A data set linking large-scale, individual semantic networks and cognitive performance

We report data from a proof-of-concept study involving the concurrent assessment of large-scale individual semantic networks and cognitive performance. The data include 10,800 free associations--collected using a dedicated web-based platform over the course of 2-4 weeks--and responses to several cognitive tasks, including verbal fluency, episodic memory, associative recall tasks, from four younger and four older native German speakers. The data are unique in scope and composition and shed light on individual and age-related differences in mental representations and their role in cognitive performance across the lifespan.

Neurodynamics and Neural Networks

Individual neurons are modelled as nonlinear oscillators that rely on bistability and homoclinic orbits to produce spiking potentials. Simplified mathematical models, like the Fitzhugh–Nagumo and NaK models, capture successively more sophisticated behavior of individual neurons, such as thresholds and spiking. Artificial neurons are introduced that are composed of three simple features: summation of inputs, referencing to a threshold, and saturating output. Artificial networks of neurons are defined through specific network architectures that included the perceptron, feedforward networks with hidden layers that are trained using the Delta Rule, and recurrent networks with feedback. A prevalent example of a recurrent network is the Hopfield network, which performs operations such as associative recall. The dynamic trajectories of the Hopfield network have basins of attraction in state space that correspond to stored memories.

Alpha rhythms reveal when, where and how memories are retrieved

SummaryOur memories for past experiences can range from vague recognition to full-blown recall of associated details. Neuroimaging research has tried to understand the brain mechanisms underlying qualitatively different memories for decades (Yonelinas, 2002). On the one hand, Electroencephalography (EEG) has shown that recall signals unfold a few hundred milliseconds after simple recognition and are hallmarked by sustained voltage deflections over left posterior sensors (Herron, 2007; Johansson & Mecklinger, 2003; Mecklinger, Rosburg, & Johansson, 2016; Rugg & Curran, 2007). However, sensor-based analyses only provide limited insights into the supporting brain networks. On the other hand, functional magnetic resonance imaging (fMRI) has revealed a ‘core recollection network’ centred on posterior parietal and medial temporal lobe (MTL) regions (Hayama, Vilberg, & Rugg, 2012; Johnson, Suzuki, & Rugg, 2013; King, de Chastelaine, Elward, Wang, & Rugg, 2015; Rugg, Johnson, & Uncapher, 2015; Rugg & Vilberg, 2013; Thakral, Benoit, & Schacter, 2017). However, due to the relatively poor time resolution of fMRI, the temporal dynamics of these regions during retrieval remain largely unknown. In order to overcome these modality-specific limitations, we here used Magnetoencephalography (MEG) in a verbal episodic memory paradigm assessing correct rejection (CR) of lures, item recognition (IR) of old words and associative recall (AR) of paired target words. We found that power decreases in the alpha frequency band (10-12 Hz) systematically track different mnemonic outcomes in both time and space: Over left posterior sensors, alpha power decreased in a stepwise fashion from 500 ms onward, first from CR to IR and then from IR to AR. When projecting alpha power into source space, the ‘core recollection network’ known from fMRI studies emerged, including posterior parietal cortex (PPC) and hippocampus. While PPC showed a linear change across conditions, hippocampal effects were specific to recall. Critically, the hippocampal recall effect emerged ∼200 ms before the PPC recall effect, suggesting a bottom-up recall signal from hippocampus to PPC. Our data thus link engagement of the core recollection network to the temporal dynamics of episodic memory and suggest that alpha rhythms constitute a fundamental oscillatory mechanism revealing when, where and how our memories are retrieved.HighlightsAlpha rhythms distinguish between different retrieval outcomesAlpha power time courses track item recognition and associative recallSource alpha power decreases track the fMRI core recollection networkHippocampal recall signal precedes parietal signal

Dynamic Neural Turing Machine with Continuous and Discrete Addressing Schemes

We extend the neural Turing machine (NTM) model into a dynamic neural Turing machine (D-NTM) by introducing trainable address vectors. This addressing scheme maintains for each memory cell two separate vectors, content and address vectors. This allows the D-NTM to learn a wide variety of location-based addressing strategies, including both linear and nonlinear ones. We implement the D-NTM with both continuous and discrete read and write mechanisms. We investigate the mechanisms and effects of learning to read and write into a memory through experiments on Facebook bAbI tasks using both a feedforward and GRU controller. We provide extensive analysis of our model and compare different variations of neural Turing machines on this task. We show that our model outperforms long short-term memory and NTM variants. We provide further experimental results on the sequential [Formula: see text]MNIST, Stanford Natural Language Inference, associative recall, and copy tasks.

Verbalization and musical memory in string players

The purpose of this study was to examine the effect of verbalization about a brief etude on collegiate string players’ short- and long-term recall of the etude in question. We examined competing hypotheses that suggest it is possible that verbalization (i.e., verbal analysis of musical features) (a) could aid in recall both by highlighting patterns and constraints that inform the music’s creation and by facilitating the creation of explicit performance cues that help to bridge gaps between associative recall chains or (b) may hinder recall by interfering with the creation of procedural and auditory memories necessary for musical performance. Participants ( N = 20) were assigned to experimental conditions in which they learned an unfamiliar etude either through repetition alone or through repetition followed by the completion of a verbalization worksheet provided by the experimenter. Recall was tested both immediately following initial practice and 24 hours later to examine the effect of verbalization on both short- and long-term retention. Findings indicated no differences between groups on immediate recall performance but significant differences at the 24-hour recall task with participants in the verbalization condition recalling more material. In addition, the patterns of errors found across groups indicated a strong primacy effect. Theoretical implications for the study of memory processes in musical contexts and practical implications regarding the preparation of memorized performance are discussed.

Multiple-object Tracking as a Tool for Parametrically Modulating Memory Reactivation

Converging evidence supports the “nonmonotonic plasticity” hypothesis, which states that although complete retrieval may strengthen memories, partial retrieval weakens them. Yet, the classic experimental paradigms used to study effects of partial retrieval are not ideally suited to doing so, because they lack the parametric control needed to ensure that the memory is activated to the appropriate degree (i.e., that there is some retrieval but not enough to cause memory strengthening). Here, we present a novel procedure designed to accommodate this need. After participants learned a list of word–scene associates, they completed a cued mental visualization task that was combined with a multiple-object tracking (MOT) procedure, which we selected for its ability to interfere with mental visualization in a parametrically adjustable way (by varying the number of MOT targets). We also used fMRI data to successfully train an “associative recall” classifier for use in this task: This classifier revealed greater memory reactivation during trials in which associative memories were cued while participants tracked one, rather than five, MOT targets. However, the classifier was insensitive to task difficulty when recall was not taking place, suggesting that it had indeed tracked memory reactivation rather than task difficulty per se. Consistent with the classifier findings, participants' introspective ratings of visualization vividness were modulated by MOT task difficulty. In addition, we observed reduced classifier output and slowing of responses in a postreactivation memory test, consistent with the hypothesis that partial reactivation, induced by MOT, weakened memory. These results serve as a “proof of concept” that MOT can be used to parametrically modulate memory retrieval—a property that may prove useful in future investigation of partial retrieval effects, for example, in closed-loop experiments.

Multiple-object tracking as a tool for parametrically modulating memory reactivation

Converging evidence supports the “non-monotonic plasticity” hypothesis that although complete retrieval may strengthen memories, partial retrieval weakens them. Yet, the classic experimental paradigms used to study effects of partial retrieval are not ideally suited to doing so, because they lack the parametric control needed to ensure that the memory is activated to the appropriate degree (i.e., that there is some retrieval, but not enough to cause memory strengthening). Here we present a novel procedure designed to accommodate this need. After participants learned a list of word-scene associates, they completed a cued mental visualization task that was combined with a multiple-object tracking (MOT) procedure, which we selected for its ability to interfere with mental visualization in a parametrically adjustable way (by varying the number of MOT targets). We also used fMRI data to successfully train an “associative recall” classifier for use in this task: this classifier revealed greater memory reactivation during trials in which associative memories were cued while participants tracked one, rather than five MOT targets. However, the classifier was insensitive to task difficulty when recall was not taking place, suggesting it had indeed tracked memory reactivation rather than task difficulty per se. Consistent with the classifier findings, participants’ introspective ratings of visualization vividness were modulated by MOT task difficulty. In addition, we observed reduced classifier output and slowing of responses in a post-reactivation memory test, consistent with the hypothesis that partial reactivation, induced by MOT, weakened memory. These results serve as a "proof of concept” that MOT can be used to parametrically modulate memory retrieval – a property that may prove useful in future investigation of partial retrieval effects, e.g., in closed-loop experiments.