Tyler D. Bancroft

Distractor frequency influences performance in vibrotactile working memory

We use a vibrotactile-delayed match-to-sample paradigm to evaluate the effects of interference on working memory. One of the suggested mechanisms through which interference affects performance in working memory is feature overwriting: Short-term representations are maintained in a finite set of feature units (such as prefrontal neurons), and distractor stimuli co-opt some or all of those units, degrading the stored representation of an earlier stimulus. Subjects were presented with two vibrotactile stimuli and were instructed to determine whether they were of the same or different frequencies. A distractor stimulus was presented between the target and probe stimuli, the frequency of which was a function of the target stimulus. Performance on the task was affected by the frequency of the distractor, with subjects making more erroneous same judgments on different trials when the distractor frequency was closer to the probe than to the target, than when the distractor was further from the probe than the target. The results suggest that the frequency of the distractor partially overwrites the stored frequency information of the probe stimulus, providing support for the feature-overwriting explanation of working memory interference.

Extensions of the picture superiority effect in associative recognition.

Previous research has shown that the picture superiority effect (PSE) is seen in tests of associative recognition for random pairs of line drawings compared to pairs of concrete words (Hockley, 2008). In the present study we demonstrated that the PSE for associative recognition is still observed when subjects have correctly identified the individual items of each pair as old (Experiment 1), and that this effect is not due to rehearsal borrowing (Experiment 2). The PSE for associative recognition also is shown to be present but attenuated for mixed picture-word pairs (Experiment 3), and similar in magnitude for pairs of simple black and white line drawings and coloured photographs of detailed objects (Experiment 4). The results are consistent with the view that the semantic meaning of nameable pictures is activated faster than that of words thereby affording subjects more time to generate and elaborate meaningful associations between items depicted in picture form.

Associative and familiarity-based effects of environmental context on memory.

Previous research has shown that hit and false alarm rates and claims of remembering are greater when test items are shown in the same context that was present at study. In the present article, the effects of environmental context (photographs of scenes shown in the background) were evaluated in a yes-no recognition task when context was manipulated on the computer screen compared with when subjects were wearing virtual reality glasses (Experiment 1), in a forced-choice recognition task to address the question of criterion changes (Experiment 2), and in a free-recall task (Experiment 3) to address the issue of generality. The results show that both specific item-context associations and the familiarity of an old context influence memory performance. We suggest that the effects of environmental context are like other instances of reconstructive memory and can both support and distort recognition memory.

Does stimulus complexity determine whether working memory storage relies on prefrontal or sensory cortex

Traditionally, working and short-term memory (WM/STM) have been believed to rely on storage systems located in prefrontal cortex (PFC). However, recent experimental and theoretical efforts have suggested that, in many cases, sensory or other task-relevant cortex is the actual storage substrate for WM/STM. What factors determine whether a given WM/STM task relies on PFC or sensory cortex? In the present article, we outline recent experimental findings and suggest that the dimensionality or complexity of the to-be-remembered property or properties of a stimulus can be a determining factor.

Mechanisms of Interference in Vibrotactile Working Memory

In previous studies of interference in vibrotactile working memory, subjects were presented with an interfering distractor stimulus during the delay period between the target and probe stimuli in a delayed match-to-sample task. The accuracy of same/different decisions indicated feature overwriting was the mechanism of interference. However, the distractor was presented late in the delay period, and the distractor may have interfered with the decision-making process, rather than the maintenance of stored information. The present study varies the timing of distractor onset, (either early, in the middle, or late in the delay period), and demonstrates both overwriting and non-overwriting forms of interference.

Can vibrotactile working memory store multiple items

Vibrotactile working memory is increasing in popularity as a model system to test theories of working memory. Notably, however, we know little about vibrotactile working memory capacity. While most other domains of working memory are able to store multiple items (for example, the seven-plus-or-minus-two capacity of verbal memory [17]), previous examinations of vibrotactile working memory suggest that stored items may suffer from high levels of interference in the form of overwriting or representation-based interference [2,4], potentially limiting capacity and also limiting our ability to draw comparisons between vibrotactile working memory and other forms of working memory. In the present study, we use a two-item delayed match-to-sample paradigm to demonstrate that subjects are able to store multiple items in vibrotactile working memory, suggesting that interference does not catastrophically limit capacity, and strengthening our ability to compare vibrotactile working memory to other working memory tasks.

Vibrotactile working memory as a model paradigm for psychology, neuroscience, and computational modeling

Generalizing the study of short-term memory across different species and paradigms (e.g., behavioral, neuroimaging, etc.) can be challenging. Studies of human memory have often focused on stimuli with substantial semantic content, which are both difficult to model computationally (at least, using biologically based models) and often impossible to convert to animal studies. In contrast, paradigms used in animal studies are sometimes too simple or insufficiently challenging to prove useful in human studies. However, recent theoretical and experimental advances have identified an experimental paradigm that is well suited to both animal, human, and computational research. Vibrotactile working memory has several properties that make it highly suitable for use as a model system: A well-defined, relatively simple neural code, straightforward experimental designs that can be translated from animal to human research (or vice versa) with little change, simple and inexpensive experimental apparatus, and similar neural correlates in both humans and animal models. As might be deduced from the name, vibrotactile working memory is working memory for vibrational stimuli applied to the hand, most commonly to the dominant index finger. The most common experimental design is the delayed match-to-sample task, in which subjects are presented with a stimulus to be remembered (the target), an unfilled delay period, followed by a second stimulus (the probe), of the same frequency or of a different frequency as the target stimulus, and are asked to decide if the target and probe are of the same frequency or different frequencies. Notably, these vibrational stimuli are non-semantic, allowing use of the same experimental designs in both humans and non-humans. While it is theoretically possible that human subjects are verbally labeling the stimuli and storing that label, thereby converting the task from a vibrotactile to a verbal working memory task, there is no evidence of subjects using such a strategy. There has been no activation of language-specific cortical regions found in a number of human imaging studies (Soros et al., 2007; Spitzer et al., 2010). Further, a recent human behavioral study implicitly tested this hypothesis, and found no evidence of verbal coding (Bancroft et al., under review). Research in both humans and non-human primates has identified a set of four regions critical for vibrotactile working memory: primary somatosensory cortex (SI), secondary somatosensory cortex (SII), medial premotor cortex (MPC), and prefrontal cortex (PFC; Romo and Salinas, 2003). Extensive single-cell recording work in macaques has been done by Romo et al. (1999) allowing the tentative assignment of roles to these regions: SI is believed to be involved in stimulus processing, SII in stimulus processing and decision-making, PFC in stimulus storage and decision-making, and MPC in converting decisions into motor responses. Functional MRI (Preuschhof et al., 2006; Soros et al., 2007; Hegner et al., 2010), EEG (Spitzer et al., 2010), and MEG (Haegens et al., 2010) research in humans has produced results that are generally consistent with single-cell recordings in macaques, suggesting there is substantial similarity between the neural correlates of vibrotactile working memory in human and non-human primates. The vibrational frequency of stimuli is encoded in the firing rates of neurons, with firing rates being monotonic functions of the stimulus frequency. This relatively simple relationship between firing rate and stimulus frequency is not only convenient for modelers, but also allows the extraction of useful information from experimental data. Previous single-cell recording research has used information-theoretic methods to determine when neuronal firing rates are carrying information about stimulus frequency (Romo et al., 1999; Hernandez et al., 2002; Romo and Salinas, 2003). More recently, Spitzer et al. (2010) and Spitzer and Blankenburg (2011) were able to extract the frequency of a stored stimulus by observing modulations of beta-band (20–25 Hz) EEG activity in human PFC during vibrotactile working memory tasks. The relatively straightforward neural code used to encode and store vibrotactile stimuli allows examination of the flow of information through the neural systems involved in working memory, and also allows disambiguation between systems involved in stimulus processing and storage, and systems involved in other cognitive functions, such as attention or motor functioning related to response. On a cognitive level, vibrotactile working memory shares many traits with other domains of working memory, including susceptibility to interference (Harris et al., 2001; Bancroft and Servos, 2011; Bancroft et al., 2011), a storage capacity of more than one item (Bancroft et al., under review), and demand on attentional systems (Hannula et al., 2010; Spitzer and Blankenburg, 2011). Further, recent EEG research has shown that subjects are able to remove individual stimuli from a stored set of stimuli when so instructed (Spitzer and Blankenburg, 2011). As vibrotactile working memory shares common qualities with other working memory systems, but does not contain semantic content, it is useful for testing theories of working memory, as we have access to both human and animal research. For example, Postle (2006) suggested that sensory cortex may be a storage substrate for working memory. However, information-theoretic analyses of single-cell recordings in primary and secondary somatosensory cortex suggest that stimuli are not represented in these areas during the delay period, but rather in non-sensory regions of PFC (Romo et al., 1999; Hernandez et al., 2002; Romo and Salinas, 2003). Similarly, research into vibrotactile working memory has begun to pose problems for the venerable multiple-components model (see Repovs and Baddeley, 2006 for a recent overview of this model). The multiple-components model postulates modality-specific components involved in the storage and processing of stimuli (for example, the visuospatial sketchpad). The present incarnation of the model, however, does not contain a component capable of storing vibrotactile stimuli, nor do the neural correlates proposed for those components overlap substantially with the neural correlates of vibrotactile working memory. Thus, findings associated with vibrotactile working memory pose problems for the generality of the multiple-components model. Experimental apparatus used in vibrotactile working memory tasks can be constructed inexpensively and with relative ease. In our lab, we use devices constructed by mounting a large nylon screw on the middle of a speaker cone, and mounting the cone inside a plastic housing such that the top of the screw is flush with the housing. The speaker is connected to a computers standard headphone jack, and driven by a sine wave of the desired frequency. Subjects place their index finger on the surface of the screw, which then vibrates when the speaker vibrates. Such devices are simple to construct, and can be built using off-the-shelf components. Standard experimental software can be used to deliver stimuli. Stimulators using piezoelectric devices or solenoids to deliver vibrations are also available (but proportionately more expensive). Such devices, however, are not necessary in most cases, and one can construct a perfectly serviceable device at a reasonable cost. Vibrotactile working memory is a paradigm that offers researchers the ability to draw on the human, animal, and computational literatures simultaneously. Research translates well between human and non-human subjects, and vibrotactile memory relies on a well-defined neural code and set of cortical regions, making it an ideal model system that can be studied using behavioral, imaging, and computational paradigms.

Irrelevant sensory stimuli interfere with working memory storage: Evidence from a computational model of prefrontal neurons

The encoding of irrelevant stimuli into the memory store has previously been suggested as a mechanism of interference in working memory (e.g., Lange & Oberauer, Memory, 13, 333–339, 2005; Nairne, Memory & Cognition, 18, 251–269, 1990). Recently, Bancroft and Servos (Experimental Brain Research, 208, 529–532, 2011) used a tactile working memory task to provide experimental evidence that irrelevant stimuli were, in fact, encoded into working memory. In the present study, we replicated Bancroft and Servos’s experimental findings using a biologically based computational model of prefrontal neurons, providing a neurocomputational model of overwriting in working memory. Furthermore, our modeling results show that inhibition acts to protect the contents of working memory, and they suggest a need for further experimental research into the capacity of vibrotactile working memory.

Diffusion modeling of interference in vibrotactile working memory.

The nature of interference in working memory has been a subject of discussion for decades. It has previously been argued that irrelevant stimuli can interfere with working memory by being encoded into memory. Previous findings have suggested that irrelevant sensory activity can interfere with the storage of information in tactile working memory. More recently, it has been suggested that this type of interference may operate through the overwriting of stored information by interfering sensory stimuli, even when participants are instructed to ignore such stimuli. Such a mechanism of interference is consistent with previous theoretical proposals. In the present study, we use a computational diffusion model to demonstrate that previous empirical findings are best explained by the encoding of irrelevant sensory information and subsequent interference.

TMS-induced neural noise in sensory cortex interferes with short-term memory storage in prefrontal cortex

In a previous study, Harris et al. (2002) found disruption of vibrotactile short-term memory after applying single-pulse transcranial magnetic stimulation (TMS) to primary somatosensory cortex (SI) early in the maintenance period, and suggested that this demonstrated a role for SI in vibrotactile memory storage. While such a role is compatible with recent suggestions that sensory cortex is the storage substrate for working memory, it stands in contrast to a relatively large body of evidence from human EEG and single-cell recording in primates that instead points to prefrontal cortex as the storage substrate for vibrotactile memory. In the present study, we use computational methods to demonstrate how Harris et al.s results can be reproduced by TMS-induced activity in sensory cortex and subsequent feedforward interference with memory traces stored in prefrontal cortex, thereby reconciling discordant findings in the tactile memory literature.