Joshua J. Foster

The topography of alpha-band activity tracks the content of spatial working memory

Working memory (WM) is a system for the online storage of information. An emerging view is that neuronal oscillations coordinate the cellular assemblies that code the content of WM. In line with this view, previous work has demonstrated that oscillatory activity in the alpha band (8-12 Hz) plays a role in WM maintenance, but the exact contributions of this activity have remained unclear. Here, we used an inverted spatial encoding model in combination with electroencephalography (EEG) to test whether the topographic distribution of alpha-band activity tracks spatial representations held in WM. Participants in three experiments performed spatial WM tasks that required them to remember the precise angular location of a sample stimulus for 1,000-1,750 ms. Across all three experiments, we found that the topographic distribution of alpha-band activity tracked the specific location that was held in WM. Evoked (i.e., activity phase-locked to stimulus onset) and total (i.e., activity regardless of phase) power across a range of low-frequency bands transiently tracked the location of the sample stimulus following stimulus onset. However, only total power in the alpha band tracked the content of spatial WM throughout the memory delay period, which enabled reconstruction of location-selective channel tuning functions (CTFs). These findings demonstrate that alpha-band activity is directly related to the coding of spatial representations held in WM and provide a promising method for tracking the content of this online memory system.

Open Data: Alpha-band oscillations enable spatially and temporally resolved tracking of covert spatial attention

Covert spatial attention is essential for humans’ ability to direct limited processing resources to the relevant aspects of visual scenes. A growing body of evidence suggests that rhythmic neural activity in the alpha frequency band (8–12 Hz) tracks the spatial locus of covert attention, which suggests that alpha activity is integral to spatial attention. However, extant work has not provided a compelling test of another key prediction: that alpha activity tracks the temporal dynamics of covert spatial orienting. In the current study, we examined the time course of spatially specific alpha activity after central cues and during visual search. Critically, the time course of this activity tracked trial-by-trial variations in orienting latency during visual search. These findings provide important new evidence for the link between rhythmic brain activity and covert spatial attention, and they highlight a powerful approach for tracking the spatial and temporal dynamics of this core cognitive process.

Spatially Selective Alpha Oscillations Reveal Moment-by-Moment Trade-offs between Working Memory and Attention

Current theories assume a functional role for covert attention in the maintenance of spatial information in working memory. Consistent with this view, both the locus of attention and positions stored in working memory can be decoded based on the topography of oscillatory alpha-band (8–12 Hz) activity on the scalp. Thus far, however, alpha modulation has been studied in isolation for covert attention and working memory tasks. Here, we applied an inverted spatial encoding model in combination with EEG to study the temporal dynamics of spatially specific alpha activity during a task that required observers to visually select a target location while maintaining another independently varying location in working memory. During the memory delay period, alpha-based spatial tuning functions shifted from the position stored in working memory to the covertly attended position and back again after the attention task was completed. The findings provide further evidence for a common oscillatory mechanism in both the selection and the maintenance of relevant spatial visual information and demonstrate the dynamic trade-off in prioritization between two spatial tasks.

Alpha-band oscillations track the retrieval of precise spatial representations from long-term memory

A hallmark of episodic memory is the phenomenon of mentally re-experiencing the details of past events, and a well-established concept is that the neuronal activity that mediates encoding is reinstated at retrieval. Evidence for reinstatement has come from multiple modalities, including functional Magnetic Resonance Imaging (fMRI) and electroencephalography (EEG). These EEG studies have shed light on the time-course of reinstatement, but have been limited to distinguishing between a few categories and/or limited measures of memory strength. The goal of this work was to investigate whether recently developed experimental and technical approaches, namely an inverted encoding model applied to alpha oscillatory power in conjunction with sensitive tests of memory retrieval in a continuous space, can track and reconstruct memory retrieval of specific spatial locations. In Experiment 1, we establish that an inverted encoding model applied to multivariate alpha topography can track retrieval of precise spatial memories. In Experiment 2, we demonstrate that the pattern of multivariate alpha activity at study is similar to the pattern observed during retrieval. Finally, we observe that these encoding models predict memory retrieval behavior, including the accuracy and latency of recall. These findings highlight the broad potential for using encoding models to characterize long-term memory retrieval.

Inverted Encoding Models Assay Population-Level Stimulus Representations, Not Single-Unit Neural Tuning

Inverted encoding models (IEMs) are a powerful tool for reconstructing population-level stimulus representations from aggregate measurements of neural activity (e.g., fMRI or EEG). In a recent report, Liu et al. (2018) tested whether IEMs can provide information about the underlying tuning of single units. Here, we argue that using stimulus reconstructions to infer properties of single neurons, such as neural tuning bandwidth, is an ill-posed problem with no unambiguous solution. Instead of interpreting results from these methods as evidence about single-unit tuning, we emphasize the utility of these methods for assaying population-level stimulus representations. These can be compared across task conditions to better constrain theories of large-scale neural information processing across experimental manipulations, such as changing sensory input or attention. Neuroscience methods range astronomically in scale. In some experiments, we record subthreshold membrane potentials in individual neurons, while in others we measure aggregate responses of thousands of neurons at the millimeter scale. A central goal in neuroscience is to bridge insights across all scales to understand the core computations underlying cognition (Churchland and Sejnowski, 1988). However, inferential problems arise when moving across scales: single-unit response properties cannot be inferred from fMRI activation in single voxels, subthreshold membrane potential cannot be inferred from extracellular spike rate, and the state of single ion channels cannot be inferred from intracellular recordings. These are all examples of an inverse problem in which an observation at a larger scale is consistent with an enormous number of possible observations at a smaller scale. Recent analytical advances have circumvented challenges inherent in inverse problems by instead transforming aggregate signals from their native “measurement space” (e.g., activation pattern across fMRI voxels) into a …

Item-specific delay activity demonstrates concurrent storage of multiple items in working memory

Abstract A longstanding view holds that information is maintained in working memory (WM) via persistent neural activity that encodes the content of WM. Recent work, however, has challenged the view that all items stored in WM are actively maintained. Instead, “activity-silent” models propose that items can be maintained in WM without the need for persistent neural activity, raising the possibility that only a subset of items – perhaps just a single item – may be actively represented at a given time. While past studies have successfully decoded multiple items stored in WM, these studies cannot rule out an active switching account in which only a single item is actively represented at a time. Here, we directly tested whether multiple representations can be held concurrently in an active state. We tracked spatial representations in WM using alpha-band (8–12 Hz) activity, which encodes spatial positions held in WM. Human observers (male and female) remembered one or two positions over a short delay while we recorded EEG. Using a spatial encoding model, we reconstructed stimulus-specific working memory representations (channel tuning functions, CTFs) from the scalp distribution of alphaband power. Consistent with past work, we found the selectivity of spatial CTFs was lower when two items were stored than when one item was stored. Critically, data-driven simulations revealed that the selectivity of spatial representations in the two-item condition could not be explained by models restricting storage to a single item at a time. Thus, our findings provide robust evidence for the concurrent storage of multiple items in visual working memory. Author Summary Working memory (WM) is a mental workspace where we temporarily hold information “online” in pursuit of our current goals. However, recent activity-silent models of WM have challenged the view that all items are held in an “online” state, instead proposing that only a subset of representations in WM – perhaps just one item – are represented by persistent activity at a time. To directly test a single-item model of persistent activity, we used a spatial encoding model to read out the strength of two representations from alpha-band (8–12 Hz) power in the human EEG signal. We provide direct evidence that both locations were maintained concurrently, ruling out the possibility that declines in stimulus-specific activity are due to storing one of two items in an activity-silent state.

Working memory implements distinct maintenance mechanisms depending on task goals

Working memory is the function by which we temporarily maintain information to achieve current task goals. Models of working memory typically debate where this information is stored, rather than how it is stored. Here we ask instead what neural mechanisms are involved in storage, and how these mechanisms change as a function of task goals. Participants either had to reproduce the orientation of a memorized bar (continuous recall task), or identify the memorized bar in a search array (visual search task). The sensory input and retention interval were identical in both tasks. Next, we used decoding and forward modeling on multivariate electroencephalogram (EEG) and time-frequency decomposed EEG to investigate which neural signals carry more informational content during the retention interval. In the continuous recall task, working memory content was preferentially carried by induced oscillatory alpha-band power, while in the visual search task it was more strongly carried by the distribution of evoked (consistently elevated and non-oscillatory) EEG activity. To show the independence of these two signals, we were able to remove informational content from one signal without affecting informational content in the other. Finally, we show that the tuning characteristics of both signals change in opposite directions depending on the current task goal. We propose that these signals reflect oscillatory and elevated firing-rate mechanisms that respectively support location-based and object-based maintenance. Together, these data challenge current models of working memory that place storage in particular regions, but rather emphasize the importance of different distributed maintenance signals depending on task goals. Significance statement (120 words) Without realizing, we are constantly moving things in and out of our mind’s eye, an ability also referred to as ‘working memory’. Where did I put my screwdriver? Do we still have milk in the fridge? A central question in working memory research is how the brain maintains this information temporarily. Here we show that different neural mechanisms are involved in working memory depending on what the memory is used for. For example, remembering what a bottle of milk looks like invokes a different neural mechanism from remembering how much milk it contains: the first one primarily involved in being able to find the object, and the other one involving spatial position, such as the milk level in the bottle.