Directed-Forgetting in Working Memory

How does the intent to remember or forget information affect working memory (WM)? To explore this question, in four experiments, we gauged the availability of the to-be-forgotten information directly. Participants remembered six words presented sequentially in separate frames. After each word offset, the frame turned either blue or orange, indicating a to-be-remembered or to-be-forgotten word, respectively. In all experiments, consistently poor recognition performance for to-be-forgotten words and facilitation of to-be-remembered words demonstrated that intent has a strong impact on WM. These directed-forgetting effects are remarkably robust: They can be observed when testing the to-be-forgotten words up to four times (Experiment 1, n=341), for both item and binding memory (Experiment 3, n=124), and even when information has to be maintained in WM up to 5s until the memory cue is presented (Experiment 2+4, n=302+321). Our study establishes a new method to jointly study the effects of intent on WM content for both relevant and irrelevant information and provides evidence for directed-forgetting in WM. Our research suggests that a combination of two processes cause directed-forgetting in WM: One process reduces memory strength of earlier memory representations as a function of subsequently encoded events. Another process rapidly encodes or boosts memory strength only when the person intends to remember that information.

Non-Markovian memory strength bounds quantum process recoverability

Generic non-Markovian quantum processes have infinitely long memory, implying an exact description that grows exponentially in complexity with observation time. Here, we present a finite memory ansatz that approximates (or recovers) the true process with errors bounded by the strength of the non-Markovian memory. The introduced memory strength is an operational quantity and depends on the way the process is probed. Remarkably, the recovery error is bounded by the smallest memory strength over all possible probing methods. This allows for an unambiguous and efficient description of non-Markovian phenomena, enabling compression and recovery techniques pivotal to near-term technologies. We highlight the implications of our results by analyzing an exactly solvable model to show that memory truncation is possible even in a highly non-Markovian regime.

I remember it now, so I'll remember it later: Working Memory Representations Guide Inaccurate Predictions of Future Memory Performance

Making accurate predictions of future memory performance (Judgements of Learning; JOLs) is a prerequisite for efficient learning. Since decades, those JOLs are assumed to be made inferentially, based on cues. This cue-utilization approach substituted the idea that JOLs are directly linked to memory quality. We criticize the reasons for the rejection of this memory-strength hypothesis because they ignore the existence of two different memory systems: working memory which holds representations immediately accessible, and long-term memory which is a more permanent store. Considering both memory systems, the current work revisited the memory-strength hypothesis: In Experiment 1, participants memorized sequences of two or four colored objects, then they provided JOLs for a long-term memory test, and performed a working memory test on the objects’ colors. After learning 200 objects, the long-term memory test on all studied objects followed. Sequence-length affected working memory, but not long-term memory performance. JOLs, however, were higher for sequences of two than four and correlated higher with working memory than long-term memory performance. Experiment 2 replicated the sequence-length effect on JOLs in the absence of a working memory test. Results of a sequence-eight condition revealed an increase in JOLs’ accuracy when the number of studied objects exceeded working memory span. Contrary to predominant theories, our findings suggest that JOLs are based on the quality of memory representations.

Sources of interference in item and associative recognition memory

A powerful theoretical framework for exploring recognition memory is the global matchingframework, in which a cue’s memory strength reflects the similarity of the retrieval cuesbeing matched against the contents of memory simultaneously. Contributions at retrievalcan be categorized as matches and mismatches to the item and context cues, including theself match (match on item and context), item noise (match on context, mismatch on item),context noise (match on item, mismatch on context), and background noise (mismatch onitem and context). We present a model that directly parameterizes the matches andmismatches to the item and context cues, which enables estimation of the magnitude ofeach interference contribution (item noise, context noise, and background noise). Themodel was fit within a hierarchical Bayesian framework to ten recognition memory datasetsthat employ manipulations of strength, list length, list strength, word frequency, study-testdelay, and stimulus class in item and associative recognition. Estimates of the modelparameters revealed at most a small contribution of item noise that varies by stimulusclass, with virtually no item noise for single words and scenes. Despite the unpopularity ofbackground noise in recognition memory models, background noise estimates dominated atretrieval across nearly all stimulus classes with the exception of high frequency words,which exhibited equivalent levels of context noise and background noise. These parameterestimates suggest that the majority of interference in recognition memory stems fromexperiences acquired prior to the learning episode.

Using Extinction-Renewal to Circumvent the Memory Strength Boundary Condition in Fear Memory Reconsolidation

Reconsolidation is a process by which memories are destabilized, updated, and then restabilized. Strong memories are resistant to undergoing reconsolidation. Here, we addressed whether an overtrained fear memory could be made susceptible to reconsolidation by first extinguishing, and then renewing, the memory. Rats were trained with ten tone-footshock pairings, followed by eight days of tone extinction in the training context. The next day, rats were placed into a second context and memory for the tone was renewed/reactivated with a single tone presentation. Immediately following reactivation, rats received an injection of midazolam or vehicle. Rats were then tested for freezing to the tone in a third context. Midazolam had no effect in rats that did not undergo tone extinction, but significantly attenuated freezing to the tone in extinguished rats. Thus, rats that received tone extinction underwent tone memory reconsolidation following its renewal. In a second experiment, we administered the reactivation session and midazolam injections prior to extinction. Midazolam had no effect and rats extinguished at a rate similar to controls. These data suggest that strong emotional memories are capable of updating following weakening of memory expression through extinction.

Fading Affect Bias in Intergroup Relations

Negative affect associated with autobiographical events fades faster over time than positive affect. This Fading Affect Bias (FAB) has been established in the individual and interpersonal domains. Two studies tested the FAB in intergroup relations with Muslims ( N= 76 White British non-Muslim) and opposite gender ( N = 242 women and men) as target outgroups. The results indicated that the FAB exists in an intergroup context, for both ingroup and outgroup memories. Mediation analyses showed that intergroup contact is related to a lower fading of positive affect associated with the outgroup memory, through greater memory strength and a more positive outgroup member evaluation. The findings are important for understanding affect associated with intergroup memories and the buffering effect of positive contact.

BehaviorDEPOT: a tool for automated behavior classification and analysis in rodents

Quantitative descriptions of animal behavior are essential to understand the underlying neural substrates. Fear conditioning in rodents is a widely used assay that allows neuroscientists to probe the neural mechanisms of memory. To date, quantification of freezing behavior, a proxy for fear memory strength, is usually performed by hand or with expensive and inflexible commercial software. To overcome these barriers, we developed BehaviorDEPOT (DEcoding behavior based on POsitional Tracking), a MATLAB-based application containing six independent modules. The Experiment Module runs fear conditioning experiments using an Arduino-based design that interfaces with commercial hardware. The Analysis Module classifies freezing and analyzes spatiotemporal behavioral statistics in user-defined ways. The remaining modules can develop custom classifiers. Of note, the Inter-Rater Module establishes reliable ground-truth human labels, making it broadly useful for scientists developing classifiers with any application. BehaviorDEPOT provides a simple, flexible, automated pipeline to move from pose tracking to reliably quantifying task-relevant behaviors.