Short-term Memory Encoding Research Articles

Deep encoder–decoder network based data-driven method for impact feedback rendering on head during earthquake

In safety training simulators, realistic haptic feedback is essential to make people construct accurate situation awareness through experiencing. In this regard, this paper presents a new and innovative system that provides the haptic experience of falling objects on user’s head during an earthquake. Special focus was on the accurate reproduction of impact feedback when various objects fall on the head. To this end, we propose a novel data-driven approach. This approach first collects 3-axis acceleration signals during real collision under several impact velocities. Afterward, 3D acceleration data is abstracted to a 1D acceleration profile using our novel max–min extraction approach. The impact signal for an arbitrary velocity is interpolated using a deep convolutional bidirectional long short-term memory encoder–decoder model. Rendering hardware is also implemented using high performance voice-coil vibrotactile actuator. Numerical and subjective evaluations are carried out to evaluate the performance of the proposed approach.Kindly check and confirm the edit made in the title.I confirm the edit is okay.Please confirm if the author names are presented accurately and in the correct sequence (given name, middle name/initial, family name). Authors Given name: [Joolekha Bibi] Last name: [Joolee], Given name: [Mohammad Shadman] Last name: [Hashem]. Also, kindly confirm the details in the metadata are correct.Yes, the author names are presented accurately and in the correct sequence.

Self-Organizing Neural Scheduler for the Flexible Job Shop Problem With Periodic Maintenance and Mandatory Outsourcing Constraints.

Scheduling is significant in improving the production efficiency and reducing delivery delays for manufacturing enterprises. Unlike the flexible job-shop scheduling problem, two special constraints are encountered in real-world power supply manufacturing systems: 1) periodic maintenance and 2) mandatory outsourcing. As the characteristics of these constraints are not considered in existing scheduling algorithms, schedules generated by most existing approaches are not optimal or even conflict with these constraints. In this article, a self-organizing neural scheduler (SoNS) is proposed to overcome this limitation. A long short-term memory encoder is developed to transform the variable-length structural information into fixed-length feature vectors. Moreover, the reinforcement learning model is proposed to automatically select policies for improving candidate schedules. To validate the effectiveness of the proposed algorithm, extensive experiments are conducted on over 300 problem instances. The nonparametric Kruskal-Wallis tests confirm that the proposed algorithm outperforms several state-of-the-art methods in terms of effectiveness and robustness within a limited computational budget. It demonstrates that the proposed SoNS can solve scheduling problems with the periodic maintenance and mandatory outsourcing constraints effectively.

Day-ahead forecasting of residential electric power consumption for energy management using Long Short-Term Memory encoder–decoder model

Energy management in smart buildings and energy communities needs short-term load demand forecasting for optimization-based scheduling, dispatch, and real-time operation. However, producing accurate forecasting for individual residential households is more challenging compared to the forecasting of load demand at the distribution level, which is smoother and benefits from statistical compensation of errors. This paper presents a day-ahead forecasting technique for individual residential load demand that is based on the Long Short-Term Memory encoder–decoder architecture, which is extended to consider possibly differing sets of past and future exogenous variables. A novel focus is posed on the validation of the proposed approach considering that it is tailored for use by energy management systems. A publicly available dataset was used for validation, and the approach was compared with three other methods, resulting in a reduction of the Mean Absolute Scaled Error by up to 8%.

Predicting drivers’ route trajectories in last-mile delivery using a pair-wise attention-based pointer neural network

In last-mile delivery, drivers frequently deviate from planned delivery routes because of their tacit knowledge of the road and curbside infrastructure, customer availability, and other characteristics of the respective service areas. Hence, the actual stop sequences chosen by an experienced human driver may be potentially preferable to the theoretical shortest-distance routing under real-life operational conditions. Thus, being able to predict the actual stop sequence that a human driver would follow can help to improve route planning in last-mile delivery. This paper proposes a pair-wise attention-based pointer neural network for this prediction task using drivers’ historical delivery trajectory data. In addition to the commonly used encoder–decoder architecture for sequence-to-sequence prediction, we propose a new attention mechanism based on an alternative specific neural network to capture the local pair-wise information for each pair of stops. To further capture the global efficiency of the route, we propose a new iterative sequence generation algorithm that is used after model training to identify the first stop of a route that yields the lowest operational cost. Results from an extensive case study on real operational data from Amazon’s last-mile delivery operations in the US show that our proposed method can significantly outperform traditional optimization-based approaches and other machine learning methods (such as the Long Short-Term Memory encoder–decoder and the original pointer network) in finding stop sequences that are closer to high-quality routes executed by experienced drivers in the field. Compared to benchmark models, the proposed model can increase the average prediction accuracy of the first four stops from around 0.229 to 0.312, and reduce the disparity between the predicted route and the actual route by around 15%.

Long–short-term memory encoder–decoder with regularized hidden dynamics for fault detection in industrial processes

The ability of recurrent neural networks (RNN) to model nonlinear dynamics of high dimensional process data has enabled data-driven RNN-based fault detection algorithms. Previous studies have focused on detecting faults by identifying the discrepancies in data distribution between the faulty and normal data, as reflected in prediction errors generated by RNN models. However, in industrial processes, variations in data distribution can also result from changes in normal control setpoints and compensatory control adjustments in response to disturbances, making it hard to differentiate between normal and faulty conditions. This paper proposes a fault detection method utilizing a long short-term memory (LSTM) encoder–decoder structure with regularized hidden dynamics and reversible instance normalization (RevIN) to compactly represent high-dimensional measurements for effective monitoring. During training, the hidden states of the model are regularized to form a low-dimensional latent space representation of the original multivariate time series data. As a result, the prediction errors of the latent states can be used to monitor the abnormal dynamic variations, while the reconstruction errors of the measured variables are used to monitor the abnormal static variations. Furthermore, the proposed indices can reflect operating conditions, even when the distribution of test data changes, which helps distinguish faults from normal adjustments and disturbances that controllers can settle. Data from numerical simulation and the Tennessee Eastman process are used to illustrate the effectiveness of the proposed fault detection method.

Nowcasting extreme rain and extreme wind speed with machine learning techniques applied to different input datasets

Predicting extreme weather events in a short time period and their developing in localized areas is a challenge. The nowcasting of severe and extreme weather events is an issue for air traffic management and control because it affects aviation safety, and determines delays and diversions. This work is part of a larger study devoted to nowcasting rain and wind speed in the area of Malpensa airport by merging different datasets. We use as reference the weather station of Novara to develop a nowcasting machine learning model which could be reusable in other locations. In this location we have the availability of ground-based weather sensors, a Global Navigation Satellite System (GNSS) receiver, a C-band radar and lightning detectors. Our analysis shows that the Long Short-Term Memory Encoder Decoder (LSTM E/D) approach is well suited for the nowcasting of meteorological variables. The predictions are based on 4 different datasets configurations providing rain and wind speed nowcast for 1 h with a time step of 10 min. The results are very promising with the extreme wind speed probability of detection higher than 90%, the false alarms lower than 2%, and a good performance in extreme rain detection for the first 30 min. The configuration using just weather stations and GNSS data in input provides excellent performances and should be preferred to the other ones, since it refers to the pre-convective environment, and thus can be adaptable to any weather conditions.

How musical rhythm training improves short-term memory for faces

Playing a musical instrument engages numerous cognitive abilities, including sensory perception, selective attention, and short-term memory. Mounting evidence indicates that engaging these cognitive functions during musical training will improve performance of these same functions. Yet, it remains unclear the extent these benefits may extend to nonmusical tasks, and what neural mechanisms may enable such transfer. Here, we conducted a preregistered randomized clinical trial where nonmusicians underwent 8 wk of either digital musical rhythm training or word search as control. Only musical rhythm training placed demands on short-term memory, as well as demands on visual perception and selective attention, which are known to facilitate short-term memory. As hypothesized, only the rhythm training group exhibited improved short-term memory on a face recognition task, thereby providing important evidence that musical rhythm training can benefit performance on a nonmusical task. Analysis of electroencephalography data showed that neural activity associated with sensory processing and selective attention were unchanged by training. Rather, rhythm training facilitated neural activity associated with short-term memory encoding, as indexed by an increased P3 of the event-related potential to face stimuli. Moreover, short-term memory maintenance was enhanced, as evidenced by increased two-class (face/scene) decoding accuracy. Activity from both the encoding and maintenance periods each highlight the right superior parietal lobule (SPL) as a source for training-related changes. Together, these results suggest musical rhythm training may improve memory for faces by facilitating activity within the SPL to promote how memories are encoded and maintained, which can be used in a domain-general manner to enhance performance on a nonmusical task.

Post-Authorship Attribution Using Regularized Deep Neural Network

Post-authorship attribution is a scientific process of using stylometric features to identify the genuine writer of an online text snippet such as an email, blog, forum post, or chat log. It has useful applications in manifold domains, for instance, in a verification process to proactively detect misogynistic, misandrist, xenophobic, and abusive posts on the internet or social networks. The process assumes that texts can be characterized by sequences of words that agglutinate the functional and content lyrics of a writer. However, defining an appropriate characterization of text to capture the unique writing style of an author is a complex endeavor in the discipline of computational linguistics. Moreover, posts are typically short texts with obfuscating vocabularies that might impact the accuracy of authorship attribution. The vocabularies include idioms, onomatopoeias, homophones, phonemes, synonyms, acronyms, anaphora, and polysemy. The method of the regularized deep neural network (RDNN) is introduced in this paper to circumvent the intrinsic challenges of post-authorship attribution. It is based on a convolutional neural network, bidirectional long short-term memory encoder, and distributed highway network. The neural network was used to extract lexical stylometric features that are fed into the bidirectional encoder to extract a syntactic feature-vector representation. The feature vector was then supplied as input to the distributed high networks for regularization to minimize the network-generalization error. The regularized feature vector was ultimately passed to the bidirectional decoder to learn the writing style of an author. The feature-classification layer consists of a fully connected network and a SoftMax function to make the prediction. The RDNN method was tested against thirteen state-of-the-art methods using four benchmark experimental datasets to validate its performance. Experimental results have demonstrated the effectiveness of the method when compared to the existing state-of-the-art methods on three datasets while producing comparable results on one dataset.

Fast Modeling of the Capacity Degradation of Lithium-Ion Batteries via a Conditional Temporal Convolutional Encoder–Decoder

Recently, deep neural networks have been successfully applied to battery degradation modeling, and onboard graphic processing units can potentially be used to host these models. However, most existing degradation models rely on recursive operations, which are slow on hardware optimized for vectorized computations. This article proposes a two-encoder, one-decoder conditional temporal convolutional network that uses no recursive operations, thus allowing for the conditional modeling of battery degradation and improving upon the prediction speed of long short-term memory encoder–decoder by a factor of up to 70 for long tests. Datasets containing 115 batteries from multiple sources and 195 000 steps are used to verify the proposed model. The data are augmented by randomly selecting the prediction starting point during training. The proposed model achieves an average prediction error of less than 4% of the maximum available capacity based on large datasets. In addition, prediction results and convolution weight analysis indicate that the model can learn the temporal and conditional dynamics of typical lithium-ion batteries and is robust to the selection of the prediction starting point. We further demonstrate a prediction under hypothetical future conditions and a Monte Carlo prediction under unknown future conditions, and the degradation trends are accurately predicted.

EEG-based investigation of the impact of room size and window placement on cognitive performance

This study investigated changes in scalp electroencephalography (EEG) features associated with short-term exposure to four virtual classroom designs, with different window placement and room dimensions. Participants engaged in five cognitive tests in each design condition: the Stroop Test, the Digit Span Test, the Benton Test, a Visual Memory Test, and an Arithmetic Test. Performance on the cognitive tests and EEG data were analyzed by contrasting various classroom design conditions. The cognitive test performance results showed no significant differences related to the design changes in the room studied. We computed frequency band-power and connectivity EEG features to identify changes in neural patterns associated to the different design conditions. A leave-one-out machine-learning classification scheme was implemented to assess the robustness of the EEG features, with the classification accuracy evaluation of the trained model iteratively performed against an unseen participant’s data for the test set. The feature selection and classification results located consistent differences in the EEG features that held predictive power (p<0.003) across participants in the different classroom design conditions: a. Neural vs One-Window, b. Neutral vs Two-Windows, c. Neutral vs Wide room. This effect was observed for the tasks that required short-term memory encoding, including the Digit Span Test (median test set classification accuracy: a. 53.8%, b. 51.3%), the Benton Test (a. 61.3%, b. 56.3%), and the Visual Memory Test (a. 55.0%, b. 57.5%, c. 61.3%). The most discriminative EEG features were primarily frequency band-power features, observed in bilateral occipital, parietal, and frontal regions in the theta (4-8 Hz) and alpha (8-12 Hz) frequency bands. The connectivity features reinforced these findings by showing that there were changes in the transfer of information from centro-parietal to frontal electrodes in the different classroom conditions. This study provides rigorous evidence that brain activity features during cognitive tasks are affected by the design elements of window placement and room dimensions in a virtual classroom. The ongoing development of this EEG-based approach has the potential to strengthen evidence-based design through evaluation of robust neurophysiological evidence.

Prediction of Yangtze River streamflow based on deep learning neural network with El Ni\xf1o\u2013Southern Oscillation

Accurate long-term streamflow and flood forecasting have always been an important research direction in hydrology research. Nowadays, climate change, floods, and other anomalies occurring more and more frequently and bringing great losses to society. The prediction of streamflow, especially flood prediction, is important for disaster prevention. Current hydrological models based on physical mechanisms can give accurate predictions of streamflow, but the effective prediction period is only about 1 month in advance, which is too short for decision making. The artificial neural network (ANN) has great potential for predicting runoff and is not only good at handling non-linear data but can also make long-period forecasts. However, most of ANN models are unstable in their predictions when faced with raw flow data, and have excessive errors in predicting extreme flows. Previous studies have shown a link between the El Niño–Southern Oscillation (ENSO) and the streamflow of the Yangtze River. In this paper, we use ENSO and the monthly streamflow data of the Yangtze River from 1952 to 2018 to predict the monthly streamflow of the Yangtze River in two extreme flood years and a small flood year by using deep neural networks. In this paper, three deep neural network frameworks are used: stacked long short-term memory, Conv long short-term memory encoder–decoder long short-term memory and Conv long short-term memory encoder–decoder gate recurrent unit. The results show that the use of ConvLSTM improves the stability of the model and increases the accuracy of the flood prediction. Besides, the introduction of ENSO to the experimental data resulted in a more accurate prediction of the time of the occurrence of flood peaks and flood flows. Furthermore, the best results were obtained on the convolutional long short-term memory + encoder–decoder gate recurrent unit model.

Speed Adaptability Assessment of Railway Balise Transmission Module Using a Deep-Adaptive-Attention-Based Encoder–Decoder Network

The accurate forecast of the number of effective telegram frames (NETF) received by the balise transmission module (BTM) at different speeds is the key to assessing its speed adaptability. However, the NETF index has a complex correlation with multiple BTM transmission parameters and exhibits highly nonlinear variations in the speed dimension, which poses a challenge to the NETF series prediction. Thus, a deep-adaptive-attention-based model is proposed in this article, which combines two enhanced attention mechanisms into the long short-term memory encoder–decoder network to thoroughly explore the spatial-speed relationship between the historical multivariate series and the future NETF series. Specifically, the proposed model consists of an encoder with a dual-phase spatial-attention mechanism and a decoder with an adaptive speed-attention mechanism, which can jointly capture the stable spatial correlation and relevant long-term speed-dependent information of the input multivariate series across all speed steps for the robust and accurate prediction of the NETF series. Experimental results on real-world BTM datasets demonstrate that the proposed model can predict the NETF series with the lowest local predicted error and highest overall prediction accuracy, compared with other state-of-the-art attention-based methods.

Anomaly detection based on machine learning in IoT-based vertical plant wall for indoor climate control

Indoor climate is closely related to human health, comfort and productivity. Vertical plant wall systems, embedded with sensors and actuators, have become a promising application for indoor climate control. In this study, we explore the possibility of applying machine learning based anomaly detection methods to vertical plant wall systems so as to enhance the automation and improve the intelligence to realize predictive maintenance of the indoor climate. Two categories of anomalies, namely point anomalies and contextual anomalies are researched. Prediction-based and pattern recognition-based methods are investigated and applied to indoor climate anomaly detection. The results show that neural network-based models, specifically the autoencoder (AE) and the long short-term memory encoder decoder (LSTM-ED) model surpass the others in terms of detecting point anomalies and contextual anomalies, respectively, therefore can be deployed into vertical plant walls systems in industrial practice. Based on the results, a new data cleaning method is proposed and a prediction-based method is deployed to the cloud in practice as a proof-of-concept. This study showcases the advancements in machine learning and Internet of things can be fully utilized by researches on building environment to accelerate the solution development.

Mimicking Short-Term Memory in Shape-Reconstruction Task Using an EEG-Induced Type-2 Fuzzy Deep Brain Learning Network

The paper attempts to model short-term memory (STM) for shape-reconstruction tasks by employing a 4-stage deep brain leaning network (DBLN), where the first two stages are built with Hebbian learning and the last two stages with Type-2 Fuzzy logic. The model is trained stage-wise independently with visual stimulus of the object-geometry as the input of the first stage, EEG acquired from different cortical regions as input and output of respective intermediate stages, and recalled object-geometry as the output of the last stage. Two error feedback loops are employed to train the proposed DBLN. The inner loop adapts the weights of the STM based on a measure of error in model-predicted response with respect to the object-shape recalled by the subject. The outer loop adapts the weights of the iconic (visual) memory based on a measure of error of the model predicted response with respect to the desired object-shape. In the test phase, the DBLN model reproduces the recalled object shape from the given input object geometry. The motivation of the paper is to test the consistency in STM encoding (in terms of similarity in network weights) for repeated visual stimulation with the same geometric object. Experiments undertaken on healthy subjects, yield high similarity in network weights, whereas patients with pre-frontal lobe Amnesia yield significant discrepancy in the trained weights for any two trials with the same training object. This justifies the importance of the proposed DBLN model in automated diagnosis of patients with learning difficulty. The novelty of the paper lies in the overall design of the DBLN model with special emphasis to the last two stages of the network, built with vertical slice based type-2 fuzzy logic, to handle uncertainty in function approximation (with noisy EEG data). The proposed technique outperforms the state-of-the-art functional mapping algorithms with respect to the (pre-defined outer loop) error metric, computational complexity and runtime.

Host load prediction in cloud computing using Long Short-Term Memory Encoder–Decoder

Cloud computing has been developed as a means to allocate resources efficiently while maintaining service-level agreements by providing on-demand resource allocation. As reactive strategies cause delays in the allocation of resources, proactive approaches that use predictions are necessary. However, due to high variance of cloud host load compared to that of grid computing, providing accurate predictions is still a challenge. Thus, in this paper we have proposed a prediction method based on Long Short-Term Memory Encoder–Decoder (LSTM-ED) to predict both mean load over consecutive intervals and actual load multi-step ahead. Our LSTM-ED-based approach improves the memory capability of LSTM, which is used in the recent previous work, by building an internal representation of time series data. In order to evaluate our approach, we have conducted experiments using a 1-month trace of a Google data centre with more than twelve thousand machines. Our experimental results show that while multi-layer LSTM causes overfitting and decrease in accuracy compared to single-layer LSTM, which was used in the previous work, our LSTM-ED-based approach successfully achieves higher accuracy than other previous models, including the recent LSTM one.

Attention guided deep audio-face fusion for efficient speaker naming

Speaker naming has recently received considerable attention in identifying the active speaking character in a movie video, and face cue alone is generally insufficient to achieve reliable performance due to its significant appearance variations. In this paper, we treat the speaker naming task as a group of matched audio-face pair finding problems, and present an efficient attention guided deep audio-face fusion approach to detect the active speakers. First, we start with VGG-encoding of face images and extract the Mel-Frequency Cepstrum Coefficients from audio signals. Then, two efficient audio encoding modules, namely two-layer Long Short-Term Memory encoding and two-dimensional convolution encoding, are addressed to discriminate the high-level audio features. Meanwhile, we train an end-to-end audio-face common attention model to discriminate the face attention vector, featuring adaptively to accommodate various face variations. Further, an efficient factorized bilinear model is presented to deeply fuse the paired audio-face features, whereby the joint audio-face representation can be reliably obtained for speaker naming. Extensive experiments highlight the superiority of the proposed approach and show its very competitive performance with the state-of-the-arts.

An ensemble code in medial prefrontal cortex links prior events to outcomes during learning

The prefrontal cortex is implicated in learning the rules of an environment through trial and error. But it is unclear how such learning is related to the prefrontal cortex’s role in short-term memory. Here we ask if the encoding of short-term memory in prefrontal cortex is used by rats learning decision rules in a Y-maze task. We find that a similar pattern of neural ensemble activity is selectively recalled after reinforcement for a correct decision. This reinforcement-selective recall only reliably occurs immediately before the abrupt behavioural transitions indicating successful learning of the current rule, and fades quickly thereafter. We could simultaneously decode multiple, retrospective task events from the ensemble activity, suggesting the recalled ensemble activity has multiplexed encoding of prior events. Our results suggest that successful trial-and-error learning is dependent on reinforcement tagging the relevant features of the environment to maintain in prefrontal cortex short-term memory.

Developing a hippocampal neural prosthetic to facilitate human memory encoding and recall

Objective. We demonstrate here the first successful implementation in humans of a proof-of-concept system for restoring and improving memory function via facilitation of memory encoding using the patient’s own hippocampal spatiotemporal neural codes for memory. Memory in humans is subject to disruption by drugs, disease and brain injury, yet previous attempts to restore or rescue memory function in humans typically involved only nonspecific, modulation of brain areas and neural systems related to memory retrieval. Approach. We have constructed a model of processes by which the hippocampus encodes memory items via spatiotemporal firing of neural ensembles that underlie the successful encoding of short-term memory. A nonlinear multi-input, multi-output (MIMO) model of hippocampal CA3 and CA1 neural firing is computed that predicts activation patterns of CA1 neurons during the encoding (sample) phase of a delayed match-to-sample (DMS) human short-term memory task. Main results. MIMO model-derived electrical stimulation delivered to the same CA1 locations during the sample phase of DMS trials facilitated short-term/working memory by 37% during the task. Longer term memory retention was also tested in the same human subjects with a delayed recognition (DR) task that utilized images from the DMS task, along with images that were not from the task. Across the subjects, the stimulated trials exhibited significant improvement (35%) in both short-term and long-term retention of visual information. Significance. These results demonstrate the facilitation of memory encoding which is an important feature for the construction of an implantable neural prosthetic to improve human memory.

P 17 Neural signatures of post-perceptual attention and selective working memory encoding

Post-perceptual cues can enhance visual short term memory encoding even after the offset of the visual stimulus. However, both the mechanisms by which the sensory stimulus characteristics are buffered as well as the mechanisms by which post-perceptual selective attention enhances short term memory encoding remain unclear. We analyzed late post-perceptual event-related potentials (ERP) in visual change detection tasks (100 ms stimulus duration) by high-resolution ERP analysis to elucidate these mechanisms. The effects of auditory post-cues (300 ms or 850 ms after visual stimulus onset) as well as the effects of a visual interference stimulus were examined in 27 healthy right-handed adults. Focusing attention with post-perceptual cues at both latencies significantly improved memory performance, i.e. sensory stimulus characteristics were available for up to 850 ms after stimulus presentation. Passive watching of the visual stimuli evoked a slow negative wave over occipito-temporal visual areas. In the memory task, this N700 was strongly reduced by a visual interference stimulus which impeded memory maintenance. Therefore N700 could reflect visual post-processing by automatic attention attraction – when visual post-processing is not impeded by interference effects. In contrast, contralateral delay activity (CDA) still developed after the application of auditory post-cues in the retroactive interference condition. Thus CDA may reflect the top-down process of searching selectively for the required information through post-perceptual attention. N700 and CDA could be dissociated by retroactive interference effects and seem to represent two different processes involved in short term memory encoding. We discuss the meaning of cortical slow negative waves during post-perceptual attention changes in different modalities (olfactory, auditory, somatosensory, motor/proprioceptive, visual) in order to elucidate general mechanisms of the selective encoding of sensory features in working Hecht et al., 2016 , Lenk et al., 2014 , Bender et al., 2014 , Bender et al., 2012 , Bender et al., 2012 , Thiemann et al., 2012 , Bender et al., 2010 .

Mathematical modelling of ICAN-mediated persistent firing in hippocampal neurons

Persistent neural activity has been the focus of neuroscientific research since it was first associated with complex cognitive behaviours. In particular, persistent firing has long been thought to be the neural mechanisms underlying short-term memory encoding and storage [1]. This activity is often elicited by short transient stimuli that have to be retained in memory for long delay periods, in the order of several seconds, after the original stimulus disappeared. In this scenario, the brain stores information for future execution of action depending on that information. Persistent activity elicited in a recurrent network comprising strong excitation in the local circuit has been extensively described in the literature (for a review see [2]). However, recent findings have shown that memory can also be encoded in brain regions which do not display such recurrent connection topology, including the CA1 hippocampal area [3]. Emerging studies are pointing towards intrinsic neural mechanisms independent of synaptic connections, as a complementary mechanism for the maintenance of persistent activity in the hippocampus [4]. These include various cytoplasmic currents flowing through the membrane, characterised by slow ion channel kinetics, and particular neurotransmitter modulation. Our work investigates persistent firing activity in networks of hippocampal neurons, elicited by leveraging intrinsic currents rather than network dynamics. Building on previous findings [5], we model a hippocampal pyramidal neuron using the Hodgkin-Huxley model, with low-threshold Ca2+ currents governing the inward flow of calcium ions in the cell membrane. The intracellular calcium concentration mediates the opening of calcium-activated non-specific (CAN) ion channels [6], which causes an increase in the ionic flow through the cell membrane. Therefore the CAN channels equip the neuron with an after-spike depolarisation mechanism, which enables it to emit action potentials in the absence of external stimulation. Given a transient 250ms 200pA current injection, the neuron is capable of eliciting and maintaining persistent activity with a firing rate of 6Hz for long delay periods (~30s). Moreover, this behaviour is in accord with that displayed in neural recordings of hippocampal slice preparations [4]. Connecting these persistent firing neurons in a network comprising strong local excitation yields a wide range of behaviours depending on the interaction between CAN and synaptic currents. Indeed, our network model is capable of displaying rhythmic behaviour in the form of short synchronised bursts with intra-burst frequencies of 20-40Hz and inter-burst frequencies of 3Hz. These results hint towards a possible mechanism for the generation of memory-related oscillatory activity in the hippocampus.