Memory Updating Research Articles

Norepinephrine Regulates Spatial Memory Consolidation Using Barnes Maze in Male, but Not Female Rats (P1-12.009)

<h3>Objective:</h3> Here, we study the role of NE on spatial learning and memory in rats using a Barnes maze assay. <h3>Background:</h3> The role of norepinephrine (NE) in learning and memory has been extensively studied, yet its contribution to different phases of memory, such as consolidation or updating, remains elusive. <h3>Design/Methods:</h3> We used a specific noradrenergic neurotoxin (N-(2-chloroethyl)-N-ethyl-2-bromobenzylamine (DSP-4)) to deplete central NE stores. In experiment one, Wistar rats (n=16/sex) were trained to reach a hidden goal box using visual and auditory cues for five days with three trials per day. Rats were administered 50 mg/kg/i.p of DSP-4 or saline ten days prior to training on the Barnes maze assay. DSP-4 mediated reduction of NE, but not serotonin or dopamine, was confirmed using HPLC in several brain regions. In experiment two, using female and male rats (n=16/sex), we replicated results of experiment one by utilizing two trials per day to increase the difficulty of learning the task and extended the findings by utilizing a single probe trial (memory consolidation) and a two-day reversal (memory updating) learning task. <h3>Results:</h3> In experiment one, female and male rats in DSP-4 and saline groups learned the location of the hidden zone, and this was measured by a reduction in latency to reach the goal box. Levels of NE were significantly lower in the dorsal hippocampus, cingulate cortex, and the striatum. Experiment two replicated these results and the subsequent behavioral analysis suggests that only male DSP-4 treated rats increased time spent in the covered hidden zone area during the probe trial. During the reversal learning task, DSP-4 treated males continued to visit the previously hidden platform despite learning the new goal location. <h3>Conclusions:</h3> These data suggest that NE plays a role in regulation of memory consolidation rather than acquisition or memory updating, and this regulation may be sex-specific. <b>Disclosure:</b> Ms. Simpson has nothing to disclose. Dr. Gheidi has nothing to disclose. Dr. Davidson has nothing to disclose. Mr. Yahya has nothing to disclose. Mrs. Trombley has nothing to disclose. Miss Sadik has received personal compensation for serving as an employee of Wayne State University . The institution of Dr. Perrine has received research support from NIH.

Influence of Strenuous Physical Activity and Cardiorespiratory Fitness on Age-Related Differences in Brain Activations During Varieties of Cognitive Control

While there is extensive literature on the beneficial effects of physical activity on age differences in cognitive control, limited research exists on comparing the contributions of strenuous physical activity (sPA) and cardiorespiratory fitness (CRF) to fluctuations in the blood oxygen level-dependent (BOLD) signals during varieties of cognitive control. The current study addresses this gap in knowledge by investigating BOLD signal differences between high-fit and low-fit older adults, determined by their sPA or CRF, during a novel fMRI task with a hybrid block and event-related design that included transient activations (during switching, updating and their combination trials) and sustained activations (during proactive and reactive control blocks). fBOLD signals from older (n = 25) were compared to more functionally efficient younger (n = 15) adults. High-sPA old showed higher task accuracy than Low-sPA old and similar accuracy as young. Whole-brain fMRI analyses identified higher BOLD activations (esp. dlPFC/MFG) in high-fit old during updating and combination trials that were similar to young, suggesting maintenance of BOLD signals in higher fit older adults during working memory updating. Additionally, both High-sPA and High-CRF related compensatory overactivation were observed in left parietal and occipital areas during sustained activations, which were positively correlated with older adults' accuracy. These results suggest that physical fitness is a modifier of age-related changes in BOLD signal modulation elicited in response to increasing cognitive control demands, with higher fitness in old contributing to both compensatory overactivations and maintenance of task-related brain activations during cognitive control, whereas lower fitness contributed to maladaptive overactivations during lower cognitive demands.

Recurrent unit augmented memory network for video summarisation

AbstractVideo summarisation can relieve the pressure on video storage, transmission, archiving, and retrieval caused by the explosive growth of online videos in recent years. Most existing supervised video summarisation methods use convolutional neural network (CNN) or recurrent neural network (RNN) to model the temporal dependencies between video frames or video shots. CNN mainly focuses on local information, and RNN loses long‐term information when the input sequence is long, both of which have limited ability to obtain long‐range memory in the video. Therefore, a recurrent unit augmented memory network (RUAMN) for video summarisation is proposed, which effectively utilises the long‐term memory extraction ability of the end‐to‐end memory network (MemN2N) and solves the problem that MemN2N is insensitive to temporal sequence information. At the same time, the proposed RUAMN enhances the process of memory update between multiple computational steps (hops), and finally generates a meaningful video summarisation result. Specifically, the proposed RUAMN is mainly composed of the input module, the global‐and‐local sampling, the memory module and the output module. The input module uses a bidirectional GRU to obtain the forward and backward information of each video frame. Then the global‐and‐local sampling performs global sampling and local sampling on the output sequence of the input module respectively to obtain several shorter sequences, so that the memory modules can capture the fine‐grained relationship features between video frames more effectively. The memory module extracts the long‐term memory information in the feature sequence, and finally the frame‐level importance scores are predicted by the output module. Extensive experiments on benchmark datasets, that is, TVSum and SumMe, demonstrate the superiority of our method over several state‐of‐the‐art supervised video summarisation methods.

152 Caudate and Dorsolateral Prefrontal Cortex Beta Bursts During Working Memory Encoding Correlate with Memory Impairment in Parkinson’s Disease

INTRODUCTION: Cognitive dysfunction is a common nonmotor symptom of Parkinson’s disease (PD), with memory impairment being a primary cause of disability. Decreased dopamine in PD elevates beta power in cortico-striato-thalamo-cortico (CSTC) motor circuits and is the basis of closed-loop deep brain stimulation (DBS) strategies. Analogous dysfunction of parallel CSTC cognitive circuitry including caudate and dorsolateral prefrontal cortex (DLPFC) may mediate PD cognitive impairment and be a neuromodulation target for cognitive symptoms. We have previously shown caudate and DLPFC beta power contributes to working memory updating, and suppression of prefrontal cortex beta bursts during working memory encoding has been linked with improved performance. However, how PD affects beta bursting in relation to memory function remains unknown. METHODS: Fourteen PD patients undergoing awake DBS surgery with electrode trajectories traversing the caudate, DLPFC, or both participated in this study. Subjects completed a 2-back working memory task where they identified whether a presented word matched the word presented two trials prior. Bursts were defined as epochs where power exceeded a Z-score of two for at least three oscillatory cycles. We correlated change in caudate and DLPFC beta burst rates during working memory encoding with subjects’ composite memory scores, which were calculated from preoperative neuropsychological assessments. RESULTS: Caudate and DLPFC beta burst rates decreased significantly during working memory encoding for correct trials. In both structures, change in beta burst rate was significantly correlated with subjects’ composite memory scores, with greater decreases in beta bursting corresponding with better memory function. CONCLUSIONS: Beta bursts are suppressed in caudate and DLPFC during working memory encoding. Reduced beta burst suppression during encoding correlated with greater memory impairment in PD patients, indicating that altered caudate and DLPFC beta bursting dynamics may contribute to PD memory impairment.

The influence of bilingual experience on executive function under emotional interference: Evidence from the N1 component.

The influence of bilingual education and experience on an individual's information-processing ability has recently been a hot issue in international studies. Previous studies have found that bilingual experience affects executive function, but the results remain controversial. Executive function refers to the conscious control of purposeful behavior. It is responsible for processing high-level action controls, including such sub-functions as inhibitory control, cognitive switching, and working memory updating. Emotion, as an essential factor in daily life, also has a complex interaction with executive function. This paper explores whether the bilingual cognitive advantage effect can continue in the more complex conditions of emotional interference. To investigate the specific electrophysiological characteristics of the participants at different stages of cognitive processing, we used a combination of the behavioral and ERP experiments in which the positive, neutral, and negative emotional stimuli were selected as emotional interference conditions and the emotional Simon paradigm, the cognitive switching of emotion paradigm, and the emotional N-back paradigm was adopted. The results show that the main effect of the N1 component amplitude is significant. Specifically, the amplitude of the N1 component in the proficient bilinguals is significantly smaller than that of the non-proficient bilinguals, while the main effects of other component groups are not significant, indicating that under the condition of emotional interference, the influence of bilingual experience on executive function only exists in the early attention stage and that the bilingual experience can improve the individual's attentional control and speed up attention processing in the early attention stage.

Association between adolescents' socioeconomic status and working memory updating functioning: Role of parental educational involvement.

Family socioeconomic status (SES) is positively associated with executive functioning. This study tested whether parental educational involvement mediates this association. Two hundred and sixty, 12-15-year-old adolescents completed working memory updating (WMU) and general intelligence tasks, and questionnaires on SES and parental educational involvement. SES and WMU ability were positively associated; there was no difference between the fathers and mothers for three types of educational involvement. The mothers' behavioural involvement positively mediated the SES-WM updating association, whereas a negative mediation was observed for the mothers' intellectual involvement. The fathers' educational involvement did not play a significant mediating role. These results might inform interventions targeting educational involvement for enhancing the cognitive development of children from low SES families.

The gerrymander sequence, or A348456

The gerrymander sequence, gL, given as A348456 in the OEIS, counts the number of ways to bisect a 2L×2L chessboard into two polyominoes of equal area. Recently Kauers, Koutschan and Spahn announced a significant increase in the length of this sequence from 3 to 7 terms. We give a further extension to 11 terms, but more significantly prove that the coefficients grow as λ4L2, where λ≈1.7445498, and is equal to the corresponding quantity for self-avoiding walks crossing a square (WCAS), or self-avoiding polygons crossing a square (PCAS). These are, respectively, OEIS sequences A007764 and A333323. Thus we have established a close connection between these previously separate problems.We have also related the sub-dominant behaviour to that of WCAS and PCAS, allowing us to conjecture that the coefficients of the gerrymander sequence grow as gL∼λ4L2+2dL+e⋅(2L)h, where d=−4.04354±0.0001, e≈8 and h=0.75±0.01, with h almost certainly 3/4 exactly.We have also generated 26 terms of the related OEIS sequence A068416, which counts the number of ways to partition a L×L square into two connected components (not necessarily of equal area). We have thus been able to predict the asymptotic behaviour of this sequence with a satisfying degree of precision. Indeed, it behaves exactly as L times the corresponding coefficient of the generalised gerrymander sequence (defined below).The improved algorithm we give for counting these sequences is a variation of that which we recently developed for extending a number of sequences for SAWs and SAPs crossing a domain of the square or hexagonal lattices. It makes use of a minimal perfect hash function and in-place memory updating of the arrays for the counts of the number of paths.

Muscarinic receptor activation overrides boundary conditions on memory updating in a calcium/calmodulin-dependent manner.

Long-term memory storage is a dynamic process requiring flexibility to ensure adaptive behavioural responding in changing environments. Indeed, it is well established that memory reactivation can "destabilize" consolidated traces, leading to various forms of updating. However, the neurobiological mechanisms rendering long-term memories labile and modifiable remain poorly described. Moreover, boundary conditions, such as the age or strength of the memory, can reduce the likelihood of this destabilization; yet, intuitively, these most behaviourally influential of memories should also be modifiable under appropriate conditions. Here, we provide evidence that salient novelty at the time of memory reactivation promotes integrative updating of resistant object memories in rats. Furthermore, blockade of muscarinic acetylcholine receptors (mAChRs; with pirenzepine) or disruption of calcium/calmodulin (Ca2+/CaM) with KN-93, a Ca2+/CaM-binding molecule that inhibits calcium/calmodulin-dependent protein kinase II (CaMKII) activation, in perirhinal cortex (PRh) prevented novelty-induced destabilization and updating of resistant object memories. Finally, PRh M1 mAChR activation (with CDD-0102A) was sufficient to destabilize resistant object memories for updating, and this effect was blocked by KN-93, possibly via inhibition of CaMKII activity. Thus, mAChRs and activation of CaMKII appear to interact as part of a mechanism to override boundary conditions on resistant object memories to ensure integrative modification with novel information. These findings therefore have important implications for understanding the dynamic nature of long-term memory storage and potential treatments for conditions characterized by maladaptive and inflexible memories.

Specific but not general declines in attention and executive function with aging: Converging cross-sectional and longitudinal evidence across the adult lifespan.

Attention and executive function (EF) are vulnerable to aging. However, whether all these functions generally decline with aging is not known. Furthermore, most evidence is based on cross-sectional data and fewer follow-up data are available in the literature. Longitudinal follow-up studies are necessary to characterize individualized and precise changes in cognitive function. Additionally, relatively few aging studies have included middle-aged adults to examine age-related differences in attention and EF. Therefore, this study aims to examine whether general or specific attention and EF decline with aging from adulthood to old age by combining cross-sectional and longitudinal follow-up approaches. This study recruited 253 participants aged 20 to 78 years. passing a prescreening procedure (see main text for detail) for the baseline session, and 123 of them were invited to return 1 ~ 2 years after their first visit to participate in the follow-up session. The participants completed a series of attention and EF tasks at both the baseline and follow-up sessions, which measured alerting, orienting, conflict control, stopping, memory updating, and switching abilities. We applied linear and nonlinear regression models to evaluate the cross-sectional age effect on attention and EF and employed a modified Brinley plot to inspect follow-up performance against baseline in attention and EF. The results of cross-sectional data showed that older adults exhibited decreased efficiency in alerting, stopping, and memory updating but paradoxically increased efficiency in conflict control and switching abilities and no changes in orienting efficiency with age. However, the results of longitudinal data showed that only alerting and memory updating continued to show decreased efficiency. Furthermore, conflict control and switching showed increased efficiency with aging, whereas the orienting network, and stopping no longer showed decreased efficiency. Thus, converging the cross-sectional and longitudinal data showed that the alerting and memory updating function exhibited the most robust deficit with age (cross-sectional) and aging (longitudinal). Alerting and memory updating abilities are crucial survival skills for human beings. Therefore, developing methods to prevent and improve an individual's alertness and working memory ability is an important practical issue in aging research.

Occupation shapes cognition? The case of restaurant ticket collectors' working memory updating ability

AbstractPrevious research revealed associations between an individual's occupation and cognitive abilities. However, the underlying causal relation is not always clear and only few studies focused on a critical component of executive functioning, namely working memory updating (WMU). Study 1 examined whether restaurant ticket collectors (N = 53) have a better WMU ability compared to a group of security guards (N = 49) that was matched on relevant variables. Study 2 examined transfer effects of a computerized working memory training program in students. The program simulated elements of a restaurant ticket collector's daily work requirements. In Study 1, the ticket collectors performed better than the guards on WMU tasks. In Study 2, using an active control group (N = 33) as comparison, the trained students (N = 33) displayed beneficial training effects on transfer WMU tasks but not on general intelligence tasks. The results support the general notion of repeated experience with occupation‐specific demands affecting specific objectively‐assessed cognitive abilities.

Learning from errors: Distinct neural networks for monitoring errors and maintaining corrects through repeated practice and feedback

How memory representations are eventually established and maintained in the brain is one of central issues in memory research. Although the hippocampus and various brain regions have been shown to be involved in learning and memory, how they coordinate to support successful memory through errors is unclear. In this study, a retrieval practice (RP) – feedback (FB) paradigm was adopted to address this issue. Fifty-six participants (27 in the behavioral group, and 29 in the fMRI group) learned 120 Swahili-Chinese words associations and underwent two RP-answer FB cycles (i.e., RP1, FB1, RP2, FB2). The responses of the fMRI group were recorded in the fMRI scanner. The trials were divided based on participant's performance (correct or incorrect, C or I) during the two RPs and the final test (i.e., trial type, CCC, ICC, IIC III). The results showed that the regions in the salience and executive control networks (S-ECN) during RP, but not during FB, was strongly predictive of final successful memory. Their activation was just before the errors were corrected (i.e., RP1 in ICC trials and RP2 in IIC trials). The anterior insula (AI) is a core region in monitoring repeated errors, and it had differential connectivity with the default mode network (DMN) regions and the hippocampus during the RP and FB phases to inhibit incorrect answers and update memory. In contrast, maintaining corrected memory representation requires repeated RP and FB, which was associated with the DMN activation. Our study clarified how different brain regions support error monitoring and memory maintenance through repeated RP and FB, and emphasized the role of the insula in learning from errors.

From Stochastic to Rough Volatility: A New Deep Learning Perspective on Hedging

The Black–Scholes model assumes that volatility is constant, and the Heston model assumes that volatility is stochastic, while the rough Bergomi (rBergomi) model, which allows rough volatility, can perform better with high-frequency data. However, classical calibration and hedging techniques are difficult to apply under the rBergomi model due to the high cost caused by its non-Markovianity. This paper proposes a gated recurrent unit neural network (GRU-NN) architecture for hedging with different-regularity volatility. One advantage is that the gating network signals embedded in our architecture can control how the present input and previous memory update the current activation. These gates are updated adaptively in the learning process and thus outperform conventional deep learning techniques in a non-Markovian environment. Our numerical results also prove that the rBergomi model outperforms the other two models in hedging.

Consistent Intra-Video Contrastive Learning With Asynchronous Long-Term Memory Bank

Unsupervised representation learning for videos has recently achieved remarkable performance owing to the effectiveness of contrastive learning. Most works on video contrastive learning (VCL) pull all snippets from the same video into the same category, even if some of them are from different actions, leading to temporal collapse, i.e., the snippet representations of a video are invariable with the evolution of time. In this paper, we introduce a novel intra-video contrastive learning (intra-VCL) that further distinguishes intra-video actions to alleviate this issue, which includes an asynchronous long-term memory bank (that caches the representations of all snippets of each video) and mines an extra positive/negative snippet within a video based on the asynchronous long-term memory bank. In addition, since an asynchronous long-term memory bank is required for performing intra-VCL and asynchronous update of the long-term memory leads to inconsistencies when performing contrastive learning, we further propose a consistent contrastive module (CCM) to perform consistent intra-VCL. Specifically, in the CCM, we propose an intra-video self-attention refinement function to reduce the inconsistencies within the asynchronously updated representations (of all snippets of each video) in the long-term memory and an adaptive loss re-weighting to reduce unreliable self-supervision produced by inconsistent contrastive pairs. We call our method as consistent intra-VCL. Extensive experiments demonstrate the effectiveness of the proposed consistent intra-VCL, which achieves state-of-the-art performance on the standard benchmarks of self-supervised action recognition, with top-1 accuracies of 64.2% and 91.0% on HMDB-51 and UCF-101, respectively.

Continual Recognition with Adaptive Memory Update

Class incremental continual learning aims to improve the ability of modern classification models to continually recognize new classes without forgetting the previous ones. Prior art in the field has largely considered using a replay buffer. In this article, we start from an observation that the existing replay-based method would fail when the stored exemplars are not hard enough to get a good decision boundary between a previously learned class and a new class. To prevent this situation, we propose a method from the perspective of remedy after forgetting for the first time. In the proposed method, a set of exemplars is preserved as a working memory, which helps to recognize new classes. When the working memory is insufficient to distinguish between new classes, more discriminating samples would be swapped from a long-term memory, which is built up during the early training process, in an adaptive way. Our continual recognition model with adaptive memory update is capable of overcoming the problem of catastrophic forgetting with various new classes coming in sequence, especially for similar but different classes. Extensive experiments on different real-world datasets demonstrate that the proposed model is superior to existing state-of-the-art algorithms. Moreover, our model can be used as a general plugin for any replay-based continual learning algorithm to further improve their performance.

Transforming experiences: Neurobiology of memory updating/editing.

Long-term memory is achieved through a consolidation process where structural and molecular changes integrate information into a stable memory. However, environmental conditions constantly change, and organisms must adapt their behavior by updating their memories, providing dynamic flexibility for adaptive responses. Consequently, novel stimulation/experiences can be integrated during memory retrieval; where consolidated memories are updated by a dynamic process after the appearance of a prediction error or by the exposure to new information, generating edited memories. This review will discuss the neurobiological systems involved in memory updating including recognition memory and emotional memories. In this regard, we will review the salient and emotional experiences that promote the gradual shifting from displeasure to pleasure (or vice versa), leading to hedonic or aversive responses, throughout memory updating. Finally, we will discuss evidence regarding memory updating and its potential clinical implication in drug addiction, phobias, and post-traumatic stress disorder.

О стартовых условиях смыслообразования в общении с письменными речевыми жанрами

The article deals with the initial stage of the meaning making process which occurs when a certain request of the reader comes into contact with the inner targeted orientation of the given written saying. The starting conditions of meaning making in written speech genres remind of the effect produced by the meeting of two communicating subjects, with one of them patiently expecting the coming interaction and the other striving to start the immediate interaction to the best of the situationally determined interest of the addressee. Their productive contact is a kind of a cognitive explosion. As a result, there is born a cloud of meanings which do not coincide completely neither with the meanings enclosed in the perceived text nor with the ones felt so far by the recipient. We focus on the factors associated with key verbal aggravators – dominant lexis in the text of the written speech genre and in the updated memory of those who are about to read this text. The reliability of meaning making is determined among other things by a number of intensive contextual coincidences of the key words and notions in the consciousness of the recipient and those which the given written text contains and reveals.

Distinct brain networks for remote episodic memory depending on content and emotional experience

Memories of life episodes are the heart of individual stories. However, modelling episodic memory is a major challenge in both humans and animals when considering all its characteristics. As a consequence, the mechanisms that underlie the storage of old nontraumatic episodic memories remain enigmatic. Here, using a new task in rodents that models human episodic memory including odour/place/context components and applying advances behavioural and computational analyses, we show that rats form and recollect integrated remote episodic memories of two occasionally encountered complex episodes occurring in their daily life. Similar to humans, the information content and accuracy of memories vary across individuals and depend on the emotional relationship with odours experienced during the very first episode. We used cellular brain imaging and functional connectivity analyses, to find out the engrams of remote episodic memories for the first time. Activated brain networks completely reflect the nature and content of episodic memories, with a larger cortico-hippocampal network when the recollection is complete and with an emotional brain network related to odours that is critical in maintaining accurate and vivid memories. The engrams of remote episodic memories remain highly dynamic since synaptic plasticity processes occur during recall related to memory updates and reinforcement.

NMDA Receptor–Arc Signaling Is Required for Memory Updating and Is Disrupted in Alzheimer’s Disease

BackgroundMemory deficits are central to many neuropsychiatric diseases. During acquisition of new information, memories can become vulnerable to interference, yet mechanisms that underlie interference are unknown. MethodsWe describe a novel transduction pathway that links the NMDA receptor (NMDAR) to AKT signaling via the immediate early gene Arc and evaluate its role in memory. The signaling pathway is validated using biochemical tools and transgenic mice, and function is evaluated in assays of synaptic plasticity and behavior. The translational relevance is evaluated in human postmortem brain. ResultsArc is dynamically phosphorylated by CaMKII (calcium/calmodulin-dependent protein kinase II) and binds the NMDAR subunits NR2A/NR2B and a previously unstudied PI3K (phosphoinositide 3-kinase) adapter p55PIK (PIK3R3) in vivo in response to novelty or tetanic stimulation in acute slices. NMDAR-Arc-p55PIK recruits p110α PI3K and mTORC2 (mechanistic target of rapamycin complex 2) to activate AKT. NMDAR-Arc-p55PIK-PI3K-mTORC2-AKT assembly occurs within minutes of exploratory behavior and localizes to sparse synapses throughout hippocampal and cortical regions. Studies using conditional (Nestin-Cre) p55PIK deletion mice indicate that NMDAR-Arc-p55PIK-PI3K-mTORC2-AKT functions to inhibit GSK3 and mediates input-specific metaplasticity that protects potentiated synapses from subsequent depotentiation. p55PIK conditional knockout mice perform normally in multiple behaviors including working memory and long-term memory tasks but exhibit deficits indicative of increased vulnerability to interference in both short-term and long-term paradigms. The NMDAR-AKT transduction complex is reduced in postmortem brain of individuals with early Alzheimer’s disease. ConclusionsA novel function of Arc mediates synapse-specific NMDAR-AKT signaling and metaplasticity that contributes to memory updating and is disrupted in human cognitive disease.

The relationship between working memory updating components and reading comprehension

The objective of this study was to determine the contribution of retrieval and substitution components of working memory updating to reading comprehension. Difficulties in reading comprehension have been related to the inability to update information in working memory. Updating is a complex process comprising various subprocesses, such as retrieving information into the focus of attention and substituting information that is no longer relevant. Various numerical subtasks requiring or not requiring the substitution and retrieval components of working memory updating, as well as reading comprehension and general cognitive measures, were administered to a sample (n = 148) of 4th grade children. Less-skilled comprehenders showed lower accuracy when information retrieval was required. In contrast, substitution was not related to reading comprehension. These findings suggest that reading comprehension difficulties are related to the efficacy of information retrieval during updating in working memory.

Enhancing discriminative appearance model for visual tracking

Most end-to-end discriminative trackers obtain the improvements by a large margin through mining more representative target features for localizing the target, yet it isn't trivial to get comprehensive features as a result of diverse challenges. Moreover, we note that prevalent discriminative trackers merely use a convolution block to process pre-trained residual features to acquire target classification features, which further is input into the appearance model to obtain a target classifier, aiming at obtaining a response map. In view of this problem, we propose a novel feature enhancement module to obtain a richer and comprehensive feature representation, which is end-to-end trainable. Furthermore, discriminative trackers equipped with an online update mechanism demand to refine the classification model with recent samples, which will more or less learn the inaccurate tracking results into the model, thus weakening its discriminative ability. To alleviate this issue, we use a metric learning method to devise a verifier, which is devoted to yielding a similarity score to make piecewise predictions. By comparing the similarity score and piecewise thresholds, this strategy can enhance the robustness of the tracker by adaptively regulating the update of the sample memory and the learning rate. It is worth mentioning that the verifier module has strong portability. We choose the recent SuperDiMP as a baseline and implement comprehensive experimental tests and analyses of our approach on six popular benchmarks, which confirm that the proposed methods perform favorably against state-of-the-art trackers.