What Are the Different Working Memory Models?

Working memory is a memory system with limited capacity for temporary processing and storage of information. It plays an important role in many complex cognitive activities. In 1974, Baddeley and Hitch proposed a three-system concept of working memory based on experiments to simulate short-term memory impairments, replacing the original "short-term memory" with "working memory" (working memory, WM). STM) concept. Since then, working memory and short-term memory have different meanings and contexts.

Working memory: refers to the temporary storage of information and its joint operation with other more complex tasks.

Working memory refers to a system with limited capacity to temporarily hold and store information.

Working memory (working memory, WM) refers to a system with limited resources for temporary storage and processing of information during the execution of cognitive tasks [1]. WM is described as the human cognitive center and is one of the most active research fields in cognitive psychology and cognitive neuroscience. Due to the prominent importance of WM in human advanced cognitive activities, Goldman-Rakic rated it as "perhaps the most important achievement in human psychological evolution" [2]. At present, for the mechanism of WM, more than a dozen influential theoretical models have been proposed internationally, the most famous of which is Baddeley's multi-component model [3]. The model considers that WM is composed of a speech loop, a visual space template, and a central execution system. The voice circuit is responsible for the storage and control of sound-based information. The visual space template is mainly responsible for storing and processing visual information. The central execution system is the core of WM. It is responsible for the connections between subsystems and their long-term memory. Coordination and strategy selection and planning. The research around the validation and improvement of this model has been the core area of WM research.

Baddeley's working memory consists of three parts:

Including the attention control system-the central execution system, and the two subsystems that serve it "the visual and air preliminary processing system for the temporary storage and processing of visual materials" and "the voice circuit for the temporary storage and processing of spoken materials"

1. Voice-based voice loop. It is mainly used to remember the order of words and keep information.

2. Void-space image processor. Mainly used for processing visual and spatial information.

3, similar to the central system of attention. It is mainly used to allocate attention resources and control the processing process, which is a key component of working memory.

However, in the study of working memory, some experimental studies cannot be explained by Baddeley's three-system concept. For example, in the experiment, the subjects can only remember about 5 irrelevant words, but they can remember about 16 common words. Based on the modification of the original working memory model, Baddeley proposed the concept of the context buffer as a supplement to the three system concept flaws. This is a secondary memory system for storing different information processing results. It maintains the processed information under the control of the central execution system and supports subsequent processing operations.

WM's research methods cover the major research techniques of cognitive psychology and cognitive neuroscience. Among them, event-related potentials (ERPs) are technologies that have been increasingly used in this field in recent years. The outstanding feature of this technology is that it can measure the brain potential caused by specific cognitive events in real time with time resolution accurate to the millisecond level. The technology can also detect changes in internal psychological processes without the need for participants to make explicit behavioral responses. These advantages enable ERP technology to clearly distinguish the different stages of the information processing process, and this distinction is undoubtedly of great value in establishing a scientific cognitive model. However, because ERP indicators are relatively sensitive to independent and irrelevant variables, they were mainly used to study relatively simple cognitive mechanisms such as attention and perception. Until the WM experimental paradigm has been continuously innovated in recent years, ERP has been increasingly used in the research of WM core areas, and has achieved many important results.

P300 effect of working memory

P300 refers to the third positive wave in the late component of ERP. Since the earliest discovered P3 appeared about 300 milliseconds after the stimulus was presented, it is called P300. Now P300 has become a family with multiple sub-components. Many studies have shown that P300 is closely related to WM. Donchin et al. Examined the ERP effect in a typical WM task and found that the fluctuation of P300 can be regarded as an indicator of context updating in WM. The higher the complexity of WM task, the larger the P300 amplitude. They also found that the latency of P300 was affected by perceptual complexity and task difficulty, and could be considered as an indicator of stimulus evaluation and classification time. Polich et al. Also found that in tasks that require more processing resources, such as tasks that quickly implement resource allocation and information retention, there is a stable relationship between the latency of P300 and cognitive ability. However, there are also some different experimental results about the significance of P300 latency. For example, ERP research on complex tasks shows that there is a close relationship between the amplitude of P300 and cognitive ability, and the latency of P300 is related to cognitive operation and cognitive ability. It doesn't matter. In a response time task of one of five choices, subjects with high reading breadth in WM produced a larger amplitude of P300; in an n-back task, as the WM load increased, the amplitude of P300 decreased, and the amplitude was Participants' Wechsler intelligence test scores were positively correlated [4]. In both cases, the latency of P300 has nothing to do with cognitive ability. In addition, some studies have found that the latency of P300 is negatively correlated with cognitive ability, while the amplitude of P300 is positively correlated with cognitive ability.

The N400 effect of working memory.

N400 refers to a negative wave with a peak latency of 400ms, and it is also the total 115 name of an ERP family. The typical N400 is the most common ERP component in the cognitive processing of human brain language, and it usually appears when the subjects read sentences with ambiguity at the end. However, the substance of the N400 is still controversial. For example, it reflects only the nature of language itself, or semantic processing in the general sense? Is it only a question related to language or is it related to other intermediary factors? These questions have not been well answered.

Gunter et al. When examining the relationship between N400 and reading comprehension performance under high WM load conditions, found that N400 is likely to be a direct indicator of WM capacity and flexibility, and has nothing to do with reading comprehension itself. The results of this study are very subversive and challenge the long-standing relationship between N400 and reading comprehension. So is this finding reliable? If it is reliable, what is the relationship between N400 and reading comprehension and WM?

Salisbury recently made an important breakthrough in the research of the above problems by examining the relationship between N400 and semantic memory test and speech WM capacity. This study uses the task of disambiguation of homographs under low WM load. The experimental material uses short sentences of the pedigree structure or subject-predicate structure, such as "noun is an adjective" or "noun + intransitive verb". Nouns are divided into homomorphic words and definite meaning words; adjectives or verbs are words that have a dispel effect on homomorphic words, and they are consistent with and inconsistent with definite meaning words. The understanding of the subordinate meanings of homographs is related to the storage of semantic memory and the amount of knowledge. If N400 reflects vocabulary access, it should also be related to semantic knowledge, because this is mainly related to crystal memory, which is responsible for the temporal-apical circuit of the cerebral cortex; if N400 reflects vocabulary integration, then it should be related to WM capacity, because This is mainly related to fluid memory, which is responsible for the prefrontal cortex of the cerebral cortex; if N400 only reflects the mechanism and capacity of WM, then not only subjects in the high WM capacity group should have significantly greater amplitudes in N400 than subjects in the low WM capacity group, but It should also have nothing to do with understanding error rates and semantic storage. It was found that N400 is related to WM capacity, but not to semantic knowledge, indicating that it mainly reflects a late processing of WM. The larger the WM capacity, the greater the N400 caused by the disambiguation word. The experimental results support the third hypothesis, so N400 is an indicator of the capacity of WM, rather than an indicator that is generally considered to be the difficulty of context integration. This research strongly supports the findings of Gunter et al., And has made breakthrough progress in the research of the N400 substance, while providing an important indicator for future WM research.

CNV effect of working memory

CNV (contingent negative variation) usually refers to the negative deflection of EEG observed between the preparatory signal and the command signal, and is one of the earliest pure psychological waves found. It is generally believed that the psychological nature of CNV is to increase the psychological load of multiple psychological factors such as expectation, orientation response, awakening, attention and motivation when completing the same task.

Casini et al. Found that when subjects were required to encode or generate information about the length of time in WM, a negative slow wave could be recorded on the scalp. They found that this negative slow wave is CNV, and its amplitude changes with the increase of required attention resources. Casini et al. Also found a close relationship between CNV levels in the prefrontal lobe and temporal information processing. Monfort et al. Recently further examined the relationship between ERP and the maintenance of temporal information in WM. In the experiment, they recorded the ERP during the time information encoding and storage, and examined the impact of WM load on these ERP components. The load is controlled by memorizing the number of time periods and the processing degree of time information. It was found that a typical CNV effect occurred during encoding. There is no significant difference in the amplitude of CNV under different resource requirements. During the holding period, the slow waves at multiple electrode positions in the frontal region showed a significant positive shift, while the slow waves in the top occipital region showed a negative shift. The amplitude of the frontal positive slow wave has different tasks, and the tasks with larger resource requirements have larger amplitudes. Therefore, the study revealed that during the preservation of time information, a slow ERP wave will appear, and the amplitude of this ERP component is related to the resource requirements of the WM task. ^[2]

Research on ERP and Visual Working Memory

The typical application of ERP in the research of visual WM is to distinguish the electrophysiological mechanisms of visual space, visual object and speech WM. As mentioned earlier, Baddeley's three-component model divides WM into two subsidiary systems: visual space WM and speech WM. The visual space system may be divided into visual object WM and spatial WM. ERP technology provides strong evidence for the distinction between WM in these two visual systems and their distinction from verbal WM.

Research on ERP and Verbal Working Memory

The application of ERP technology in the study of speech WM is reflected in the discussion of its specific components, such as the research of Ruchkin et al. Mentioned above; on the other hand, it focuses on the study of the structure and mechanism of speech WM. This article focuses on the second aspect of research. In recent years, some researchers have proposed that speech WM can be further distinguished like visual WM. As Caplan et al. Point out, verbal WM can be divided into interpretive WM (interpretative working memory) and post-interpretative working memory (11). The former is responsible for the automatic processing of speech modularity, such as syntactic processing; the latter is responsible for the central control processing of speech, such as the exploration of external world knowledge. Caplan et al. Provided evidence for the dual structure of their speech WM through PET technology. To sum up, ERP technology has been widely and deeply applied in the research of WM as an advanced cognitive function. ERP has been applied to the research of many key issues in the WM field and has achieved significant results. These studies provide accurate time-history information for the internal processes of WM, enabling researchers to understand the start and end time of different types of WM encoding, saving, updating, and central execution control and their approximate distribution in the brain area. Development and improvement provided important data. However, there are still some shortcomings in the current WM ERP research. The WM connotation of some ERP components is not very clear, the differences between different research results are relatively large, and the study of some issues is not systematic enough.

Research on ERP and Central Executive System

Central executive system (CE) is the most important but the most difficult field in WM model research. The current research in this area mainly focuses on the ERP effect of CE update function and attention transfer function.

Donchin et al. Pointed out that the P300 component in human ERP is caused by the update of WM content, which is the earliest research on the relationship between ERP and CE. Donchin's scenario updating hypothesis was later supported by many ERP studies. Fabiani et al. Found that the P300 volatility caused by the presentation of speech stimuli was significantly related to the memory performance. However, the correlation was significant only when the subjects used the method of mechanical memory, but not when the subjects used memory strategies such as finishing. Because of the use of mechanical memory, participants need to update the content of WM with the representation of the target item; while the finishing strategy involves other processing processes, the success of the recall mainly depends on the effectiveness of the strategy, not the highlight of the target stimulus Sex, thereby eliminating the effect of memory update on the target's stimulus properties, leading to the inability to observe the relationship between P300 and memories disappearing. Therefore, the research provided support for the updated relationship between P300 and WM.

However, the above studies have obvious shortcomings, because the tasks they use are not typical WM tasks, but most of them are short-term memory scanning tasks of Sternberg, and the relationship between memory update and P300 is obtained through indirect derivation. To this end, Kiss et al. Used Baddeley's model as the theoretical background and directly introduced the ERP effect of CE update function by introducing sophisticated experimental tasks. The research set up a control task matching the update task. In the update task, the participants were required to dynamically update the content stored in the WM. The corresponding control task did not require the participants to keep changing memory stimuli. The sequence and reaction pattern are exactly the same. By subtracting the ERP in the update task from the ERP in the control task, an ERP mode specifically related to WM update can be obtained. For example, in a specific experiment, subjects are presented with a single number in turn, and the length of the stimulation sequence is randomly changed between 2 and 5, and a pair of numbers is presented after each sequence as a memory target. Under the update condition, half of these number pairs exactly match the order and content of the last two numbers in the sequence just presented, and the other half do not match.The participants were required to press the right mouse button when they found a match. Press the left mouse button for a matching number pair. Under controlled conditions, half of the stimulus pairs are equivalent to unmatched stimuli under update conditions, and the other half are two "*" signs, asking participants to press the left mouse button on all single digits and "*" pairs and Right-click. Subjects under update conditions need to encode, store and mark a series of items in WM. When the sequence items exceed the items to be memorized, the subjects also need to dynamically update the memory content. It was found that this update process triggered a wider forward ERP difference wave, as shown in Figure 1. Although the update conditions and control conditions involve WM processes such as encoding, storage, and serial marking, the ERP effects of these factors in the difference wave have been reduced. So the forward-difference wave shown in Figure 1 is an ERP component corresponding to the specific process of WM update. The study pioneeringly revealed the ERP effects related to the central executive function of WM.