Separate Memory Research Articles

What is shared and what is not: How the Cold War era in northeastern Thailand is remembered on social media

This article investigates memories of the Cold War era in Thailand, through online practices on Facebook pages administered by residents of a borderland province in the Northeast that was a key front line during the conflict. The possibilities for greater access to, and dissemination of, information afforded by digital media technologies have created a new environment for the production of shared knowledge about the past. However, on these pages, instead of converging plural memories, participatory online culture and the use of the language of memory have enabled the creation of distinct and separated memory spaces. This article calls attention in particular to what is silenced and absent in what people are sharing online. It argues that the deep-seated ethnic politics of the Cold War have an afterlife in Thai society—especially in the country's ‘margins’ where the conflict was most violent—that is reflected as much by what is not said as by what is said online.

Treatment with Tinospora cordifolia ameliorates prenatal stress and maternal separation-induced memory deficits by restoring amygdaloid neuronal architecture

Introduction: Tinospora cordifolia (TC) possesses antioxidative properties and has been shown to improve cognition. In this study, the effects of TC were investigated on prenatal vibratory stress, maternal separation-induced amygdala plasticity, and memory retention in pregnant rats and their neonates. Methods: Four experimental animal groups of pregnant Wistar rats were employed in this research, including normal & vehicle controls, the stressed group, which received 3 hours of vibratory stress/day, and the TC-treated group, which received 6 mg/kg TC before vibratory stress. The neonates born to these mothers were then subjected to maternal separation, followed by postnatal TC treatment. After 6 weeks, the animals were assessed to evaluate memory retention, serum cortisol levels, and neuronal structural changes in the amygdala. Results: Neonates exposed to prenatal vibratory stress and maternal separation entered the dark compartment more quickly during the retention test (P<0.001), indicating reduced memory retention. In contrast, the TC-treated groups showed longer latencies (P<0.001), suggesting improved memory retention. The TC-treated mothers and neonates had longer dendrites with more branching points and intersections (P<0.001). However, these dendrites underwent pruning, indicating a complex structural response to stress and TC treatment. Conclusion: The results indicate that TC exerts neuroprotective effects against prenatal vibratory stress and maternal separation and aids memory retrieval in rats by restoring amygdala neuronal damage.

Enhancing spatial memory and pattern separation: Long-term effects of stimulant treatment in individuals with ADHD

This study explores the under-researched domain of long-term stimulant treatment in children and adolescents diagnosed with attention deficit hyperactivity disorder (ADHD). The necessity for extended treatment duration, often accompanied by safety concerns and side effects leading to treatment discontinuation, underscores the significance of this investigation. Concurrently, comparative studies have revealed adverse impacts on vulnerable regions within the hippocampal formation, accompanied by behavioral perturbations. We employed computerized tests and virtual reality to assess spatial memory, pattern separation, and object recognition memory in a cohort of children diagnosed with ADHD receiving stimulant treatment. We compared their performance to a group of neurotypical peers. Our findings indicate that the ADHD group exhibited a lower performance in spatial memory, pattern separation, and object recognition memory than ND group. Intriguingly, a positive relationship emerged between the duration of stimulant treatment and performance in these variables. Notably, this improvement was not immediate to MPH treatment but becomes significant after 24 months of treatment. In contrast to previous comparative investigations, our study did not reveal a detrimental impact on spatial navigation, object recognition memory, or pattern separation, despite the known interplay of these cognitive processes with the hippocampal formation. These results shed new light on the nuanced effects of stimulant treatment in ADHD, underscoring the need for a more comprehensive understanding of long-term treatment outcomes.

Air pollution prediction using blind source separation with Greylag Goose Optimization algorithm

Particularly, environmental pollution, such as air pollution, is still a significant issue of concern all over the world and thus requires the identification of good models for prediction to enable management. Blind Source Separation (BSS), Copula functions, and Long Short-Term Memory (LSTM) network integrated with the Greylag Goose Optimization (GGO) algorithm have been adopted in this research work to improve air pollution forecasting. The proposed model involves preprocessed data from the urban air quality monitoring dataset containing complete environmental and pollutant data. The application of Noise Reduction and Isolation techniques involves the use of methods such as Blind Source Separation (BSS). Using copula functions affords an even better estimate of the dependence structure between the variables. Both the BSS and Copula parameters are then estimated using GGO, which notably enhances the performance of these parameters. Finally, the air pollution levels are forecasted using a time series employing LSTM networks optimized by GGO. The results reveal that GGO-LSTM optimization exhibits the lowest mean squared error (MSE) compared to other optimization methods of the proposed model. The results underscore that certain aspects, such as noise reduction, dependence modeling and optimization of parameters, provide much insight into air quality. Hence, this integrated framework enables a proper approach to monitoring the environment by offering planners and policymakers information to help in articulating efficient environment air quality management strategies.

Effects of the Regular Use of Virtual Environments on Spatial Navigation and Memory.

Introduction: The cognitive effects of video games have garnered increasing attention due to their potential applications in cognitive rehabilitation and evaluation. However, the underlying mechanisms driving these cognitive modifications remain poorly understood. Objectives: This study investigates the fundamental mnemonic processes of spatial navigation, pattern separation, and recognition memory, closely associated with the hippocampus. Our objective is to elucidate the interaction of these cognitive processes and shed light on rehabilitation mechanisms that could inform the design of video games aimed at stimulating the hippocampus. Method: In this study, we assessed 48 young adults, including both video game players and non-players. We utilized virtual reality and cognitive tasks such as the Lobato Virtual Water Maze and the Mnemonic Similarity Task to evaluate their cognitive abilities. Results: Our key findings highlight that gamers exhibit heightened pattern separation abilities and demonstrate quicker and more accurate spatial learning, attributed to the cognitive stimulation induced by video games. Additionally, we uncovered a significant relationship between spatial memory, guided by environmental cues, and pattern separation, which serves as the foundation for more efficient spatial navigation. Conclusions: These results provide valuable insights into the cognitive impact of video games and offer potential for monitoring changes in rehabilitation processes and early signs of cognitive decline through virtual reality-based assessments. Ultimately, we propose that examining the relationships between cognitive processes represents an effective method for evaluating neurodegenerative conditions, offering new possibilities for early diagnosis and intervention.

The Effect of Playback Speed and Distractions on the Comprehension of Audio and Audio-Visual Materials

In the modern age, we often consume content at faster than its normal speed. Prior research suggests that watching lecture videos at speeds up to 2x does not significantly affect performance, but the mechanisms by which comprehension is preserved at faster playback speeds are not fully understood. Therefore, we sought to investigate whether there is an effect of speed when the content is audio only, varies in modality (audio-only, audio-visual) and content (textual, pictorial), or is accompanied by distractions. In four experiments, we found that: (1) increasing playback speed to 2.5x speed did not impair test performance (though we still do not advise exceeding 2x speed); (2) having a visual aspect (i.e., presentation slides, instructor images) to learning can be advantageous, especially when processing information at faster speeds; (3) there was a small benefit of receiving textual over pictorial presentations, and the effect did not vary by speed; (4) computer-based distractions (i.e., phone calls, text messages, email notifications) did not impact performance at 1x or 2x speed. Hence, students are more adept at learning at faster speeds than conventional wisdom would suggest, even with distractions present, demonstrating intact comprehension at double the natural speed of to-be-learned material. Furthermore, multimedia presentations can help mitigate the negative costs of accelerated speeds, especially when information is processed by separate working memory components (i.e., narration in the auditory channel, text or pictures in the visual channel), which reduces cognitive load.

Kdm4a is an activity downregulated barrier to generate engrams for memory separation

Memory engrams are a subset of learning activated neurons critical for memory recall, consolidation, extinction and separation. While the transcriptional profile of engrams after learning suggests profound neural changes underlying plasticity and memory formation, little is known about how memory engrams are selected and allocated. As epigenetic factors suppress memory formation, we developed a CRISPR screening in the hippocampus to search for factors controlling engram formation. We identified histone lysine-specific demethylase 4a (Kdm4a) as a negative regulator for engram formation. Kdm4a is downregulated after neural activation and controls the volume of mossy fiber boutons. Mechanistically, Kdm4a anchors to the exonic region of Trpm7 gene loci, causing the stalling of nascent RNAs and allowing burst transcription of Trpm7 upon the dismissal of Kdm4a. Furthermore, the YTH domain containing protein 2 (Ythdc2) recruits Kdm4a to the Trpm7 gene and stabilizes nascent RNAs. Reducing the expression of Kdm4a in the hippocampus via genetic manipulation or artificial neural activation facilitated the ability of pattern separation in rodents. Our work indicates that Kdm4a is a negative regulator of engram formation and suggests a priming state to generate a separate memory.

Research on Grain Futures Price Prediction Based on a Bi-DSConvLSTM-Attention Model

Grain is a commodity related to the livelihood of the nation’s people, and the volatility of its futures price affects risk management, investment decisions, and policy making. Therefore, it is very necessary to establish an accurate and efficient futures price prediction model. Aiming at improving the accuracy and efficiency of the prediction model, so as to support reasonable decision making, this paper proposes a Bi-DSConvLSTM-Attention model for grain futures price prediction, which is based on the combination of a bidirectional long short-term memory neural network (BiLSTM), a depthwise separable convolutional long short-term memory neural network (DSConvLSTM), and an attention mechanism. Firstly, the mutual information is used to evaluate, sort, and select the features for dimension reduction. Secondly, the lightweight depthwise separable convolution (DSConv) is introduced to replace the standard convolution (SConv) in ConvLSTM without sacrificing its performance. Then, the self-attention mechanism is adopted to improve the accuracy. Finally, taking the wheat futures price prediction as an example, the model is trained and its performance is evaluated. Under the Bi-DSConvLSTM-Attention model, the experimental results of selecting the most relevant 1, 2, 3, 4, 5, 6, and 7 features as the inputs showed that the optimal number of features to be selected was 4. When the four best features were selected as the inputs, the RMSE, MAE, MAPE, and R2 of the prediction result of the Bi-DSConvLSTM-Attention model were 5.61, 3.63, 0.55, and 0.9984, respectively, which is a great improvement compared with the existing price-prediction models. Other experimental results demonstrated that the model also possesses a certain degree of generalization and is capable of obtaining positive returns.

Psilocybin and 2C-B at Encoding Distort Episodic Familiarity

BackgroundAs research on psychedelics (hallucinogenic serotonin receptor 2A agonists) progresses, it is important to delineate the reliability of supposedly unique effects across this drug class. One such effect is how psychedelics impair the formation (i.e., encoding) of hippocampal-dependent recollections (retrieval of specific details) while potentially enhancing the encoding of cortical-dependent familiarity (a feeling of knowing that a stimulus has been previously experienced). MethodsIn a double-blind, placebo-controlled, within-participants study (N = 20), we tested the acute effects of 2 distinct psychedelics, psilocybin and 2C-B, on the encoding of emotional episodic memories. During acute drug effects, participants viewed negative, neutral, and positive pictures. The following day (while sober), participants completed 2 separate memory tests for these pictures. ResultsUsing computational models of memory confidence, we found trends for psilocybin and 2C-B at encoding to impair estimates of recollection that were supported by other measures/analyses. Surprisingly, psilocybin and 2C-B at encoding impaired estimates of familiarity, but these impairments were likely due to a misattribution of heightened familiarity, because both drugs at encoding selectively increased familiarity-based false alarms, especially for negative and positive stimuli. Psilocybin and 2C-B at encoding also tended to impair estimates of metamemory (understanding one’s own memory) for negative and neutral memories but enhanced estimates of metamemory for positive memories, although these effects were less reliable in additional analyses. ConclusionsDespite differences in their chemistry, pharmacology, and subjective effects, both psilocybin and 2C-B distorted episodic familiarity, suggesting a common neurocognitive mechanism across psychedelics that may drive other phenomena.

Prospects for memristors with hysteretic memristance as so-far missing core hardware element for transfer-less data computing and storage

We like and need Information and Communication Technologies (ICTs) for data processing. This is measurable in the exponential growth of data processed by ICT, e.g., ICT for cryptocurrency mining and search engines. So far, the energy demand for computing technology has increased by a factor of 1.38 every 10 years due to the exponentially increasing use of ICT systems as computing devices. Energy consumption of ICT systems is expected to rise from 1500 TWh (8% of global electricity consumption) in 2010 to 5700 TWh (14% of global electricity consumption) in 2030 [A. S. G. Andrae, Eng. Appl. Sci. Lett. 3, 19–31 (2020)]. A large part of this energy is required for the continuous data transfer between separated memory and processor units, which constitute the main components of ICT computing devices in von-Neumann architecture. This, at the same time, massively slows down the computing power of ICT systems in von-Neumann architecture. In addition, due to the increasing complexity of artificial intelligence (AI) compute algorithms, since 2010, the AI training compute time demand for computing technology has increased tenfold every year, for example, from 1 × 10−6 to 1 × 10+4 Petaflops/day in the period from 2010 to 2020 [J. Wang, see https://ark-invest.com/articles/analyst-research/ai-training/ for information about the cost to train an AI interference system (2020)]. It has been theoretically predicted that ICT systems in the neuromorphic computer architecture will circumvent all of this through the use of merged memory and processor units. However, the core hardware element for this has not yet been realized so far. In this work, we discuss the perspectives for non-volatile resistive switches with hysteretic memristance as the core hardware element for merged memory and processor units in neuromorphic computers.

Nociceptor‐Enhanced Spike‐Timing‐Dependent Plasticity in Memristor with Coexistence of Filamentary and Non‐Filamentary Switching

AbstractIn the era of big data, traditional computing architectures face limitations in handling vast amounts of data owing to the separate processing and memory units, thus causing bottlenecks and high‐energy consumption. Inspired by the human brain's information exchange mechanism, neuromorphic computing offers a promising solution. Resistive random access memory devices, particularly those with bilayer structures like Pt/TaOx/TiOx/TiN, show potential for neuromorphic computing owing to their simple design, low‐power consumption, and compatibility with existing technology. This study investigates the synaptic applications of Pt/TaOx/TiOx/TiN devices for neuromorphic computing. The unique coexistence of nonfilamentary and filamentary switching in the Pt/TaOx/TiOx/TiN device enables the realization of reservoir computing and the functions of artificial nociceptors and synapses. Additionally, the linkage between artificial nociceptors and synapses is examined based on injury‐enhanced spike‐time‐dependent plasticity paradigms. This study underscores the Pt/TaOx/TiOx/TiN device's potential in neuromorphic computing, providing a framework for simulating nociceptors, synapses, and learning principles.

A Novel Deep Learning Model for Student Performance Prediction Using Engagement Data

Technology-enhanced learning supported by virtual learning environments (VLEs) facilitates tutors and students. VLE platforms contain a wealth of information that can be used to mine insight regarding students’ learning behaviour and relationships between behaviour and academic performance, as well as to model data-driven decision-making. This study introduces a system that we termed ASIST: a novel Attention-aware convolutional Stacked BiLSTM network for student representation learning to predict their performance. ASIST exploits student academic registry, VLE click stream, and midterm continuous assessment information for their behaviour representation learning. ASIST jointly learns the student representation using five behaviour vectors. It processes the four sequential behaviour vectors using a separate stacked bidirectional long short term memory (LSTM) network. A deep convolutional neural network models the diurnal weekly interaction behaviour. It also employs the attention mechanism to assign weight to features based on their importance. Next, five encoded feature vectors are concatenated with the assessment information, and, finally, a softmax layer predicts the high-performer (H), moderate-performer (M), and at-risk (F) categories of students. We evaluate ASIST over three datasets from an Irish university, considering five evaluation metrics. ASIST achieves an area under the curve (AUC) score of 0.86 to 0.90 over the three datasets. It outperforms three baseline deep learning models and four traditional classification models. We also found that the attention mechanism has a slight impact on ASIST’s performance. The ablation analysis reveals that weekly event count has the greatest impact on ASIST, whereas diurnal weekly interaction has the least impact. The early prediction using the first seven weeks of data achieves an AUC of 0.83 up to 0.89 over the three datasets. In yearly analysis, ASIST performs best over the 2018/19 dataset and worst over the 2020/21 dataset.

High energy efficient and configurable CIM macro for image processing

Traditional von Neumann architecture, characterized by separate memory and processing units connected by a limited-capacity memory bus, faces challenges in handling tasks with high energy efficiency requirements. In contrast, the compute-in-memory (CIM) architecture offers a promising alternative, facilitating high parallelism in data processing while integrating storage functions, thereby significantly reducing memory access frequency and power consumption. This study presents a fully digital CIM macro featuring a novel self-write-back 12T cell. This bitcell is capable of performing Boolean logic operations and autonomously writing back results into the in-situ cell, thereby significantly improving energy efficiency.This can be used for binary logical operations between each pixel of two images without requiring additional storage area. A bidirectional read/write architectural design enables matrix transposition. This can achieve image rotation.Additionally, this novel 12T cell also offers the option to choose not to write back the results of logical operations, thereby providing the flexibility and configurability for image processing. Combined with an adder tree, it enables multiply-and-accumulate (MAC) operation for convolutional neural networks(CNNs). This can be used for feature extraction. We propose an 18T full adder structure with lower power-delay product than current state-of-the-art full adders, which is advantageous for improving the synthesis performance of the adder tree.The CIM macro supports a 4-bit × 4-bit MAC operation. The proposed CIM macro, designed and simulated using a 28 nm CMOS process with a 16 Kb static random access memory (SRAM), demonstrates promising results. At VDD = 0.9 V, the logic operation energy consumption is 1.35 fJ/bit with an energy efficiency of 740TOPS/W, and MAC operation energy consumption is 15.91 fJ/bit with an energy efficiency of 62.85 TOPS/W.

N‐Type GaSe Thin Flake for Field Effect Transistor, Photodetector, and Optoelectronic Memory

AbstractThe family of 2D chalcogenide semiconductors has been growing rapidly. Metal monochalcogenides, for instance, can enable new possibilities in functional electronics and optoelectronics. A Gallium Selenide (GaSe) thin flake is used to fabricate a back gated field effect transistor (FET) with n‐type conduction behavior and wide hysteresis at the ambient conditions. The device shows high mobility up to 28 cm2 V−1 s−1 with Ion/Ioff ratio over 103. Under the laser exposure, the device shows a decrease in the threshold voltage and a left‐shift of the transfer characteristic with a slight increase in the current. The transfer characteristic exhibits a hysteretic behavior with hysteresis width linearly dependent on the applied gate voltage. Moreover, the GaSe‐based FET shows a photo response with a photoresponsivity of 475 mAW−1 and detectivity of 4.6 × 1012 Jones. The photocurrent rise and decay times are 0.1 and 1.3 s, respectively. Furthermore, the GaSe FET device can be used as a performant memory device with well separated states and memory window enhanced by the laser exposure, confirming an optoelectronic memory class.

Long-short diffeomorphism memory network for weakly-supervised ultrasound landmark tracking

Ultrasound is a promising medical imaging modality benefiting from low-cost and real-time acquisition. Accurate tracking of an anatomical landmark has been of high interest for various clinical workflows such as minimally invasive surgery and ultrasound-guided radiation therapy. However, tracking an anatomical landmark accurately in ultrasound video is very challenging, due to landmark deformation, visual ambiguity and partial observation. In this paper, we propose a long-short diffeomorphism memory network (LSDM), which is a multi-task framework with an auxiliary learnable deformation prior to supporting accurate landmark tracking. Specifically, we design a novel diffeomorphic representation, which contains both long and short temporal information stored in separate memory banks for delineating motion margins and reducing cumulative errors. We further propose an expectation maximization memory alignment (EMMA) algorithm to iteratively optimize both the long and short deformation memory, updating the memory queue for mitigating local anatomical ambiguity. The proposed multi-task system can be trained in a weakly-supervised manner, which only requires few landmark annotations for tracking and zero annotation for deformation learning. We conduct extensive experiments on both public and private ultrasound landmark tracking datasets. Experimental results show that LSDM can achieve better or competitive landmark tracking performance with a strong generalization capability across different scanner types and different ultrasound modalities, compared with other state-of-the-art methods.

Retrieval-based inference in the acquired equivalence paradigm

IntroductionGeneralization is fundamental to cognition. In acquired equivalence, two stimuli that share a common association become treated as equivalent, with information acquired about one stimulus generalizing to the other. Acquired equivalence has been thought to rely on integrating related memories as they are encoded, resulting in fast spontaneous generalization, but other studies suggested effortful on-demand recombination of initially separate memories at retrieval. Here, we tested whether the tendency to separate vs. integrate related information may depend on a methodological detail of a traditional acquired equivalence paradigm.MethodsHuman participants underwent feedback-based learning of overlapping face-scene associations, choosing a correct scene for a face from two options on each trial. Foil (incorrect) scenes were controlled for half of the participants to ensure that they can only learn from corrective feedback. The other half had foils selected randomly on each trial, allowing statistical learning of face-scene co-occurrence to supplement feedback-based learning. We hypothesized that the opportunity for statistical learning would boost learning and generalization and facilitate memory integration.ResultsThe opportunity for statistical learning increased associative learning and generalization. However, rather than integrated memories, generalization was increased through learning during test.DiscussionThe results indicate that the tendency for generalization in the acquired equivalence is rather small, with no evidence for integrative encoding irrespective of group. The results inform current debates regarding encoding-based vs. retrieval-based mechanisms of generalization. They also highlight how methodological details may alter performance and the involvement of cognitive processes that underlie it.

The role of input vs. output phonological working memory in narrative production: Evidence from case series and case study approaches

ABSTRACT Separable input and output phonological working memory (WM) capacities have been proposed, with the input capacity supporting speech recognition and the output capacity supporting production. We examined the role of input vs. output phonological WM in narrative production, examining speech rate and pronoun ratio – two measures with prior evidence of a relation to phonological WM. For speech rate, a case series approach with individuals with aphasia found no significant independent contribution of input or output phonological WM capacity after controlling for single-word production. For pronoun ratio, there was some suggestion of a role for input phonological WM. Thus, neither finding supported a specific role for an output phonological buffer in speech production. In contrast, two cases demonstrating dissociations between input and output phonological WM capacities provided suggestive evidence of predicted differences in narrative production, though follow-up research is needed. Implications for case series vs. case study approaches are discussed.

Atypical pattern separation memory and its association with restricted interests and repetitive behaviors in autistic children.

Memory challenges remain understudied in childhood autism. Our study investigates one specific aspect of memory function, known as pattern separation memory, in autistic children. Pattern separation memory refers to the critical ability to store unique memories of similar stimuli; however, its role in childhood autism remains largely uncharted. Our study first uncovered that the pattern separation memory was significantly reduced in autistic children, and then showed that reduced memory performance was linked to their symptoms of repetitive, restricted interest and behavior. We also identified distinct subgroups with profiles of reduced and increased generalization for pattern separation memory. More than 72% of autistic children showed a tendency to reduce memory generalization, focusing heavily on unique details of objects for memorization. This focus made it challenging for them to identify commonalities across similar entities. Interestingly, a smaller proportion of autistic children displayed an opposite pattern of increased generalization, marked by challenges in differentiating between similar yet distinct objects. Our findings advance the understanding of memory function in autism and have practical implications for devising personalized learning strategies that align with the unique memory patterns exhibited by autistic children. This study will be of broad interest to researchers in psychology, psychiatry, and brain development as well as teachers, parents, clinicians, and the wider public.

Nonvolatile Memristive Materials and Physical Modeling for In‐Memory and In‐Sensor Computing

Separate memory and processing units are utilized in conventional von Neumann computational architectures. However, regarding the energy and the time, it is costly to shuffle data between the memory and the processing entity, and for data‐intensive applications associated with artificial intelligence, the demand is ever increasing. A paradigm shift in traditional architectures is required, and in‐memory computing is one of the non‐von‐Neumann computing strategies. By harnessing physical signatures of the memory, computing workloads are administered in the same memory element. For in‐memory computing, a wide range of memristive material (MM) systems have been examined. Moreover, developing computing schemes that perform in the same sensory network and that minimize the data shuffle between the processing unit and the sensing element is a requirement, to process large volumes of data efficiently and decrease the energy consumption. In this review, an overview of the switching character and system signature harnessed in three archetypal MM systems is rendered, along with an integrated application survey for developing in‐sensor and in‐memory computing, viz., brain‐inspired or analogue computing, physical unclonable functions, and random number generators. The recent progress in theoretical studies that reveal the structural origin of the fast‐switching ability of the MM system is further summarized.

Fine-grain Quantitative Analysis of Demand Paging in Unified Virtual Memory

The abstraction of a shared memory space over separate CPU and GPU memory domains has eased the burden of portability for many HPC codebases. However, users pay for ease of use provided by system-managed memory with a moderate-to-high performance overhead. NVIDIA Unified Virtual Memory (UVM) is currently the primary real-world implementation of such abstraction and offers a functionally equivalent testbed for in-depth performance study for both UVM and future Linux Heterogeneous Memory Management (HMM) compatible systems. The continued advocacy for UVM and HMM motivates improvement of the underlying system. We focus on UVM-based systems and investigate the root causes of UVM overhead, a non-trivial task due to complex interactions of multiple hardware and software constituents and the desired cost granularity. In our prior work, we delved deeply into UVM system architecture and showed internal behaviors of page fault servicing in batches. We provided quantitative evaluation of batch handling for various applications under different scenarios, including prefetching and oversubscription. We revealed that the driver workload depends on the interactions among application access patterns, GPU hardware constraints, and host OS components. Host OS components have significant overhead present across implementations, warranting close attention. This extension furthers our prior study in three aspects: fine-grain cost analysis and breakdown, extension to multiple GPUs, and investigation of platforms with different GPU-GPU interconnects. We take a top-down approach to quantitative batch analysis and uncover how constituent component costs accumulate and overlap, governed by synchronous and asynchronous operations. Our multi-GPU analysis shows reduced cost of GPU-GPU batch workloads compared to CPU-GPU workloads. We further demonstrate that while specialized interconnects, NVLink, can improve batch cost, their benefits are limited by host OS software overhead and GPU oversubscription. This study serves as a proxy for future shared memory systems, such as those that interface with HMM, and the development of interconnects.