Memory Structure Research Articles

Ion trap and release dynamics enables nonintrusive tactile augmentation in monolithic sensory neuron.

An iontronic-based artificial tactile nerve is a promising technology for emulating the tactile recognition and learning of human skin with low power consumption. However, its weak tactile memory and complex integration structure remain challenging. We present an ion trap and release dynamics (iTRD)-driven, neuro-inspired monolithic artificial tactile neuron (NeuroMAT) that can achieve tactile perception and memory consolidation in a single device. Through the tactile-driven release of ions initially trapped within iTRD-iongel, NeuroMAT only generates nonintrusive synaptic memory signals when mechanical stress is applied under voltage stimulation. The induced tactile memory is augmented by auxiliary voltage pulses independent of tactile sensing signals. We integrate NeuroMAT with an anthropomorphic robotic hand system to imitate memory-based human motion; the robust tactile memory of NeuroMAT enables the hand to consistently perform reliable gripping motion.

On the issue of circumstances, aggravating punishment in the commission of crime, under Article 2434 of the Criminal Code of the Russian Federation

The article studies problematic theoretical issues of application of the criminal law in case of presence of circumstances aggravating punishment in case of destruction or damage of military burial places, as well as monuments, steles, obelisks, other memorial structures or objects perpetuating the memory of those who died in defense of the Fatherland or its interests or dedicated to the days of military glory of Russia, according to article 2434 of the Criminal Code of the Russian Federation (CC RF). In particular, the most common circumstances aggravating punishment in committing an act under article 2434 of the Criminal Code of the Russian Federation are investigated, namely: in cases of recidivism of crimes, occurrence of grave consequences, committing a crime during mobilization. The author applied the methods of statistical, systemic and comparative legal analysis, formal logic. It is argued that the recognition of recidivism of a crime does not affect the social danger of a deed, but this circumstance should be taken into account in penalization for the appointment of fair punishment based on the personality of the offender. The author’s classification of grave consequences to be taken into account in the case of committing an offense under Art. 2434 of the Criminal Code of the Russian Federation is proposed. A number of aspects of such a circumstance aggravating punishment as committing a crime during mobilization, especially in relation to negligent crimes, are criticized. The author concludes that it is justified to take into account the commission of a crime during mobilization in relation to article 2434 of the Criminal Code of the RF. This will have a positive impact on the practice of imposing criminal punishment for a crime against the historical memory of those who died in defense of the Fatherland, especially since the purpose of imposing a just punishment — including the prevention of committing socially dangerous acts.

Dynamic emotional states shape the episodic structure of memory

Human emotions fluctuate over time. However, it is unclear how these shifting emotional states influence the organization of episodic memory. Here, we examine how emotion dynamics transform experiences into memorable events. Using custom musical pieces and a dynamic emotion-tracking tool to elicit and measure temporal fluctuations in felt valence and arousal, our results demonstrate that memory is organized around emotional states. While listening to music, fluctuations between different emotional valences bias temporal encoding process toward memory integration or separation. Whereas a large absolute or negative shift in valence helps segment memories into episodes, a positive emotional shift binds sequential representations together. Both discrete and dynamic shifts in music-evoked valence and arousal also enhance delayed item and temporal source memory for concurrent neutral items, signaling the beginning of new emotional events. These findings are in line with the idea that the rise and fall of emotions can sculpt unfolding experiences into memories of meaningful events.

MoViT: Memorizing Vision Transformers for Medical Image Analysis.

The synergy of long-range dependencies from transformers and local representations of image content from convolutional neural networks (CNNs) has led to advanced architectures and increased performance for various medical image analysis tasks due to their complementary benefits. However, compared with CNNs, transformers require considerably more training data, due to a larger number of parameters and an absence of inductive bias. The need for increasingly large datasets continues to be problematic, particularly in the context of medical imaging, where both annotation efforts and data protection result in limited data availability. In this work, inspired by the human decision-making process of correlating new "evidence" with previously memorized "experience", we propose a Memorizing Vision Transformer (MoViT) to alleviate the need for large-scale datasets to successfully train and deploy transformer-based architectures. MoViT leverages an external memory structure to cache history attention snapshots during the training stage. To prevent overfitting, we incorporate an innovative memory update scheme, attention temporal moving average, to update the stored external memories with the historical moving average. For inference speedup, we design a prototypical attention learning method to distill the external memory into smaller representative subsets. We evaluate our method on a public histology image dataset and an in-house MRI dataset, demonstrating that MoViT applied to varied medical image analysis tasks, can outperform vanilla transformer models across varied data regimes, especially in cases where only a small amount of annotated data is available. More importantly, MoViT can reach a competitive performance of ViT with only 3.0% of the training data. In conclusion, MoViT provides a simple plug-in for transformer architectures which may contribute to reducing the training data needed to achieve acceptable models for a broad range of medical image analysis tasks.

Deep learning with data preprocessing methods for water quality prediction in ultrafiltration

Ultrafiltration (UF) has been widely used to remove colloidal substances and suspended solids in feed water. However, UF membrane breakage can lead to downstream impurities flow, hindering subsequent filtration such as reverse osmosis. Preliminary detection for abnormal water quality after UF is vital for cost-efficient operations, but current predictive models lack accuracy. This study investigated the predictive models using deep learning algorithms, specifically convolutional neural network (CNN) and long short-term memory (LSTM) structures. One month of data was provided from a UF system in a real seawater desalination plant. Unfortunately, conventional CNN and LSTM models struggled to predict sudden turbidity spikes caused by UF membrane damage (R2 < 0.2351). To address this challenge, we proposed a novel approach coupling wavelet signals and raw data. This technique enriched turbidity data with abundant waveform signals, resulting in a significant improvement in predictive accuracy (R2 < 0.9203). Shapley additive explanation demonstrated that the wavelet signals emphasized turbidity spikes, helping models in recognizing the extent of changes. This outcome of this study is the development of highly accurate predictive models for outflow turbidity after UF. These models will enhance the safety and efficiency of UF and subsequent filtration systems, improving their overall performance.

Self-Organizing Memory Based on Adaptive Resonance Theory for Vision and Language Navigation

Vision and Language Navigation (VLN) is a task in which an agent needs to understand natural language instructions to reach the target location in a real-scene environment. To improve the model ability of long-horizon planning, emerging research focuses on extending the models with different types of memory structures, mainly including topological maps or a hidden state vector. However, the fixed-length hidden state vector is often insufficient to capture long-term temporal context. In comparison, topological maps have been shown to be beneficial for many robotic navigation tasks. Therefore, we focus on building a feasible and effective topological map representation and using it to improve the navigation performance and the generalization across seen and unseen environments. This paper presents a S elf-organizing Memory based on Adaptive Resonance Theory (SMART) module for incremental topological mapping and a framework for utilizing the SMART module to guide navigation. Based on fusion adaptive resonance theory networks, the SMART module can extract salient scenes from historical observations and build a topological map of the environmental layout. It provides a compact spatial representation and supports the discovery of novel shortcuts through inferences while being explainable in terms of cognitive science. Furthermore, given a language instruction and on top of the topological map, we propose a vision–language alignment framework for navigational decision-making. Notably, the framework utilizes three off-the-shelf pre-trained models to perform landmark extraction, node–landmark matching, and low-level controlling, without any fine-tuning on human-annotated datasets. We validate our approach using the Habitat simulator on VLN-CE tasks, which provides a photo-realistic environment for the embodied agent in continuous action space. The experimental results demonstrate that our approach achieves comparable performance to the supervised baseline.

Low-temperature enhanced OFF-state telegraph noise in defect engineered ReRAMs

The OFF-state retention characteristics of Pt/NiOx–(Ar)/NiOx–(Ar + O2)/Pt stacking resistive random access memory structures were measured as a function of temperature between 300 and 190 K. The devices show random telegraph noise effects whose noise amplitude increases upon lowering the temperature. Interestingly, the application of forward and reverse biases produces up and downresistance jumps. These results present important insights into the thermally activated charge trapping/de-trapping phenomenon that is responsible for resistance fluctuations.

Adaptive Memorization With Group Labels for Unsupervised Person Re-Identification

Re-identification (re-ID) aims to identify a person’s images across different cameras. However, the domain differences between different datasets make it a challenge for re-ID models trained on one dataset to be adapted to another. A variety of unsupervised domain adaptation methods tend to transfer learned knowledge from one domain to another by optimizing with pseudo-labels. Though impressive performances have been achieved, there are still some limitations. To be specific, these methods always generate one pseudo label for each unlabeled sample, which is hard to describe a person accurately and introduces a large number of noisy labels by one-shot clustering, thus hindering the retraining process and limiting generalization. To build more comprehensive descriptions of samples and mitigate the effects of noisy pseudo labels, this paper proposes an Adaptive Memorization with Group labels (AdaMG) framework for unsupervised person re-ID, which resists noisy labels and exploits the diversity of samples by developing a multi-branch structure with the adaptive memorization. In particular, group labels are generated for one sample in the unseen domain to learn more complementary and diverse features through clustering. Meanwhile, to better optimize the neural networks with noisy data, multiple memory structures are designed in AdaMG, which are updated adaptively according to the confidence of samples. Comprehensive experimental results have demonstrated that our proposed method can achieve excellent performances on benchmark datasets.

Comparison of 1T1R and 1C1R ReRAM Arrays

The rise of memristors with potential applications in memory has attracted wide interests. Memristors are typically assembled in crossbar arrays with data bits encoded by the resistance of individual cells. The 1T1R cell structure is the most popular for memristive memory as it eliminates sneak path currents. The transistor not only allows for flexible selection of memory cells but also facilitates the programming for computing-in-memory applications. In this paper, we replace the selector in the 1T1R configuration with a capacitor to form a selectorless and passive combination of ReRAM structure. Moreover, we evaluate the merits of the two structures in SkyWater 130nm CMOS technology by comparing the writing technique, power consumption, switching speed, and memory density. Furthermore, we compared the complexity of the readout method between 1T1R and 1C1R for a memory application. This work shows that the 1C1R configuration is a promising memory structure that consumes less energy, switches faster, has higher density, and has a simpler readout method when compared to 1T1R.

THE MEMORY MODE IN THE LITERARY WORKS OF NATALENA KOROLEWA

The article examines the specifics and functions of memories in the structure of children prose genres of Natalena Koroleva. The paper is focused on the features of modeling the story unfolded in the form of memories. The purpose of the research is to analytically comprehend the place and functions of memories in the structure of the plot in the short prose by Natalena Koroleva.&#x0D; As a result of the analysis, it has been found that the memory should be considered as an important component of the autobiographical memory. After all, these memories are private ones and demonstrate the personal experience of the narrator. In these autobiographical memories Natalena Koroleva introduces her personal experience of being the subject of experience. It can be easily identified due to the constant usage of personal pronouns such as "I", " me".&#x0D; It has been noticed that the memory compositionally used in the structure of Natalena Koroleva’s stories is usually found in the prologue, where the writer justifies the reasons, motives for its resuscitation and presentation, as it notes its significance for her autobiographical memory. That is why the writer indicates the date, the year or the specific place of events that supply such memories with persuasiveness and credibility. This way of writing also emphasizes their importance in the formation of author’s life experience.&#x0D; Episodic memories prevail in most stories, but there are also those in which significant events in the storyteller’s life became important for her as decisive ones. In addition to memories of her own travels, Natalena Koroleva easily shares with the reader her memories of personalities who she was familiar with. Moreover, she offers her own thoughts about their fate, characterizes their life, reveals their preferences and displeasures. She is as truthful as possible with the reader while sharing these memories.&#x0D; It should be pointed out that in her memoirs the writer uses intertextual inclusions that significantly emphasize all the emotions experienced by Natalena Koroleva, especially during her distant archaeological journey. The paper also highlights that intertextual inclusions in the structure of memories perform certain functions, namely: accentual, pictorial, illustrative ones.&#x0D; In her memoirs, the writer reflects on her own life experience, and her memories act as peculiar markers of the past life, which she misses nostalgically.

Memory capacity of recurrent neural networks with matrix representation

It is well known that canonical recurrent neural networks (RNNs) face limitations in learning long-term dependencies, which have been addressed by memory structures in long short-term memory (LSTM) networks. Neural Turing machines (NTMs) are a variant of RNNs that implement the notion of programmable memory with neural network controllers that can learn simple algorithmic tasks. Matrix neural networks, on the other hand, feature matrix representations which inherently has the potential to preserve the spatial structure of data when compared to canonical neural networks that use only vector-based representation. One may then argue that neural networks with matrix representations may have the potential to provide better memory capacity. In this paper, we define and study a probabilistic notion of memory capacity based on Fisher information for matrix-based RNNs. We find bounds on memory capacity for such networks under various hypotheses and compare them with their conventional (vector) counterparts. In particular, we show that the memory capacity of such networks is bounded by N2, for N×N state matrix which generalizes and provides similar results for vector networks. We also show and analyze the increase in memory capacity for such networks which is introduced when one exhibits an external state memory, such as NTMs. Consequently, we construct NTMs with RNN controllers with matrix-based representation of external memory, leading us to introduce Matrix NTMs. We demonstrate the performance of this class of memory networks under certain algorithmic learning tasks such as copying and recall and compare it with Matrix RNNs. We find an improvement in the performance of Matrix NTMs by the addition of external memory, in comparison to Matrix RNNs.

A deep implicit memory Gaussian network for time series forecasting

In recent years, significant achievements have been made in time series forecasting using deep learning methods, particularly the Long Short-Term Memory Network (LSTM). However, time series often exhibit complex patterns and relationships like trends, seasonal patterns and irregularities, and LSTM networks fail to effectively strengthen long-term dependence in time series data, which may lead to inaccurate forecasting effect or performance degradation. Therefore, building a model to explore the temporal dependence within the time series data completely still remains a challenge. In this paper, we propose a novel Deep Implicit Memory Gaussian (DIMG) Network based on bidirectional deep memory kernel process and the implicit features enhancement method for time series forecasting. We first use the implicit features enhancement method to obtain the hidden features of the data according to the nonlinear mapping characteristics of encoder. Then, a new deep learning process called bidirectional deep memory kernel has been developed, which merges the structural properties of deep learning with the adaptability of kernel methods to capture intricate information and memory structures in sequential data. This process fully encapsulates the structure of Bi-LSTM and Gaussian process regression (GPR). Finally, the performance of the proposed model is evaluated on two real-world datasets. The experimental results verify that our model outperforms other reported methods in terms of prediction accuracy.

Monolithic 3D integration of 2D transistors and vertical RRAMs in 1T–4R structure for high-density memory

Emerging data-intensive computation has driven the advanced packaging and vertical stacking of integrated circuits, for minimized latency and energy consumption. Yet a monolithic three-dimensional (3D) integrated structure with interleaved logic and high-density memory layers has been difficult to achieve due to challenges in managing the thermal budget. Here we experimentally demonstrate a monolithic 3D integration of atomically-thin molybdenum disulfide (MoS2) transistors and 3D vertical resistive random-access memories (VRRAMs), with the MoS2 transistors stacked between the bottom-plane and top-plane VRRAMs. The whole fabrication process is integration-friendly (below 300 °C), and the measurement results confirm that the top-plane fabrication does not affect the bottom-plane devices. The MoS2 transistor can drive each layer of VRRAM into four resistance states. Circuit-level modeling of the monolithic 3D structure demonstrates smaller area, faster data transfer, and lower energy consumption than a planar memory. Such platform holds a high potential for energy-efficient 3D on-chip memory systems.

Temperature and Humidity Data Evaluation of Tight Sportswear during Motion Based on Intelligent Modeling

Abstract A neural network structure of Long Short Term Memory (LSTM) is proposed which could be used to predict the temperature and humidity of other key parts from the temperature and humidity data of some parts of the human body when wearing tight sportswear, so as to realize the temperature and humidity data prediction of all key points of the human body. The temperature and humidity of different people wearing tights were collected by DHT sensors. The experimental results show that the LSTM neural network structure proposed has higher prediction accuracy than other algorithms, and the model evaluates the feasibility of temperature and humidity data of tights in a state of motion, which facilitates the study of dynamic thermal and humid comfort and reduces the time cost of analyzing the temperature and humidity distribution and changing the law during human movement. It will effectively promote the study of temperature and humidity changes when people wear sports tights, provide theoretical reference for the study of human skin temperature in the field of sports medicine, and provide practical guidance for the application of human skin temperature changes in sports clothing production, diagnosis and prevention of sports injuries.

A 47 TOPS/W 10T SRAM-Based Multi-Bit Signed CIM With Self-Adaptive Bias Voltage Generator for Edge Computing Applications

This paper presents a 10T static random-access memory (SRAM) -computing in memory (CIM) structure of 32kb macro unit for convolutional neural networks (CNNs). The proposed CIM handles ± computation with signed input signals in a single bitcell and obtains 3 times BL range for a larger sensing margin. Although the area overhead of the SRAM bitcell is 27 larger due to the positive branch PMOS, the array size is reduced to half for the same computation. A serial multi-bit weight technique is used to extend the number of weight bits with improved linearity and no significant time penalty. A bias voltage generator is implemented to regulate the discharge current under different PT variations. Both weight and input can be adjusted to 1, 2, and 4-b. For 4-b weight and 4-b input, the proposed structure achieves 85.7% accuracy with the CIFAR-10 dataset and 47 TOPS/W energy consumption in CMOS 28nm technology.

Acceleration of Control Intensive Applications on Coarse-Grained Reconfigurable Arrays for Embedded Systems

Embedded systems confront two opposite goals: low-power operation and high performance. The current trend to reach these goals is toward heterogeneous platforms, including multi-core architectures with heterogeneous cores and hardware accelerators. The latter can be divided into custom accelerators (e.g., ASICs) and programmable domain-specific cores (e.g., DSIPs). VWR2A [1] is a programmable architecture that integrates high computational density and low power memory structures. The flexibility of VWR2A allows a large portion of applications to be covered, resulting in better performance and energy efficiency than ASICs and general-purpose processors. However, while this has been well studied for data-intensive kernels, this is not the case for control-intensive kernels —code with complex if-else and nested loop structures. Traditionally, control-intensive code is left to be executed by the host processor. This situation unnecessarily restricts the potential impact of energy-efficient acceleration, especially at the application level. In this paper, we evaluate the performance and energy consumption of VWR2A for control-intensive code and compare it with an ARM Cortex-M4 processor and a RISC-V Ibex processor. The performance and energy consumption are evaluated at the kernel and application levels. Our results confirm that VWR2A is faster and more energy-efficient than the two considered general-purpose processors also for control-intensive code.

Pulsed Rapid Thermal Annealing as a Low Thermal Budget Semiconductor Fabrication Process

Thermal annealing is required in the fabrication of many semiconductor devices. A proper annealing condition, e.g., temperature, atmosphere, and time, can improve the device performance, e.g., by reducing defect densities in bulk films and at interfaces. In addition, the annealing step can generate new structures in previously deposited thin films for various new functions. However, when the annealing condition is improper, e.g., the temperature is too high or the annealing time is too long, the device performance can deteriorate or dysfunction due to damage to the composing material, structure, or even the substrate. The thermal budget is critical in the thermal annealing step.Pulsed rapid thermal annealing (PRTA), which exposes the sample to repeated cycles of short-period heating and cooling, is a low thermal budget process. In this presentation, the principle of PRTA is presented. Examples of using PRTA in the crystallization of amorphous silicon on the glass substrate to polycrystalline thin films for thin film transistors (TFTs) and solar cells, improving high-k gate dielectric properties for MOS capacitors and solid state incandescent light emitting devices (SSI-LEDs), and preparation of nanocrystals embedded high-k structure for nonvolatile memories will be shown (1-4). Limitations of PRTA and possible improvements will be discussed. Kuo and P. M. Kozlowski, Appl. Phys. Lett., 69(8), 1092-1094 (1996).Kuo, C.-H. Lin, and M. Zhu, IEEE Photovoltaic Specialist Conf. PVSC 35, 3698-3701 (2010).Y. Kuo and C. H. Lin, ECS 237th Meeting # 130648, May 10-15, 2020.Y. Kuo, ECS 227th Meeting # 47306 (1333), May 24-28, 2015.

Exploring Deep Learning Approaches for Short-Term Passenger Demand Prediction

An accurate short-term passenger demand forecast makes a contribution to the coordination of traffic supply and demand. Forecasting the short-term passenger demand for the on-demand transportation service platform is of utmost significance since it might incentivize empty cars to relocate from over-supply regions to over-demand regions. Yet, because spatial, temporal, and exogenous dependencies need to be evaluated concurrently, short-term passenger demand forecasting may be rather difficult. This article aims to investigate several methods that can be utilized to forecast short-term traffic demand, with a primary emphasis on deep learning approaches. We examine varying degrees of temporal aggregation and how these levels affect various architectural configurations. In addition, by analyzing 22 models representing 5 distinct architectural configurations, we illustrate the influence of varying layer configurations within each architecture. The findings indicate that the long-term short memory (LSTM) structures perform the best for short-term time series forecasting, but more complex architectures do not significantly enhance the outcomes. Moreover, considering the spatiotemporal aspects results in an improvement in the prediction of more than fifty percent. In addition, we investigate the vectorization of time, also known as Time2Vec, as a way of embedding to make it possible for a selected algorithm to recognize periodic characteristics in time series, and we show that the outcome is improved by fifteen percent.

Evaluation Of Development Of Clinical Reasoning Skills In Dental Students Through Diagnostic Thinking Inventory.

To assessthe clinical reasoning skills among dentalstudents, and to compare the diagnostic thinking ability of male and female students. The cross-sectionalstudy was conducted from May to July 2021 at a public-sector dental school in Karachi, and comprised students of both genders from across allthe four years ofdentalschool.Datawas collectedusing thediagnostic thinking inventory which also assessed flexibility in thinking and knowledge structure inmemory.Datawas analysedusing SPSS 18. Of the 111 forms distributed, 108(98.2%) were received duly filled; 32 (29.35%) from male students and 76(69.72) from female students. Participants from the first year were 24(22%), second year 23(21%), third year 33 (30.27%) and the fourth year 29(26.6%).The reliability of diagnostic thinking inventory was 0.906, for knowledge structure in memory 0.661, andflexibility inthinking0.573.The correlationwassignificantforstudentsintheir 2nd, 3rdand4thyearsofstudies(p=0.01) The overall mean score was <150 which was poor. Both the knowledge structure in memory, and flexibility in thinking scores were higher for females compared to males(Significant at the level of 0.01). The clinical reasoning skills of students were low, and the diagnostic thinking ability of female students was better than that of their male counterparts.

A two-level parallel decomposition-based artificial bee colony method for dynamic multi-objective optimization problems

Many real-world multiple-objective optimization problems have objectives that change over time. These multiple-objective optimization problems are called dynamic multiple-objective optimization problems (DMOPs) and have received an increased attention. To track the changing Pareto front in DMOPs, the Pareto front at a certain moment needs to be obtained as efficiently as possible, which is challenging for most of existing methods. To this end, we propose a two-level parallel decomposition-based artificial bee colony method for solving DMOPs. To sufficiently accelerate the process of obtaining the Pareto front, a two-level parallel structure is designed in our method. In the first-level parallel structure, the dynamic multi-objective optimization problem at a certain moment is decomposed into a set of single-objective optimization problems that could be solved in parallel. In the second-level parallel structure, a parallel artificial bee colony algorithm is applied to solve each decomposed single-objective optimization problem. Specially, the parallel bee colony algorithm in our method is improved to support the exchange of information among neighbor problems, which is widely accepted to be effective in improving the efficiency of obtaining optimal solutions. To support the implementation of our improved parallel artificial bee colony algorithm, a two-level shared memory structure is designed. Our proposed method is compared with 4 widely used methods on CEC’ 2018 multi-objective optimization benchmarks and two constrained dynamic multi-objective optimization problems. The experimental results show that our method outperforms other compared methods in efficiency while maintaining good scalability and convergence.