the input to the somatosensory system remains constant in terms of cues and targets, the systematic varying of proportions and information provided to the subjects results in qualitatively different patterns of data.

Evidence for exogenous orienting: These experiments demonstrate the phenomenon of IOR in the tactile system. The effect is observed at a wide range of intervals from 100 to 1200 ms. The inhibition increases in magnitude between 200 and 700 ms, indicating that the effect is not produced by low level masking processes (Bolanowski et al., 1988). However, in order to support this, a series of experiments was carried out to test whether the initial cue was capable of peripherally masking the target vibration in order to cause the effects of inhibition observed in these results. Cue intensities of a 50 ms linear ramp stimulus and a 70 ms 200 Hz sine wave were presented in a counterbalanced blocked design. If the inhibition observed was due to low level masking, it would increase in size with a more intense cue stimulus.

The results showed no main effect of SOA (F(1 ,11) = 1.37), but the main effect of cueing (F(1 ,11) = 51.13, p

Evidence for endogenous orienting: The are two important features of these data. First, in sharp contrast to experiment 1a, facilitation was observed at the short SOA where there had been inhibition with exogenous cues. This contrast is a potential marker for the role of higher level attentional processes influencing low level perception. Second, the facilitation appears to have dissipated and to have been replaced by inhibition at the longer SOA of 700 ms. So, although the contrast between exogenous and endogenous orienting was as predicted from visual attention studies at short SOAs, the effects at long SOAs were not, therefore indicating sensory modality specificity. An interpretation would be that in these particular procedures, the subjects are unable to maintain attention at the cued location for long periods. By 700 ms attention has shifted away from the cued locus, as demonstrated by IOR. The similar inhibition effects at the longer SOA whether the subjects ignore the cues or actively try to maintain attention upon them, demonstrates strong (p

Evidence for visual orienting: The effects of visual orienting to the cued hand have consistently shown that, where previously, inhibition had been observed at the longer SOAs (for both exogenous and endogenous orienting), facilitation effects have now been obtained. The reason for this may be that if inhibition of saccades to a body site can cause inhibition of subsequent tactile stimuli presented to that body site, then removal of the saccade inhibition may produce the facilitation effects. It may be that the saccade system is rapidly activated but cannot be maintained, and if not released quickly, the eye movement is inhibited. In our task the subjects were required to maintain fixation at a central locus, thus, after cueing, which may trigger a saccade to the stimulated body site, the saccade is suppressed. Therefore, this saccade inhibition may mediate our effects. An interaction between visual orienting and tactile attention is, however, clearly demonstrated, illustrating that visual orienting to a body site is sufficient to facilitate the focusing and maintenance of attention at that site.

The previous results also show that, where purely strategic tactile orienting failed to facilitate RTs, visual orienting to the stimulated site facilitated all RTs, demonstrating again an improvement in somatosensory perception derived from visual input. The evidence of excitatory links between the spatiotopic maps of the different sensory modalities could provide a speculative explanation of the source of the facilitation effects seen in this study. When looking towards a particular locus, activity in somatosensory systems responsible for the eye/head orienting, and visual maps receiving visual input from that location, may project excitatory signals to other modalities, in this case the tactile. This encoding of information from the same spatial source may lead to the speeded processing of tactile information observed in this study.

Somatosensory attention - Humans vs Robots

New IROS 2014 paper about touch attention mechanisms in robotic systems. Read more here.

Some notes on attention in sense of touch by Matthias M. Müller and Claire-Marie Giabbiconi. Full-text Springer

Attention in somatosensation

In everyday life the human brain is confronted with an enormous amount of sensory input at any given moment. To guarantee coherent and adaptive behaviour, selective attention is needed to focus the limited processing resources on the relevant part of the available information while ignoring the rest [1, 2]. This chapter provides an overview on some topics of current research inattention and the sense of touch. We will mainly focus on studies using mechanical stimuli rather than electrical stimulation. The physical characteristics of electrical stimuli (sharp and short duration of only a fraction of a millisecond) makes them closer to pain stimuli and, thus, suboptimal to mimic the complex interactions between different mechanoreceptors of the glabrous skin [3–5]. In contrast, mechanical stimuli have not such a sharp onset (mostly they are delivered in form of a sinusoid, see below) and make contact with the skin for much longer. Therefore, mechanical stimuli seem to mimic everyday experience of touch more closely as opposed to electrical stimuli. A second aim of this chapter is to discuss possible neuronal mechanisms that explain how to-be-attended tactile stimuli are processed preferentially in the brain.

Research on attention has a long history and definitions of attention varied from time to time.In 1890, William James [6] wrote: “my experience is what I agree to attend to”. Along the same line,John Driver defined attention as “a generic term for those mechanisms, which lead our experience to be dominated by one thing rather than another”[7]. The main behavioural signature of attention is the improved accuracy in analysing and speeded detection or discrimination of attended stimuli.

But what makes somatosensation so different from other senses to look exclusively into the impact of attention onto that sensory system? Hsiao and Vega-Bermudez [8] nicely illustrated that point with the following example:“if you switch your focus of attention to your foot, you immediately become conscious of sensations arising from receptors in your foot that were non-existent a moment earlier. This simple observation demonstrates the power of selective attention”.

Based on that illustration the authors concluded that attention in somatosensation plays its role in the selection of specific sensory inputs at certain body locations. While this isnot particularly different from other modalities when we focus on spatial accounts of attention,Forster and Eimer [9], among others, discussed an important difference compared to other sen-sory modalities. Contrary to vision for instance,in touch one interacts with proximal stimuli impinging on our body surface. In other words,while visual and auditory stimuli might be miles away from our body, somatosensory stimuli are not. They have an immediate impact onto our body surface, which makes somatosensory stimuli very different from auditory or visual ones. A further difference might lay in the fact that to a great extent primary somatosensory areas also appear to represent non-spatial attributes of tac-tile events [10, 11]. This feature not only enablesa fast analysis, which is ecologically quite useful,it also makes non-spatial attentional selection accessible at a very early level of stimulus processing.

Tactile Attention - Neuroscience vs Robotics

New Neurocomputing journal paper about touch attention mechanisms in robotic systems. Read more here.

Some notes on tactile attention by Christine Elaine Chapman, Group de recherchezurle système nerveux central, Département de Physiologie & École de Réadaptation, Faculté de Médecine, Université de Montréal, Montréal, QC, Canada. Full-text Springer.


Attention allows one to focus awareness and the processing capacities of the brain on objects and events relevant to one’s immediate, behavioural goals, and this within the context of a central nervous system (CNS) that is constantly bombarded by a steady stream of sensory inputs. Tactile attention specifically refers to attention to somatic sensory stimuli. While this term can encompass all of the submodalities that contribute to somatic sensation (touch/pressure, position/movement, temperature, pain), most experimental studies have concentrated on characterizing sensory responsiveness to touch during manipulations of attention, and this is the focus of this essay.


Attention can be directed either voluntarily (top-down, endogenous) or involuntarily (bottom-up, exogenous) towards a specific stimulus. Typical examples would be, respectively, searching through a hand bag to find a pen versus the attention-grabbing elicited by the onset of vibration from a cell phone in one’s pocket. Attention can be focused on any number of attributes of the stimulus, including its spatial location, features (texture, shape, size, consistency etc) and/or modality. Attention can be directed to a single stimulus (selective attention), or can be shared across multiple stimuli (divided attention). Finally, attention can be directed towards a stimulus either covertly or overtly (respectively, with or without orienting the receptors towards the stimulus). Overt orientation can include orienting the head and eyes towards an object of interest, so as to facilitate interactions. For example, an insect might land on one’s arm, but the eventual motor response would depend on whether the insect might potentially bite (wasp) or not (housefly). For touch, the hand is appropriately shaped and positioned during the transport phase of a reach-to-grasp movement, thereby ensuring that the object will be secured in an efficient manner. Vision of the object is not essential for orienting to occur. Examples include the coordinated tongue and jaw movements that accompany chewing of a food morsel, or the adjustments in hand and finger posture that occur when grasping an object in the
dark. Finally, orienting behaviour can be aversive, in the case of, for example, a painful tactile stimulus.

Description of the Process

In comparison with the visual system, there is much less information available about the effects of attention on the perception of tactile stimuli. For all modalities, attentional effects are inferred by measuring the time taken to evaluate the stimulus (reaction time), response accuracy and also measures of sensory sensitivity (detection or discrimination thresholds). Faster reaction times are presumed to reflect speeded up processing and decision making, although the measures can also reflect increased readiness to move. Improved accuracy and reductions in sensory thresholds are likewise considered to reflect enhanced processing of sensory inputs. Some caution is necessary, however, because such changes can also reflect a change in response criterion or bias (willingness to report the presence of a stimulus when measuring detection threshold, for example).

Spatial Attention

When attention is directed with spatial cues to different fingers of the hands, then the ability to discriminate surface texture and also vibrotactile stimulation shows a modest improvement with cueing (attention towards or away from the location of the stimulus). The effects are, however, dependent on the task design only being obvious when the task is more difficult, detecting the absence rather than the presence of a change in stimulus intensity. Using reaction time measures, more robust attentional enhancements are seen in simpler tasks in which subjects discriminate the spatial location of vibrotactile stimuli applied to the fingers of each hand (thumb and index finger), in this case within the context of a cross-modal manipulation of attention (tactile and visual). Reaction times to vibrotactile stimuli are faster when attention is correctly directed to the general location of the stimulation (right or left hand) as compared to when attention is misdirected to the unstimulated side.Interestingly, an asymmetry is observed for targets located on the left as compared to the right, with cueing having larger effects for the former, potentially reflecting a default rightward attentional bias.

Cross-Modal Attention

To what extent are attentional controls similar across different modalities? This question can be approached by measuring the ability of subjects to perform similar sensory tasks under different attentional states: attention directed towards or away from each modality, or divided across modalities. Figure 1a shows typical results obtained in a cross-modal manipulation of attention across touch (vibrotactile stimuli to the fingertip) and vision (illumination of a central light),based on simple reaction time measures.Subject attention was redirected on a trial-by-trial basis with instruction lights: directed attention (valid cue) versus divided attention (neutral cue). The subject’s task was to respond as quickly as possible to the presentation of the stimulus (tactile or visual). The design included a small number of invalid cues(attention misdirected). Typically, reaction times are shortest with directed attention, intermediate for divided attention, and longest when attention is misdirected.Interestingly, attention has proportionally similar effects on the detection of weak and stronger vibrotactile stimuli, as well as on detection of the visual stimulus(Fig. 1b), suggesting that attention may exert a generalized effect on perceptual abilities across touch and vision. Although the experimental design could be criticized since the visual and tactile stimuli did not come from the same spatial location, an advantage for directed attention over divided attention is also seen when the spatial confound is controlled, in this case in the context of experiments contrasting attentional influences across three modalities, touch, vision and audition. In a task involving spatial discrimination, attention has proportionally larger effects on tactile than visual or auditory stimuli. While this observation might reflect different mechanisms when attention is divided across multiple modalities (versus two modalities), it should be stressed that touch requires physical contact between the surround and the receptors, and that this contact may serve as a reference point for interpreting incoming inputs. In contrast, audition and vision have no such reference.


An involuntary or reflexive shift of attention to a tactile cue is referred to as orienting. There is evidence that a preceding vibration can shorten reaction times to vibrotactile stimuli, even though the subjects had been instructed that the vibration had no relation to the target location (right or left hand) and that it should be ignored. Whether such effects are reflexive in nature is,however, not clear: for example, shortened reaction times have been shown using relatively long delays between the alerting cue and the stimulus to discriminate (200–400 ms), but such delays fall well within the time required for voluntary movement in response to a stimulus. Another approach, somewhat akin to studies of cross-modal attention, has been to show that vision of the stimulated body part (non informative since the subjects can not see the actual stimulus application), accompanied or not by orienting the head and eyes, improves the ability of subjects to perform tactile discrimination tasks. These results have been challenged since changes in response bias (willingness to report stimuli as being present) might have occurred,and contributed to the apparent enhancement of tactile sensitivity. The underlying mechanisms are likely complex since performance on other somatic sensory tasks, specifically haptic shape and orientation, are modified simply by providing non informative visual feedback or orienting the head (and eyes) towards the stimuli. Moreover, the nature of the interaction(performance enhanced or degraded) is critically dependent on the task itself, specifically whether the task calls upon an intrinsic (body-centred) or extrinsic (centred on external coordinates) reference frame.

The Vaccine War - Frontline PBS

The Vaccine War - Frontline PBS

Watch full documentary here.

Vaccines have changed the world, largely eradicating a series of terrible diseases, from smallpox to polio to diphtheria, and likely adding decades to most of our life spans. But despite the gains -- and numerous scientific studies indicating vaccine safety -- a growing movement of parents remains fearful of vaccines. And in some American communities, significant numbers of parents have been rejecting vaccines altogether, raising new concerns about the return of vaccine-preventable diseases like measles and whooping cough.

In The Vaccine War (originally broadcast on PBS in April 2010) FRONTLINE lays bare the science of vaccine safety and examines the increasingly bitter debate between the public health establishment and a formidable populist coalition of parents, celebrities, politicians and activists who are armed with the latest social media tools -- including Facebook, YouTube and Twitter -- and are determined to resist pressure from the medical and public health establishments to vaccinate, despite established scientific consensus about vaccine safety.

High-speed 3D printing using stereolithography

The new technology has some features in common with 3-D printing, but it makes objects continuously rather than in discrete layers, making it much faster. In a video of the process, it looks as if an object gradually emerges from a thin layer of liquid.

The new process is related to stereolithography, in which lasers trace a pattern on a liquid that is engineered to solidify when exposed to light. Normally, to form each layer, the laser has to be turned off so that more liquid can be spread out. This slows the process, and the “interfaces” between layers create weak points in a finished object.

Demonstration video here.

Machine learning: some common implementation mistakes

Suggestions by Cheng-Tao Chu's blog.

  • Take default loss function for granted
  • Use plain linear models for non-linear interaction
  • Forget about outliers
  • Use high variance model when n << p
  • L1/L2/... regularization without standardization
  • Use linear model without considering multi-collinear predictors
  • Interpreting absolute value of coefficients from linear or logistic regression as feature importance
  • The algorithm behind Google Maps

    ​ Algorithms are a science of cleverness. A natural manifestation of logical reasoning—​mathematical induction, in particular—a good algorithm is like a fleeting, damning snapshot into the very soul of a problem. A jungle of properties and relationships becomes a simple recurrence relation, a single-line recursive step producing boundless chaos and complexity. And to see through deep complexity, it takes cleverness.

    It was the programming pioneer Edsger W. Dijkstra that really figured this out, and his namesake algorithm remains one of the cleverest things in computer science. A relentless advocate of simplicity and elegance in mathematics, he more or less believed that every complicated problem had an accessible ground floor, a way in, and math was a tool to find it and exploit it.

    In 1956, Dijkstra was working on the ARMAC, a parallel computing machine based at the Netherlands’ Mathematical Center. It was a successor to the ARRA and ARRA II machines, which had been essentially the country’s first computers. His job was programming the thing, and once ARMAC was ready for its first public unveiling—after two years of concerted effort—Dijkstra needed a problem to solve.

    Full-text here.

    A Google for DNA

    In 2005, next-generation sequencing began to change the field of genetics research. Obtaining a person’s entire genome became fast and relatively cheap. Databases of genetic information were growing by the terabyte, and doctors and researchers were in desperate need of a way to efficiently sift through the information for the cause of a particular disorder or for clues to how patients might respond to treatment.

    Companies have sprung up over the past five years that are vying to produce the first DNA search engine. All of them have different tactics—some even have their own proprietary databases of genetic information—but most are working to link enough genetic databases so that users can quickly identify a huge variety of mutations. Most companies also craft search algorithms to supplement the genetic information with relevant biomedical literature. But as in the days of the early Web, before Google reigned supreme, a single company has yet to emerge as the clear winner.

    Read more here.


    Subscribe to Ricardo Martins RSS