Complex Learning Tasks Research Articles

Sketching Data Sets for Large-Scale Learning: Keeping only what you need

Big data can be a blessing: with very large training data sets it becomes possible to perform complex learning tasks with unprecedented accuracy. Yet, this improved performance comes at the price of enormous computational challenges. Thus, one may wonder: Is it possible to leverage the information content of huge data sets while keeping computational resources under control? Can this also help solve some of the privacy issues raised by large-scale learning? This is the ambition of compressive learning, where the data set is massively compressed before learning. Here, a "sketch" is first constructed by computing carefully chosen nonlinear random features [e.g., random Fourier (RF) features] and averaging them over the whole data set. Parameters are then learned from the sketch, without access to the original data set. This article surveys the current state of the art in compressive learning, including the main concepts and algorithms, their connections with established signal processing methods, existing theoretical guarantees on both information preservation and privacy preservation, and important open problems. For an extended version of this article that contains additional references and more in-depth discussions on a variety of topics, see [1].

CÁC LOẠI DỰ ÁN HỌC TẬP TRONG MÔN KĨ THUẬT Ở TIỂU HỌC

Project - based learning is a teaching model in which learners perform a complex learning task, with a combination of theory and practice to create a specific product. One of the advantages of this teaching model is to develop students' abilities, especially communication and cooperation, problem solving, self-study, and critical thinking... as the current educational goal is aiming. In the context of the transition from knowledge-oriented education to competency-oriented education, the research and application of project teaching in subjects in general and technical subjects in primary schools in particular has profound practical significance. By analyzing the concept, characteristics, project teaching process and systematizing the project classifications; Based on the characteristics of engineering subjects in elementary school, the article divides engineering subject projects into 4 types: inquiry project, research project, practice project, mixed project, concept clarification , characteristics, application process of each type of project and propose specific examples suitable to the characteristics of grade 4 and grade 5 technical subjects according to the approach to the 2018 Technology curriculum.

Vestibular Function in Children with Type 1 Diabetes: Videonystagmography Testing

Background: Vestibular system is critical for maintaining balance and learning complex tasks. This study aimed to determine the frequencies, types, and predictors of vestibular dysfunctions (VDs) in children with type 1 diabetes (T1D) using videonystagmography (VNG). Patients and Methods: This study included 65 patients (children with T1D = 40; controls = 25). The patients underwent VNG. Results: Patients (boys = 15; girls = 25) had a mean age of 14.05 ± 1.82 years and duration of illness of 6.30 ± 2.84 years. The majority had frequent attacks of diabetic ketoacidosis (DKA) (65%) and hypoglycemia (40%). Dizziness was reported in 20%. VNG abnormalities were reported in 70% (n = 28), of them 71.43 and 28.57% had central and peripheral VDs, respectively. Dizziness was associated with peripheral VD. Compared to patients without VDs, those with VDs were older and had earlier age at onset and longer duration of diabetes (>5 years), higher levels of HbA1c (>7%), higher frequencies of DKA and hypoglycemic attacks, comorbid medical conditions, and diabetic complications. Multiple logistic regression analysis showed that presence of VNG abnormalities (VDs) was independently correlated with diabetes duration >5 years (odds ratio [OR] = 4.52 [95% confidence interval [CI] = 3.55–7.04], p = 0.001), HbA1c% levels >7% (OR = 3.42 [95% CI = 2.84–5.75], p = 0.001), and presence of hypoglycemic attacks (OR = 4.65 [95% CI = 2.85–7.55]). Conclusions: -VDs are prevalent in children with T1D and correlated with the duration and severity of diabetes and the occurrence of hypoglycemic attacks. Therefore, optimizing glycemic control and prevention and treatment of diabetic complications and comorbidities are important. Multidisciplinary follow-ups are required for early detection and management of diabetic VDs.

Comparative cognition-Conceptual and methodological advancements.

This special issue originally placed a Call for Papers that emphasized the importance of "Conceptual and Methodological" advances in the field of Comparative Cognition. Represented here is a collection of 14 papers that helps to display some of the diversity of ideas and approaches within this flourishing research area. The first paper in this issue, by Gazes and Lazareva (2021), discusses transitive inference learning from the perspectives of: identifying the problems of contextual variables in studying different species; whether associative processes can or cannot fully account for the behavior and, if not, what alternative representational mechanisms might be at work; and, finally, how ecological considerations may support comparative research by suggesting novel theoretical and empirical questions. The next paper, by Loy et al. (2021) investigates questions related to the complexity of learning in invertebrate species, single-celled organisms, and plants. The paper by Rawlings et al. (2021) reviews the literature on cumulative cultural evolution, primarily in nonhuman primate species, and critically evaluates the importance of identifying the essential conceptual and methodological issues in what many have deemed to be a uniquely human form of behavior. The paper by Goto and Watanabe (2021) explores whether the mouse visual system is sensitive to Gestalt principles, using operant discrimination learning tasks similar to those used previously to document Gestalt processing in chimpanzees and humans. Qadri and Cook (2021) use the innovative approach of "adaptive genetic algorithms" to assess the relative importance of different features of a stimulus in controlling organisms' discrimination learning performance. Wittek et al. (2021) introduce a novel method for studying the importance of visual accumulation processes in pigeons when information is presented to a single hemisphere at a time. The paper by Cowie et al. (2021) focuses on a misallocation model of two-step sequence learning in young children and explores from a behavioranalytic viewpoint the implications of assuming that reinforcement might be misattributed to a misremembered response at the beginning of the behavioral sequence. The paper by López-Tolsa and Pellón (2021) explores whether the opportunity to display schedule-induced drinking as an early response within a behavioral sequence might alter the accuracy of temporal control in different-length fixed-interval schedule tasks with rats. Crystal (2021) reviews the literature examining episodic memory in nonhuman species and considers a variety of criteria and methods thought to be crucial for establishing empirical evidence for episodic memory in nonhumans, in general, and rats, in particular. Vila et al. (2021) discuss the use a novel 'hide-and-seek' task in preschool age children to study episodic-like memory; their work illustrates how memory dynamics can change over time in a manner not very unlike what has been demonstrated in other nonhuman research paradigms. The paper by Krichbaum et al. (2021) discusses some of the methodological difficulties one faces in studying spatial cognition in canines. The paper by Castro et al. (2021) uses a complex categorization learning task in which different sets of display features are diagnostic, or not, of category mastery depending upon context. The paper by Vernouillet et al. (2021) explores the formation of same/different concept learning in two species of corvids (pinyon jays and California scrub jays). The final paper, by Lazarowski et al. (2021) examines the possibility of abstract same/different learning in canines using a trial-unique training matching-to-sample procedure with olfactory stimuli. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Avalanches and edge-of-chaos learning in neuromorphic nanowire networks

The brain’s efficient information processing is enabled by the interplay between its neuro-synaptic elements and complex network structure. This work reports on the neuromorphic dynamics of nanowire networks (NWNs), a unique brain-inspired system with synapse-like memristive junctions embedded within a recurrent neural network-like structure. Simulation and experiment elucidate how collective memristive switching gives rise to long-range transport pathways, drastically altering the network’s global state via a discontinuous phase transition. The spatio-temporal properties of switching dynamics are found to be consistent with avalanches displaying power-law size and life-time distributions, with exponents obeying the crackling noise relationship, thus satisfying criteria for criticality, as observed in cortical neuronal cultures. Furthermore, NWNs adaptively respond to time varying stimuli, exhibiting diverse dynamics tunable from order to chaos. Dynamical states at the edge-of-chaos are found to optimise information processing for increasingly complex learning tasks. Overall, these results reveal a rich repertoire of emergent, collective neural-like dynamics in NWNs, thus demonstrating the potential for a neuromorphic advantage in information processing.

Burst-dependent synaptic plasticity can coordinate learning in hierarchical circuits.

Synaptic plasticity is believed to be a key physiological mechanism for learning. It is well established that it depends on pre- and postsynaptic activity. However, models that rely solely on pre- and postsynaptic activity for synaptic changes have, so far, not been able to account for learning complex tasks that demand credit assignment in hierarchical networks. Here we show that if synaptic plasticity is regulated by high-frequency bursts of spikes, then pyramidal neurons higher in a hierarchical circuit can coordinate the plasticity of lower-level connections. Using simulations and mathematical analyses, we demonstrate that, when paired with short-term synaptic dynamics, regenerative activity in the apical dendrites and synaptic plasticity in feedback pathways, a burst-dependent learning rule can solve challenging tasks that require deep network architectures. Our results demonstrate that well-known properties of dendrites, synapses and synaptic plasticity are sufficient to enable sophisticated learning in hierarchical circuits.

Real-Time Sound and Motion Feedback for Violin Bow Technique Learning: A Controlled, Randomized Trial.

The production of good sound generation in the violin is a complex task that requires coordination and spatiotemporal control of bowing gestures. The use of motion-capture technologies to improve performance or reduce injury risks in the area of kinesiology is becoming widespread. The combination of motion accuracy and sound quality feedback has the potential of becoming an important aid in violin learning. In this study, we evaluate motion-capture and sound-quality analysis technologies developed inside the context of the TELMI, a technology-enhanced music learning project. We analyzed the sound and bow motion of 50 participants with no prior violin experience while learning to produce a stable sound in the violin. Participants were divided into two groups: the experimental group (N = 24) received real-time visual feedback both on kinematics and sound quality, while participants in the control group (N = 26) practiced without any type of external help. An additional third group of violin experts performed the same task for comparative purposes (N = 15). After the practice session, all groups were evaluated in a transfer phase without feedback. At the practice phase, the experimental group improved their bowing kinematics in comparison to the control group, but this was at the expense of impairing the sound quality of their performance. At the retention phase, the experimental group showed better results in sound quality, especially concerning control of sound dynamics. Besides, we found that the expert group improved the stability of their sound while using the technology. All in all, these results emphasize the importance of feedback technologies in learning complex tasks, such as musical instrument learning.

Posttraumatic learning deficits correlate with initial trauma severity and chronic cellular reactions after closed head injury in male mice

Traumatic brain injury (TBI) is often associated with sustained attention and memory deficits. As persisting neuroinflammation and neurodegeneration may contribute to posttraumatic psychomotor dysfunction, we studied the relationship of brain cellular reactions three months after a weight-drop closed head injury in male mice with posttraumatic learning and memory using automated home-cage monitoring of socially housed mice in IntelliCages as well as tests for locomotor activity, anxiety and forepaw fine motor skills. One month after TBI, deficits in place learning and cognitive flexibility in reverse learning were clearly detectable in IntelliCages and these memory deficits correlated with the initial trauma severity on the functional neuroscore. While sucrose preference or its extinction were not influenced by TBI, traumatized mice performed significantly worse in a complex episodic memory learning task. In consecutive locomotor and forepaw skilled use tests, posttraumatic hyperactivity and impairment of contralateral paw use were evident. Analysis of cellular reactions to TBI three months after injury in selected defined regions of interest in the immediate lesion, ipsi- and contralateral frontoparietal cortex and hippocampus revealed a persistent microgliosis and astrogliosis which were accompanied by iron-containing macrophages and myelin degradation in the lesion area as well as with axonal damage in the neighboring cortical regions. Microglial and astroglial reactions in cortex showed a positive correlation with the initial trauma severity and a negative correlation with the spatial and episodic memory indicating a role of brain inflammatory reactions in posttraumatic memory deficits.

Understanding the relation between boredom and academic performance in postsecondary students.

Prior research has proposed that boredom and academic performance are reciprocally causal of one another. The present study sought to better understand the relationship between boredom and academic performance by, for the first time: distinguishing between boredom proneness, state boredom, and judgments of task boringness; conducting experiments in the laboratory where extraneous variables could be better controlled; and using experimental manipulation for causal conclusions. Study 1 examined the naturally occurring relationship between state boredom and performance on a word list recall task in the laboratory. Study 2 tested whether manipulating state boredom resulted in changes in word list recall, and Study 3 tested whether manipulating perceived word list recall resulted in changes in state boredom. State boredom and performance had a reciprocal relationship only for participants who memorized “interesting” word lists and only after repeated trials (Study 1); trait boredom predicted performance but state boredom did not (Study 2); and manipulating perceptions of performance had no effect on state boredom but did affect participants’ judgments of how boring the learning task was (Study 3). Thus, students seem to be able to weather changes in performance or boredom in the moment without one affecting the other. It is when the situation persists, or state boredom crystallizes into the form of a judgment (course-related boredom) or way of being (trait boredom), that problems emerge. Guidelines for educators are offered. Future research work is proposed, most pressingly work to replicate the current findings with more complex learning tasks. (PsycInfo Database Record (c) 2021 APA, all rights reserved)

How to train a complex skill in surgery: Qualitative evaluation of a simulation-based strategy

Background: Learning complex tasks in surgical requires the coordination and integration of technical and non-technical skills have an impact on the performance of work teams. Objective: The aim of this study is to report the results of a simulation-based educational strategy for training in complex surgical skills considering the participants’ perceptions. Material and methods: In 2019, 10 healthcare professionals participated in a 20-hour course divided in 6 hours of online training and 14 hours of onsite training. The strategy designed included the integration of case resolution activities, role-playing, practice with synthetic and virtual simulators and high-fidelity simulation. At the end of the course, a questionnaire was administered to explore participants’ perceptions on what they had learned and on their attitude changes. Results: Fifty percent of the participants perceived their skills and knowledge improved at the end of the course compared with their perception at the beginning of the course while 80% perceived the impact of the course on their professional activity was good or excellent. All the participants agreed with the need for improving non-technical skills. The experience was rated as positive or very positive by all participants, who were eager to repeat it. Conclusion: The participants’ perceptions of this educational program demonstrates that this method is highly accepted. Raising awareness of non-technical skills during the reflection stage suggests the need for changes in attitude and in self-perception of efficacy. We believe that simulation-based training offers the possibility of improving the overall performance of the surgical team. Future studies should focus on this goal.

Towards bridging the neuro-symbolic gap: deep deductive reasoners

Symbolic knowledge representation and reasoning and deep learning are fundamentally different approaches to artificial intelligence with complementary capabilities. The former are transparent and data-efficient, but they are sensitive to noise and cannot be applied to non-symbolic domains where the data is ambiguous. The latter can learn complex tasks from examples, are robust to noise, but are black boxes; require large amounts of –not necessarily easily obtained– data, and are slow to learn and prone to adversarial examples. Either paradigm excels at certain types of problems where the other paradigm performs poorly. In order to develop stronger AI systems, integrated neuro-symbolic systems that combine artificial neural networks and symbolic reasoning are being sought. In this context, one of the fundamental open problems is how to perform logic-based deductive reasoning over knowledge bases by means of trainable artificial neural networks. This paper provides a brief summary of the authors’ recent efforts to bridge the neural and symbolic divide in the context of deep deductive reasoners. Throughout the paper we will discuss strengths and limitations of models in term of accuracy, scalability, transferability, generalizabiliy, speed, and interpretability, and finally, will talk about possible modifications to enhance desirable capabilities. More specifically, in terms of architectures, we are looking at Memory-augmented networks, Logic Tensor Networks, and compositions of LSTM models to explore their capabilities and limitations in conducting deductive reasoning. We are applying these models on Resource Description Framework (RDF), first-order logic, and the description logic $\mathcal {E}{\mathscr{L}}^{+}$ respectively.

Distributed and Democratized Learning: Philosophy and Research Challenges

Due to the availability of huge amounts of data and processing abilities, current artificial intelligence (AI) systems are effective in solving complex tasks. However, despite the success of AI in different areas, the problem of designing AI systems that can truly mimic human cognitive capabilities such as artificial general intelligence, remains largely open. Consequently, many emerging cross-device AI applications will require a transition from traditional centralized learning systems towards large-scale distributed AI systems that can collaboratively perform multiple complex learning tasks. In this paper, we propose a novel design philosophy called democratized learning (Dem-AI) whose goal is to build large-scale distributed learning systems that rely on the self-organization of distributed learning agents that are wellconnected, but limited in learning capabilities. Correspondingly, inspired by the societal groups of humans, the specialized groups of learning agents in the proposed Dem-AI system are selforganized in a hierarchical structure to collectively perform learning tasks more efficiently. As such, the Dem-AI learning system can evolve and regulate itself based on the underlying duality of two processes which we call specialized and generalized processes. In this regard, we present a reference design as a guideline to realize future Dem-AI systems, inspired by various interdisciplinary fields. Accordingly, we introduce four underlying mechanisms in the design such as plasticity-stability transition mechanism, self-organizing hierarchical structuring, specialized learning, and generalization. Finally, we establish possible extensions and new challenges for the existing learning approaches to provide better scalable, flexible, and more powerful learning systems with the new setting of Dem-AI.

An Edge Computing‐Enabled Train Obstacle Detection Method Based on YOLOv3

Rail transit is developing towards intelligence which takes lots of computation resource to perform deep learning tasks. Among these tasks, object detection is the most widely used, like track obstacle detection, catenary wear, and defect detection and looseness detection of train wheel bolts. But the limited computation capability of the train onboard equipment prevents running deep and complex detection networks. The limited computation capability of the train onboard equipment prevents conducting complex deep learning tasks. Cloud computing is widely utilized to make up for the insufficient onboard computation capability. However, the traditional cloud computing architecture will bring in uncertain heavy traffic load and cause high transmission delay, which makes it fail to complete real‐time computing intensive tasks. As an extension of cloud computing, edge computing (EC) can reduce the pressure of cloud nodes by offloading workloads to edge nodes. In this paper, we propose an edge computing‐based method. The onboard equipment on a fast‐moving train is responsible for acquiring real‐time images and completing a small part of the inference task. Edge computing is used to help execute the object detection algorithm on the trackside and carry most of the computing power. YOLOv3 is selected as the object detection model, since it can balance between the real‐time and accurate performance on object detection compared with two‐stage models. To save onboard equipment computation resources and realize the edge‐train cooperative interface, we propose a model segmentation method based on the existing YOLOv3 model. We implement the cooperative inference scheme in real experiments and find that the proposed EC‐based object detection method can accomplish real‐time object detection tasks with little onboard computation resources.

Moments of Confusion in Simulation-Based Learning Environments

Confusion is an important epistemic emotion because it can help students focus their attention and effort when solving complex learning tasks. However, unresolved confusion can be detrimental because it may result in students’ disengagement. This is especially concerning in simulation environments using discovery-based learning, which puts more of the onus for learning on the students. Thus, students with misconceptions may become confused. In this study, the possible moments of confusion in a simulation-based predict-observe-explain (POE) environment were investigated. Log-based interaction patterns of undergraduate students from a fully online course were analyzed. It was found that POE environments can offer a level of difficulty that potentially triggers some confusion, and a likely moment of students’ confusion was the observe task. It was also found that confidence in prior knowledge is an important factor that can contribute to students’ confusion. Students mostly struggled when they discovered a mismatch between the subjective and objective correctness of their responses. The effects of such a mismatch were more pronounced when confusion markers were analyzed than when students’ learning outcomes were observed. These findings may guide future works to bridge the knowledge gaps that lead to confusion in POE environments.

Pseudo-rehearsal: Achieving deep reinforcement learning without catastrophic forgetting

Neural networks can achieve excellent results in a wide variety of applications. However, when they attempt to sequentially learn, they tend to learn the new task while catastrophically forgetting previous ones. We propose a model that overcomes catastrophic forgetting in sequential reinforcement learning by combining ideas from continual learning in both the image classification domain and the reinforcement learning domain. This model features a dual memory system which separates continual learning from reinforcement learning and a pseudo-rehearsal system that “recalls” items representative of previous tasks via a deep generative network. Our model sequentially learns Atari 2600 games without demonstrating catastrophic forgetting and continues to perform above human level on all three games. This result is achieved without: demanding additional storage requirements as the number of tasks increases, storing raw data or revisiting past tasks. In comparison, previous state-of-the-art solutions are substantially more vulnerable to forgetting on these complex deep reinforcement learning tasks.

Weight and bias initialization routines for Sigmoidal Feedforward Network

The success of the Sigmoidal Feedforward Networks in the solution of complex learning task can be attributed to their Universal Approximation Property. These networks are trained using non-linear iterative optimization method (of first-order or second-order) to solve a learning task. The convergence rate in Sigmoidal Feedforward Network training is affected by the initial choice of weights, therefore, in this paper, we propose two new weight initialization routines (Routine-1 and Routine-2) using characteristics of input and output data and property of activation function. Routine-1 uses the linear dependency of weight update step size on derivative of activation function and thus, initialize weights and bias to activate the activation function region near zero (input), where the derivative is maximum, therefore, increasing the weight update step size, and hence, the convergence speed. The same principle is used to derive Routine-2, that initialize weights and bias to activate distinct point in the significant range of activation function (where significant range defines the non-saturated region in activation function), such that, each node evolves independently of each other, and act as distinct feature identifier. Initializing weights in significant range reduces chances of (hidden) nodes getting stuck in saturated state. The networks initialized using proposed routines has higher convergence and higher probability to achieve deeper minima. The efficiency of proposed routines is evaluated by comparing them to conventional random weight initialization routine and 11 weight initialization routines proposed in literature (4 well established routines and 7 recently proposed routines) for several benchmark problems. The proposed routine is also tested for larger networks sizes and larger datasets such as MNIST. The results show that the performance of proposed routines is better than conventional random weight initialization routine and 11 established weight initialization routines.

Emotion-Performance Relationships in the Acquisition and Adaptation of a Complex Skill: Are Relationships Dynamic and Dependent on Activation Potential?

ABSTRACT Although the importance of emotions to self-regulation has been noted in the extant literature, little empirical research has examined how fluctuations across a range of emotions are related to performance. Accordingly, this laboratory study of 214 undergraduates (58% male) learning a complex videogame examined the dynamics of within-person emotion–performance relationships during skill acquisition and adaptation. We observed the direction of relationships was consistent with the valence of emotions at both the within- and between-person levels. In addition, we observed dynamic relationships and an interplay between valence and activation potential. Dynamic relationships were observed for negative but not positive emotions. Activation potential made a difference for positive but not negative emotions. Implications for emotion control strategies in complex task learning are discussed.

Zero-shot learning and its applications from autonomous vehicles to COVID-19 diagnosis: A review

The challenge of learning a new concept, object, or a new medical disease recognition without receiving any examples beforehand is called Zero-Shot Learning (ZSL). One of the major issues in deep learning based methodologies such as in Medical Imaging and other real-world applications is the requirement of large annotated datasets prepared by clinicians or experts to train the model. ZSL is known for having minimal human intervention by relying only on previously known or trained concepts plus currently existing auxiliary information. This is ever-growing research for the cases where we have very limited or no annotated datasets available and the detection / recognition system has human-like characteristics in learning new concepts. This makes the ZSL applicable in many real-world scenarios, from unknown object detection in autonomous vehicles to medical imaging and unforeseen diseases such as COVID-19 Chest X-Ray (CXR) based diagnosis. In this review paper, we introduce a novel and broaden solution called Few / one-shot learning, and present the definition of the ZSL problem as an extreme case of the few-shot learning. We review over fundamentals and the challenging steps of Zero-Shot Learning, including state-of-the-art categories of solutions, as well as our recommended solution, motivations behind each approach, their advantages over each category to guide both clinicians and AI researchers to proceed with the best techniques and practices based on their applications. Inspired from different settings and extensions, we then review through different datasets inducing medical and non-medical images, the variety of splits, and the evaluation protocols proposed so far. Finally, we discuss the recent applications and future directions of ZSL. We aim to convey a useful intuition through this paper towards the goal of handling complex learning tasks more similar to the way humans learn. We mainly focus on two applications in the current modern yet challenging era: coping with an early and fast diagnosis of COVID-19 cases, and also encouraging the readers to develop other similar AI-based automated detection / recognition systems using ZSL.

Images of Learning through Music: The Sounds of Cognition

This innovative practice uses classical string instruments as the tools, and music as the medium for exploring learning new and complex tasks. In a face-to-face setting people experience playing violins, violas, and cellos as their learning is guided by visual and aural imagery, enabling engagement with complex cognitive and affective concepts through what appears to be play. By accessing concept knowledge directly through representation and then experience, the need for translation of technical or theoretical material is bypassed. The progressive introduction of mental and physical skills actively illustrates the cumulative impact of the cognitive, the coordination, and the interactive demands of playing the instruments. Active reflection in and on the experience of playing a musical instrument, working collaboratively, and allowing and enjoying public learning (and failure) is contextualised in terms of people’s self-efficacy beliefs, their attitudes toward learning and skill development, and the relevant transferability of these realised and learned ideas to their personal work and life situations.

CSLM: Convertible Short-Term and Long-Term Memory in Differential Neural Computers

External memory-based neural networks, such as differentiable neural computers (DNCs), have recently gained importance and popularity to solve complex sequential learning tasks that pose challenges to conventional neural networks. However, a trained DNC usually has a low-memory utilization efficiency. This article introduces a variation of DNC architecture with a convertible short-term and long-term memory, named CSLM-DNC. Unlike the memory architecture of the original DNC, the new scheme of short-term and long-term memories offers different importance of memory locations for read and write, and they can be converted over time. This is mainly motivated by the human brain where short-term memory stores large amounts of noisy and unimportant information and decays rapidly, while long-term memory stores important information and lasts for a long time. The conversion of these two types of memory is allowed and is able to be learned according to their reading and writing frequency. We quantitatively and qualitatively evaluate the proposed CSLM-DNC architecture on the tasks of question answering, copy and repeat copy, showing that it can significantly improve memory efficiency and learning performance.