Deep Encoding Research Articles

The Effectiveness of Item-Specific Encoding and Conservative Responding to Reduce False Memories in Patients with Mild Cognitive Impairment and Mild Alzheimer's Disease Dementia.

Patients with mild Alzheimer's disease dementia are more susceptible to false memories than healthy older adults. Evidence that these patients can use cognitive strategies to reduce false memory is inconsistent. In the present study, we examined the effectiveness of conservative responding and item-specific deep encoding strategies, alone and in combination, to reduce false memory in a categorized word list paradigm among participants with mild Alzheimer's disease dementia (AD), amnestic single-domain mild cognitive impairment (MCI), and healthy age-matched older controls (OCs). A battery of clinical neuropsychological measures was also administered. Although use of conservative responding alone tended to reduce performance in the MCI and OC groups, both deep encoding alone and deep encoding combined with conservative strategies led to improved discrimination for both gist memory and item-specific recollection for these two groups. In the AD group, only gist memory benefited from the use of strategies, boosted equally by deep encoding alone and deep encoding combined with conservative strategies; item-specific recollection was not improved. No correlation between the use of these strategies and performance on neuropsychological measures was found. These results suggest that further evaluation of these strategies is warranted as they have the potential to reduce related and unrelated memory errors and increase both gist memory and item-specific recollection in healthy older adults and individuals with amnestic MCI. Patients with AD were less able to benefit from such strategies, yet were still able to use them to reduce unrelated memory errors and increase gist memory.

0084 Sleep Consolidates Incidentally Encoded Information After Deep but Not Shallow Encoding

Abstract Introduction Slow wave sleep (SWS) strengthens declarative memory for information studied for a later test. However, research on the effect of sleep on information that is not intentionally remembered is scare. Previous research from our lab suggests sleep consolidates some, but not all, information that has been encoded incidentally, meaning that it has been acted on but not intentionally remembered. It remains unclear what determines which information benefits from sleep-dependent consolidation processes and what aspects of sleep are related to these mnemonic benefits. In two experiments, we test the hypothesis that sleep consolidates strong but not weak memory traces following incidental encoding, and assess the relationship between memory performance and objective sleep characteristics. Methods In Experiment 1, participants rated words one (weak traces) or three times (strong traces) in a deep or shallow incidental encoding task. Participants either rated words on a scale from ‘concrete’ to ‘abstract’ (deep) or counted the vowels in the words (shallow). Following a 12-hour period containing sleep or wakefulness, participants took a surprise memory test. In Experiment 2, participants rated words one or three times in the deep encoding task, received an 8-hour sleep opportunity with polysomnography, and took the surprise memory test. Results In Experiment 1, participants remembered words better after sleep than wake regardless of whether words were encoded one or three times, but only after deep encoding. Sleep did not consolidate information following shallow encoding. Experiment 2 is ongoing, but we predict that the amount of SWS will correlate positively with memory. Conclusion Results thus far suggest sleep may have consolidated information based on the strength of memory traces. Because deep encoding results in stronger memory traces than shallow encoding, this work is broadly consistent with theories of memory consolidation that predict sleep is more beneficial for strong memory traces than weak, such as the synaptic downscaling hypothesis. Support N/A

Single Residential Load Forecasting Using Deep Learning and Image Encoding Techniques

The integration of more renewable energy resources into distribution networks makes the operation of these systems more challenging compared to the traditional passive networks. This is mainly due to the intermittent behavior of most renewable resources such as solar and wind generation. There are many different solutions being developed to make systems flexible such as energy storage or demand response. In the context of demand response, a key factor is to estimate the amount of load over time properly to better manage the demand side. There are many different forecasting methods, but the most accurate solutions are mainly found for the prediction of aggregated loads at the substation or building levels. However, more effective demand response from the residential side requires prediction of energy consumption at every single household level. The accuracy of forecasting loads at this level is often lower with the existing methods as the volatility of single residential loads is very high. In this paper, we present a hybrid method based on time series image encoding techniques and a convolutional neural network. The results of the forecasting of a real residential customer using different encoding techniques are compared with some other existing forecasting methods including SVM, ANN, and CNN. Without CNN, the lowest mean absolute percentage of error (MAPE) for a 15 min forecast is above 20%, while with existing CNN, directly applied to time series, an MAPE of around 18% could be achieved. We find the best image encoding technique for time series, which could result in higher accuracy of forecasting using CNN, an MAPE of around 12%.

DeepFocus: Deep Encoding Brainwaves and Emotions with Multi-Scenario Behavior Analytics for Human Attention Enhancement

Evaluation of the degree of attention has very important practical significance and application prospects in our lives. It plays considerably important functions in the fields of education, medical treatment, and automatic driving, and assists people in assessing psychological states automatically and gives early warning. However, there are some problems in the existing research, such as simplex data modality and simple modeling, which make the algorithm difficult to deploy in real circumstances. To solve these problems, a new system of attention degree evaluation, called DeepFocus, is put forward in this article. We propose an attention evaluation method based on multi-modal data and multi-scenario modeling for the first time. In addition, the relationship between emotional data and attention is analyzed in depth, and labels are corrected with emotional data. Based on the DeepFocus system, a completely new attention enhancement system is constructed, and the algorithm is deployed in a practical application for students to perceive and enhance their attention. It can be foreseen that our algorithm could explore people's inward world and assess users' attention degree accurately and comprehensively to help people work, study, and live better with higher efficiency in the near future.

DE-Net: Deep Encoding Network for Building Extraction from High-Resolution Remote Sensing Imagery

Deep convolutional neural networks have promoted significant progress in building extraction from high-resolution remote sensing imagery. Although most of such work focuses on modifying existing image segmentation networks in computer vision, we propose a new network in this paper, Deep Encoding Network (DE-Net), that is designed for the very problem based on many lately introduced techniques in image segmentation. Four modules are used to construct DE-Net: the inception-style downsampling modules combining a striding convolution layer and a max-pooling layer, the encoding modules comprising six linear residual blocks with a scaled exponential linear unit (SELU) activation function, the compressing modules reducing the feature channels, and a densely upsampling module that enables the network to encode spatial information inside feature maps. Thus, DE-Net achieves state-of-the-art performance on the WHU Building Dataset in recall, F1-Score, and intersection over union (IoU) metrics without pre-training. It also outperformed several segmentation networks in our self-built Suzhou Satellite Building Dataset. The experimental results validate the effectiveness of DE-Net on building extraction from aerial imagery and satellite imagery. It also suggests that given enough training data, designing and training a network from scratch may excel fine-tuning models pre-trained on datasets unrelated to building extraction.

Manipulating the depth of processing reveals the relevance of second eye fixations for recollection but not familiarity.

It is well known that memory affects eye movements. However, the role of individual eye fixations for recognition memory processes has hardly been investigated. Recent findings show that second fixations are especially relevant for recollection, a process associated with the retrieval of context information, but less for recognition based solely on item familiarity. The aim of the present study was to overcome limitations of a previous study (Schwedes and Wentura in Memory, 2019. https://doi.org/10.1080/09658211.2019.1567789 ) and to provide further evidence that second fixations are especially relevant for recollection-based recognition. Whereas recollection- and familiarity-based recognition was an unconstrained quasi-experimental variable in a previous study, here we manipulated the depth of stimulus processing in the encoding phase to experimentally manipulate the probability of subsequent item recollection. In the old/new recognition memory test, presentation of test probes was terminated after one or two stimulus fixations. "Old" responses in the recognition test were followed by a remember/know/guess procedure to assess recollection-based versus familiarity-based recognition. We found the expected depth of processing effect, with better recognition and more recollection-based responses after deep encoding. This effect, however, was significantly larger if two fixations instead of just one were allowed. There were no corresponding effects for familiarity-based recognition. Thus, a second fixation seems to play an important role only for recollection-based recognition.

Combining High Speed ELM Learning with a Deep Convolutional Neural Network Feature Encoding for Predicting Protein-RNA Interactions.

Emerging evidence has shown that RNA plays a crucial role in many cellular processes, and their biological functions are primarily achieved by binding with a variety of proteins. High-throughput biological experiments provide a lot of valuable information for the initial identification of RNA-protein interactions (RPIs), but with the increasing complexity of RPIs networks, this method gradually falls into expensive and time-consuming situations. Therefore, there is an urgent need for high speed and reliable methods to predict RNA-protein interactions. In this study, we propose a computational method for predicting the RNA-protein interactions using sequence information. The deep learning convolution neural network (CNN) algorithm is utilized to mine the hidden high-level discriminative features from the RNA and protein sequences and feed it into the extreme learning machine (ELM) classifier. The experimental results with 5-fold cross-validation indicate that the proposed method achieves superior performance on benchmark datasets (RPI1807, RPI2241, and RPI369) with the accuracy of 98.83, 90.83, and 85.63 percent, respectively. We further evaluate the performance of the proposed model by comparing it with the state-of-the-art SVM classifier and other existing methods on the same benchmark data set. In addition, we predicted the independent NPInter v2.0 data set using the model trained on RPI369. The experimental results show that our model can serve as a useful tool for predicting RNA-protein interactions.

Deep salient-Gaussian Fisher vector encoding of the spatio-temporal trajectory structures for person re-identification

In this paper, we propose a deep spatio-temporal appearance (DSTA) descriptor for person re-identification (re-ID). The proposed descriptor is based on the deep Fisher vector (FV) encoding of the trajectory spatio-temporal structures. These have the advantage of robustly handling the misalignment in the pedestrian tracklets. The deep encoding exploits the richness of the spatio-temporal structural information around the trajectories. This is achieved by hierarchically encoding the trajectory structures leveraging a larger tracklet neighborhood scale when moving from one layer to the next one. In order to eliminate the noisy background located around the pedestrian and model the uniqueness of its identity, the deep FV encoder is further enriched towards the deep Salient-Gaussian weighted FV (deepSGFV) encoder by integrating the pedestrian Gaussian and saliency templates in the encoding process, respectively. The proposed descriptor produces competitive accuracy with respect to state-of-the art methods and especially the deep CNN ones without necessitating either pre-training or data augmentation on four challenging pedestrian video datasets: PRID2011, i-LIDS-VID, Mars and LPW. The further combination of DSTA with deep CNN boosts the current state-of-the-art methods and demonstrates their complementarity.

Emotional arousal and memory after deep encoding

Emotion often enhances long-term memory. One mechanism for this enhancement is heightened arousal during encoding. However, reducing arousal, via emotion regulation (ER) instructions, has not been associated with reduced memory. In fact, the opposite pattern has been observed: stronger memory for emotional stimuli encoded with an ER instruction to reduce arousal. This pattern may be due to deeper encoding required by ER instructions. In the current research, we examine the effects of emotional arousal and deep-encoding on memory across three studies. In Study 1, adult participants completed a writing task (deep-encoding) for encoding negative, neutral, and positive picture stimuli, whereby half the emotion stimuli had the ER instruction to reduce the emotion. Memory was strong across conditions, and no memory enhancement was observed for any condition. In Study 2, adult participants completed the same writing task as Study 1, as well as a shallow-encoding task for one-third of negative, neutral, and positive trials. Memory was strongest for deep vs. shallow encoding trials, with no effects of emotion or ER instruction. In Study 3, adult participants completed a shallow-encoding task for negative, neutral, and positive stimuli, with findings indicating enhanced memory for negative emotional stimuli. Findings suggest that deep encoding must be acknowledged as a source of memory enhancement when examining manipulations of emotion-related arousal.

SAR Image Classification via Deep Recurrent Encoding Neural Networks

Synthetic aperture radar (SAR) image classification is a fundamental process for SAR image understanding and interpretation. With the advancement of imaging techniques, it permits to produce higher resolution SAR data and extend data amount. Therefore, intelligent algorithms for high-resolution SAR image classification are demanded. Inspired by deep learning technology, an end-to-end classification model from the original SAR image to final classification map is developed to automatically extract features and conduct classification, which is named deep recurrent encoding neural networks (DRENNs). In our proposed framework, a spatial feature learning network based on long–short-term memory (LSTM) is developed to extract contextual dependencies of SAR images, where 2-D image patches are transformed into 1-D sequences and imported into LSTM to learn the latent spatial correlations. After LSTM, nonnegative and Fisher constrained autoencoders (NFCAEs) are proposed to improve the discrimination of features and conduct final classification, where nonnegative constraint and Fisher constraint are developed in each autoencoder to restrict the training of the network. The whole DRENN not only combines the spatial feature learning power of LSTM but also utilizes the discriminative representation ability of our NFCAE to improve the classification performance. The experimental results tested on three SAR images demonstrate that the proposed DRENN is able to learn effective feature representations from SAR images and produce competitive classification accuracies to other related approaches.

An ERP study of multidimensional source retrieval in depression

We collected event-related potentials (ERPs) from 24 unmedicated adults with Major Depressive Disorder (MDD) and 24 controls during source memory retrieval. Words were encoded on the left or right during animacy and mobility judgments. Mobility judgments were slower than animacy judgments, suggesting deeper encoding. Participants then recalled the encoding judgment (Question cue) and position (Side cue) for each word. Depressed adults, but not controls, showed better accuracy for words from the mobility task presented under the Question vs. Side Cue. Furthermore, depressed adults showed larger left parietal ERPs to words from the mobility task presented under the Question vs. the Side Cue from 400 to 800 ms and 800–1400 ms. This ERP effect was negatively correlated with sleep quality. Thus, deep encoding followed by retrieval of the encoding judgment supported memory in MDD and augmented left parietal ERPs that have been linked to recollection and that appear sensitive to sleep disturbance.

The impact of familiarization strategies on the missing-letter effect.

When reading a text and searching for a target letter, readers make more omissions of the target letter if it is embedded in frequent function words than if it is in rare content words. While word frequency effects are consistently found, few studies have examined the impacts of passage familiarity on the missing-letter effect and studies that have present conflicting evidence. The present study examines the effects of passage familiarity, as well as the impacts of passage familiarization strategy promoting surface or deep encoding, on the missing-letter effect. Participants were familiarized with a passage by retyping a text, replacing all common nouns with synonyms, or generating a text on the same topic as that of the original text, and then completed a letter search task on the familiar passage as well as an unfamiliar passage. In Experiment 1, when both familiar and unfamiliar passages use the same words, results revealed fewer omissions for the retyping and synonyms conditions. However, in Experiment 2, when different words are used in both types of texts, no effect of familiarization strategy was observed. Furthermore, the missing-letter effect is maintained in all conditions, adding support to the robustness of the effect regardless of familiarity with the text.

Is the Levels of Processing effect language-limited?

Abstract The concept of Levels of Processing (LOP), proposing that deep coding enhances retention, has played a central role in the study of episodic memory. Evidence has however been based almost entirely on retention of individual words. Across five experiments, we compare LOP effects between visual and verbal stimuli, using judgments of pleasantness as a method of inducing deep encoding and a range of shallow encoding judgments selected so as to be applicable to both verbal and visual stimuli. LOP effects were consistent but modest across the visual stimuli (mean effect size 0.5). In contrast, LOP effects for verbal stimuli varied widely, from modest for people’s names and unfamiliar animals (mean effect size 0.6) to large for familiar animals and household items (mean effect size 1.4), typical of the dramatic LOP effects that characterize the existing verbal literature. We interpret our data through the Gibsonian concept of “affordance”, proposing that visual and verbal stimuli vary in the number and richness of features they afford, and that access to such features will in turn depend on encoding strategy. Our hypothesis links readily with Nairne’s feature model of long-term memory.

Short theta burst stimulation to left frontal cortex prior to encoding enhances subsequent recognition memory.

Deep semantic encoding of verbal stimuli can aid in later successful retrieval of those stimuli from long-term episodic memory. Evidence from numerous neuropsychological and neuroimaging experiments demonstrate regions in left prefrontal cortex, including left dorsolateral prefrontal cortex (DLPFC), are important for processes related to encoding. Here, we investigated the relationship between left DLPFC activity during encoding and successful subsequent memory with transcranial magnetic stimulation (TMS). In a pair of experiments using a 2-session within-subjects design, we stimulated either left DLPFC or a control region (Vertex) with a single 2-s train of short theta burst stimulation (sTBS) during a semantic encoding task and then gave participants a recognition memory test. We found that subsequent memory was enhanced on the day left DLPFC was stimulated, relative to the day Vertex was stimulated, and that DLPFC stimulation also increased participants' confidence in their decisions during the recognition task. We also explored the time course of how long the effects of sTBS persisted. Our data suggest 2s of sTBS to left DLPFC is capable of enhancing subsequent memory for items encoded up to 15s following stimulation. Collectively, these data demonstrate sTBS is capable of enhancing long-term memory and provide evidence that TBS protocols are a potentially powerful tool for modulating cognitive function.

What-where-when memory and encoding strategies in healthy aging.

Older adults exhibit disproportionate impairments in memory for item-associations. These impairments may stem from an inability to self-initiate deep encoding strategies. The present study investigates this using the “treasure-hunt task”; a what–where–when style episodic memory test that requires individuals to “hide” items around complex scenes. This task separately assesses memory for item, location, and temporal order, as well as bound what–where–when information. The results suggest that older adults are able to ameliorate integration memory deficits by using self-initiated encoding strategies when these are externally located and therefore place reduced demands on working memory and attentional resources.

You can go your own way: effectiveness of participant-driven versus experimenter-driven processing strategies in memory training and transfer

ABSTRACTCognitive training programs that instruct specific strategies frequently show limited transfer. Open-ended approaches can achieve greater transfer, but may fail to benefit many older adults due to age deficits in self-initiated processing. We examined whether a compromise that encourages effort at encoding without an experimenter-prescribed strategy might yield better results. Older adults completed memory training under conditions that either (1) mandated a specific strategy to increase deep, associative encoding, (2) attempted to suppress such encoding by mandating rote rehearsal, or (3) encouraged time and effort toward encoding but allowed for strategy choice. The experimenter-enforced associative encoding strategy succeeded in creating integrated representations of studied items, but training-task progress was related to pre-existing ability. Independent of condition assignment, self-reported deep encoding was associated with positive training and transfer effects, suggesting that the most beneficial outcomes occur when environmental support guiding effort is provided but participants generate their own strategies.

Relational Memory as a Possible Neurocognitive Marker of Schizophrenia.

Long-term or episodic memory can be understood along several dimensions, including verbal vs abstract, spatial vs nonspatial, and item vs relational memory. The latter has represented a field of growing interestover thepast 15 years, particularly in cognitive neuroscience and neuroimaging. Inbasic terms, itemmemory is a metric related to a single piece of information, whereas relational memory involves forming a relationship between 2 or more piecesof information.Relationalmemorycanoccurwithinmanycontexts such as linkingmultiple objects together (eg, “I saw xwith y.”) orwithin a spatial or temporal context (eg, “I saw x at thepark in the afternoon.”). However, relational memory cannot be considered successful in the absence of item memory (eg, “I saw someone at the park.”). Relationalmemory requiresmore elaborate processing to create associations that can later be recalled together. By consequence, memory performance for individual items can be improved with deep semantic encoding (eg, combining items into groups) asopposed toshallowencoding (eg, remembering items individually).Not surprisingly, forming relationshipsbetween items is easiest when items share semantic features such as category (eg, hammer and screwdriver are tools). The foundation of understanding the neural basis of item and relational memory was based on lesion studies in which relational memorywas selectively impaired,despite relativelypreserved item memory followinghippocampal or prefrontal damage.Nowa technique, suchas functionalmagnetic resonance imaging (fMRI),makes it possible to continue advancing our knowledge and to further our understandingof a specific cognitiveprocess (eg, itemvs relational memory) at the neuronal level. It also allows us to explore specific cognitive deficits in psychiatric research with relative ease.

Unsupervised Joint Feature Learning and Encoding for RGB-D Scene Labeling.

Most existing approaches for RGB-D indoor scene labeling employ hand-crafted features for each modality independently and combine them in a heuristic manner. There has been some attempt on directly learning features from raw RGB-D data, but the performance is not satisfactory. In this paper, we propose an unsupervised joint feature learning and encoding (JFLE) framework for RGB-D scene labeling. The main novelty of our learning framework lies in the joint optimization of feature learning and feature encoding in a coherent way, which significantly boosts the performance. By stacking basic learning structure, higher level features are derived and combined with lower level features for better representing RGB-D data. Moreover, to explore the nonlinear intrinsic characteristic of data, we further propose a more general joint deep feature learning and encoding (JDFLE) framework that introduces the nonlinear mapping into JFLE. The experimental results on the benchmark NYU depth dataset show that our approaches achieve competitive performance, compared with the state-of-the-art methods, while our methods do not need complex feature handcrafting and feature combination and can be easily applied to other data sets.

Does long-term object priming depend on the explicit detection of object identity at encoding?

It is currently unclear whether objects have to be explicitly identified at encoding for reliable behavioral long-term object priming to occur. We conducted two experiments that investigated long-term object and non-object priming using a selective-attention encoding manipulation that reduces explicit object identification. In Experiment 1, participants either counted dots flashed within an object picture (shallow encoding) or engaged in an animacy task (deep encoding) at study, whereas, at test, they performed an object-decision task. Priming, as measured by reaction times (RTs), was observed for both types of encoding, and was of equivalent magnitude. In Experiment 2, non-object priming (faster RTs for studied relative to unstudied non-objects) was also obtained under the same selective-attention encoding manipulation as in Experiment 1, and the magnitude of the priming effect was equivalent between experiments. In contrast, we observed a linear decrement in recognition memory accuracy across conditions (deep encoding of Experiment 1 > shallow encoding Experiment 1 > shallow encoding of Experiment 2), suggesting that priming was not contaminated by explicit memory strategies. We argue that our results are more consistent with the identification/production framework than the perceptual/conceptual distinction, and we conclude that priming of pictures largely ignored at encoding can be subserved by the automatic retrieval of two types of instances: one at the motor level and another at an object-decision level.

Emotional memory enhancement in respect of positive visual stimuli in Alzheimer's disease emerges after rich and deep encoding

Healthy participants remember emotional stimuli better than neutral stimuli. In normal, older adults this emotional enhancement of memory (EEM) was mostly seen in respect of positive stimuli (positivity bias) after rich and deep encoding. The results relating to this effect in Alzheimer's disease (AD) patients are fairly inconsistent. The goal of the present study was to ascertain whether EEM in AD patients depends on the depth and the richness of encoding. Twenty-one patients with mild-to-moderate AD and 23 age-and-education-matched controls completed two study phases, each followed by a retrieval phase. The first study phase consisted of a natural/man-made categorization task followed by a recognition task, whereas the second, which involved the stimuli used in the recognition task, allowed for richer and deeper encoding as a result of repetition and naming of the stimuli and presentation of the same semantic cues at encoding and retrieval. The second study phase was followed by free and cued recall tasks and a recognition task. After the first study phase we observed EEM in respect of negative and positive stimuli in controls, but not in AD patients. After the second study phase, the positivity bias was observed in the free recall task in controls but not in AD patients. In the cued recall and recognition tasks, however, both groups showed the positivity bias. Based on our results, AD patients present a positivity memory bias when encoding is sufficiently rich and deep, and when support is provided at the time of retrieval (cued recall or recognition tasks).