LiDAR-as-Camera for End-to-End Driving

The core task of any autonomous driving system is to transform sensory inputs into driving commands. In end-to-end driving, this is achieved via a neural network, with one or multiple cameras as the most commonly used input and low-level driving commands, e.g., steering angle, as output. However, simulation studies have shown that depth-sensing can make the end-to-end driving task easier. On a real car, combining depth and visual information can be challenging due to the difficulty of obtaining good spatial and temporal alignment of the sensors. To alleviate alignment problems, Ouster LiDARs can output surround-view LiDAR images with depth, intensity, and ambient radiation channels. These measurements originate from the same sensor, rendering them perfectly aligned in time and space. The main goal of our study is to investigate how useful such images are as inputs to a self-driving neural network. We demonstrate that such LiDAR images are sufficient for the real-car road-following task. Models using these images as input perform at least as well as camera-based models in the tested conditions. Moreover, LiDAR images are less sensitive to weather conditions and lead to better generalization. In a secondary research direction, we reveal that the temporal smoothness of off-policy prediction sequences correlates with the actual on-policy driving ability equally well as the commonly used mean absolute error.

The timescales of word learning in children with language delays: In-the-moment mapping, retention, and generalization

Learning new words and, subsequently, a lexicon, is a time-extended process requiring encoding of word-referent pairs, retention of that information, and generalization to other exemplars of the category. Some children, however, fail in one or more of these processes resulting in language delays. The present study examines the abilities of children who vary in vocabulary size (including both children with normal language (NL) and late talking (LT) children) across multiple timescales/processes - known and novel word mapping, novel word retention, and novel noun generalization. Results indicate that children with lower language skills suffer from deficits in quick in-the-moment mapping of known words compared to their NL peers, but age and vocabulary size rather than normative vocabulary ranking or NL/LT status better predicts performance on retention and generalization processes. Implications for understanding language development as a holistic process with multiple interacting variables are discussed.

SAR Image Ship Target Detection Adversarial Attack and Defence Generalization Research

The synthetic aperture radar (SAR) image ship detection system needs to adapt to an increasingly complicated actual environment, and the requirements for the stability of the detection system continue to increase. Adversarial attacks deliberately add subtle interference to input samples and cause models to have high confidence in output errors. There are potential risks in a system, and input data that contain confrontation samples can be easily used by malicious people to attack the system. For a safe and stable model, attack algorithms need to be studied. The goal of traditional attack algorithms is to destroy models. When defending against attack samples, a system does not consider the generalization ability of the model. Therefore, this paper introduces an attack algorithm which can improve the generalization of models by based on the attributes of Gaussian noise, which is widespread in actual SAR systems. The attack data generated by this method have a strong effect on SAR ship detection models and can greatly reduce the accuracy of ship recognition models. While defending against attacks, filtering attack data can effectively improve the model defence capabilities. Defence training greatly improves the anti-attack capacity, and the generalization capacity of the model is improved accordingly.

Generalizability of Soft Sensors for Bioprocesses through Similarity Analysis and Phase-Dependent Recalibration

A soft sensor concept is typically developed and calibrated for individual bioprocesses in a time-consuming manual procedure. Following that, the prediction performance of these soft sensors degrades over time, due to changes in raw materials, biological variability, and modified process strategies. Through automatic adaptation and recalibration, adaptive soft sensor concepts have the potential to generalize soft sensor principles and make them applicable across bioprocesses. In this study, a new generalized adaptation algorithm for soft sensors is developed to provide phase-dependent recalibration of soft sensors based on multiway principal component analysis, a similarity analysis, and robust, generalist phase detection in multiphase bioprocesses. This generalist soft sensor concept was evaluated in two multiphase bioprocesses with various target values, media, and microorganisms. Consequently, the soft sensor concept was tested for biomass prediction in a Pichia pastoris process, and biomass and protein prediction in a Bacillus subtilis process, where the process characteristics (cultivation media and cultivation strategy) were varied. High prediction performance was demonstrated for P. pastoris processes (relative error = 6.9%) as well as B. subtilis processes in two different media during batch and fed-batch phases (relative errors in optimized high-performance medium: biomass prediction = 12.2%, protein prediction = 7.2%; relative errors in standard medium: biomass prediction = 12.8%, protein prediction = 8.8%).

Investigating Generalized Performance of Data-Constrained Supervised Machine Learning Models on Novel, Related Samples in Intrusion Detection

Recently proposed methods in intrusion detection are iterating on machine learning methods as a potential solution. These novel methods are validated on one or more datasets from a sparse collection of academic intrusion detection datasets. Their recognition as improvements to the state-of-the-art is largely dependent on whether they can demonstrate a reliable increase in classification metrics compared to similar works validated on the same datasets. Whether these increases are meaningful outside of the training/testing datasets is rarely asked and never investigated. This work aims to demonstrate that strong general performance does not typically follow from strong classification on the current intrusion detection datasets. Binary classification models from a range of algorithmic families are trained on the attack classes of CSE-CIC-IDS2018, a state-of-the-art intrusion detection dataset. After establishing baselines for each class at various points of data access, the same trained models are tasked with classifying samples from the corresponding attack classes in CIC-IDS2017, CIC-DoS2017 and CIC-DDoS2019. Contrary to what the baseline results would suggest, the models have rarely learned a generally applicable representation of their attack class. Stability and predictability of generalized model performance are central issues for all methods on all attack classes. Focusing only on the three best-in-class models in terms of interdataset generalization, reveals that for network-centric attack classes (brute force, denial of service and distributed denial of service), general representations can be learned with flat losses in classification performance (precision and recall) below 5%. Other attack classes vary in generalized performance from stark losses in recall (-35%) with intact precision (98+%) for botnets to total degradation of precision and moderate recall loss for Web attack and infiltration models. The core conclusion of this article is a warning to researchers in the field. Expecting results of proposed methods on the test sets of state-of-the-art intrusion detection datasets to translate to generalized performance is likely a serious overestimation. Four proposals to reduce this overestimation are set out as future work directions.

Short-term effects of visuomotor discrepancies on multisensory integration, proprioceptive recalibration, and motor adaptation

Information about the position of our hand is provided by multisensory signals that are often not perfectly aligned. Discrepancies between the seen and felt hand position or its movement trajectory engage the processes of 1) multisensory integration, 2) sensory recalibration, and 3) motor adaptation, which adjust perception and behavioral responses to apparently discrepant signals. To foster our understanding of the coemergence of these three processes, we probed their short-term dependence on multisensory discrepancies in a visuomotor task that has served as a model for multisensory perception and motor control previously. We found that the well-established integration of discrepant visual and proprioceptive signals is tied to the immediate discrepancy and independent of the outcome of the integration of discrepant signals in immediately preceding trials. However, the strength of integration was context dependent, being stronger in an experiment featuring stimuli that covered a smaller range of visuomotor discrepancies (±15°) compared with one covering a larger range (±30°). Both sensory recalibration and motor adaptation for nonrepeated movement directions were absent after two bimodal trials with same or opposite visuomotor discrepancies. Hence our results suggest that short-term sensory recalibration and motor adaptation are not an obligatory consequence of the integration of preceding discrepant multisensory signals.NEW & NOTEWORTHY The functional relation between multisensory integration and recalibration remains debated. We here refute the notion that they coemerge in an obligatory manner and support the hypothesis that they serve distinct goals of perception.

A Semi-Supervised Approach for Improving Generalization in Non-Intrusive Load Monitoring

Non-intrusive load monitoring (NILM) considers different approaches for disaggregating energy consumption in residential, tertiary, and industrial buildings to enable smart grid services. The main feature of NILM is that it can break down the bulk electricity demand, as recorded by conventional smart meters, into the consumption of individual appliances without the need for additional meters or sensors. Furthermore, NILM can identify when an appliance is in use and estimate its real-time consumption based on its unique consumption patterns. However, NILM is based on machine learning methods and its performance is dependent on the quality of the training data for each appliance. Therefore, a common problem with NILM systems is that they may not generalize well to new environments where the appliances are unknown, which hinders their widespread adoption and more significant contributions to emerging smart grid services. The main goal of the presented research is to apply a domain adversarial neural network (DANN) approach to improve the generalization of NILM systems. The proposed semi-supervised algorithm utilizes both labeled and unlabeled data and was tested on data from publicly available REDD and UK-DALE datasets. The results show a 3% improvement in generalization performance on highly uncorrelated data, indicating the potential for real-world applications.

Parent-Implemented Telepractice Autism Intervention: A Case Study of Maintenance and Generalization

The extent to which people maintain new skills and generalize those skills to new contexts without support are two aspects of intervention research that can be difficult to examine, especially over a sustained period of time and across a variety of contexts. In past research, we have explored teaching parents and caregivers to implement evidence-based communication strategies with their young children with autism who are minimally verbal. When a former research participant contacted us with a request to participate in our project again, four years later and with a different son, we used this as an opportunity to ask questions about her maintenance of the skills in using the targeted strategies, and her generalization of those skills to a different child. Using the data collected with her older son, Ali, and new data collected four years later with her younger son, Rami, we present a case study of this mother. We discuss the implications of the findings on interpreting the efficacy of the telepractice intervention's programming for generalization, identifying opportunities for refining the intervention, and insights useful for other intervention research.

What determines location specificity or generalization of transsaccadic learning?

Humans incorporate knowledge of transsaccadic associations into peripheral object perception. Several studies have shown that learning of new manipulated transsaccadic associations leads to a presaccadic perceptual bias. However, there was still disagreement whether this learning effect was location specific (Herwig, Weiß, & Schneider, 2018) or generalizes to new locations (Valsecchi & Gegenfurtner, 2016). The current study investigated under what conditions location generalization of transsaccadic learning occurs. In all experiments, there were acquisition phases in which the spatial frequency (Experiment 1) or the size (Experiment 2 and 3) of objects was changed transsaccadically. In the test phases, participants judged the respective feature of peripheral objects. These could appear either at the location where learning had taken place or at new locations. All experiments replicated the perceptual bias effect at the old learning locations. In two experiments, transsaccadic learning remained location specific even when learning occurred at multiple locations (Experiment 1) or with the feature of size (Experiment 2) for which a transfer had previously been shown. Only in Experiment 3 was a transfer of the learning effect to new locations observable. Here, learning only took place for one object and not for several objects that had to be discriminated. Therefore, one can conclude that, when specific associations are learned for multiple objects, transsaccadic learning stays location specific and when a transsaccadic association is learned for only one object it allows a generalization to other locations.

Distinct signatures of loss of consciousness in focal impaired awareness versus tonic-clonic seizures

Loss of consciousness is a hallmark of many epileptic seizures and carries risks of serious injury and sudden death. While cortical sleep-like activities accompany loss of consciousness during focal impaired awareness seizures, the mechanisms of loss of consciousness during focal to bilateral tonic-clonic seizures remain unclear. Quantifying differences in markers of cortical activation and ictal recruitment between focal impaired awareness and focal to bilateral tonic-clonic seizures may also help us to understand their different consequences for clinical outcomes and to optimize neuromodulation therapies. We quantified clinical signs of loss of consciousness and intracranial EEG activity during 129 focal impaired awareness and 50 focal to bilateral tonic-clonic from 41 patients. We characterized intracranial EEG changes both in the seizure onset zone and in areas remote from the seizure onset zone with a total of 3386 electrodes distributed across brain areas. First, we compared the dynamics of intracranial EEG sleep-like activities: slow-wave activity (1-4 Hz) and beta/delta ratio (a validated marker of cortical activation) during focal impaired awareness versus focal to bilateral tonic-clonic. Second, we quantified differences between focal to bilateral tonic-clonic and focal impaired awareness for a marker validated to detect ictal cross-frequency coupling: phase-locked high gamma (high-gamma phased-locked to low frequencies) and a marker of ictal recruitment: the epileptogenicity index. Third, we assessed changes in intracranial EEG activity preceding and accompanying behavioural generalization onset and their correlation with electromyogram channels. In addition, we analysed human cortical multi-unit activity recorded with Utah arrays during three focal to bilateral tonic-clonic seizures. Compared to focal impaired awareness, focal to bilateral tonic-clonic seizures were characterized by deeper loss of consciousness, even before generalization occurred. Unlike during focal impaired awareness, early loss of consciousness before generalization was accompanied by paradoxical decreases in slow-wave activity and by increases in high-gamma activity in parieto-occipital and temporal cortex. After generalization, when all patients displayed loss of consciousness, stronger increases in slow-wave activity were observed in parieto-occipital cortex, while more widespread increases in cortical activation (beta/delta ratio), ictal cross-frequency coupling (phase-locked high gamma) and ictal recruitment (epileptogenicity index). Behavioural generalization coincided with a whole-brain increase in high-gamma activity, which was especially synchronous in deep sources and could not be explained by EMG. Similarly, multi-unit activity analysis of focal to bilateral tonic-clonic revealed sustained increases in cortical firing rates during and after generalization onset in areas remote from the seizure onset zone. Overall, these results indicate that unlike during focal impaired awareness, the neural signatures of loss of consciousness during focal to bilateral tonic-clonic consist of paradoxical increases in cortical activation and neuronal firing found most consistently in posterior brain regions. These findings suggest differences in the mechanisms of ictal loss of consciousness between focal impaired awareness and focal to bilateral tonic-clonic and may account for the more negative prognostic consequences of focal to bilateral tonic-clonic.

Assessment of the capacity to modulate brain signals in a home-based SMR neurofeedback training setting

Electroencephalogram (EEG)-based neurofeedback (NF) is mainly used in clinical settings as a therapeutic intervention or to optimize performance in healthy individuals. Home-based NF systems are available and might facilitate general access to NF training, especially when repeated training sessions are necessary. However, it remains an open question whether NF training at home is possible without remote monitoring. In the present study, we assessed the capacity of healthy individuals to modulate their own EEG activity when using a home-based NF training system in a comparable manner as if participants had purchased a commercially available NF system. Participants' face-to-face contact with experimenters was reduced to a minimum, and instructions were provided only in the form of written information or videos. Initially, 38 participants performed 9 sessions of sensorimotor rhythm (SMR) (12-15 Hz) based NF training (three generalization sessions, six training sessions). An active control group (n = 19) received feedback on random EEG frequencies. Because of technical problems, bad EEG data quality, or non-compliance, 21 participants had to be excluded from the final data analysis, providing first evidence for the difficulties of non-supervised home-based NF training. In this study, participants were not able to modulate their own brain activity in a desired direction during NF training. Our results indicate that personal interaction with a NF expert might be of relevance and that remote supervision of the training data and more direct communication with the NF users are necessary to enable successful NF training performance. We provide suggestions for the development and implementation of home-based NF systems.

mGluR5 in amygdala modulates fear memory generalization

Introduction

Fear memory generalization is regarded as the core characteristic of posttraumatic stress disorder (PTSD) development. However, the mechanism that contributes to the generalization of conditioned fear memory is still unclear. The generalization is generally considered to be a mismatch that occurs during memory consolidation.

Methods

Foot shocks and tones were given as unconditioned stress and conditioned stress, respectively for fear conditioning training. Immunofluorescence staining, western blotting and qPCR were performed to determine the expression of different genes in amygdala of mice after fear conditioning training. Cycloheximide was used as a protein synthesis inhibitor and 2-methyl-6-phenylethynyl-pyridine was injected for mGluR5 inhibition.

Results

Fear conditioning using caused incremental generalization, which was clearly observed during training. The density of c-Fos⁺ cells or the synaptic p-NMDAR expression did not differ with stress intensities. Strong-shock fear conditioning could induce significant mGluR5 de novo synthesis in the amygdala, which was not observed in the weak-shock group. Inhibition of mGluR5 impaired fear memory generalization induced by strong-shock fear conditioning, but the generalization level induced by weak-shock training was enhanced.

Discussion

These results indicated that mGluR5 in the amygdala is critical to the function of inappropriate fear memory generalization and suggested that this may be a potential target for the treatment of PTSD.

Targeted neurorehabilitation strategies in post-stroke aphasia

BACKGROUND

Aphasia is a debilitating language impairment, affecting millions of people worldwide. About 40% of stroke survivors develop chronic aphasia, resulting in life-long disability.

OBJECTIVE

This review examines extrinsic and intrinsic neuromodulation techniques, aimed at enhancing the effects of speech and language therapies in stroke survivors with aphasia.

METHODS

We discuss the available evidence supporting the use of transcranial direct current stimulation (tDCS), repetitive transcranial magnetic stimulation, and functional MRI (fMRI) real-time neurofeedback in aphasia rehabilitation.

RESULTS

This review systematically evaluates studies focusing on efficacy and implementation of specialized methods for post-treatment outcome optimization and transfer to functional skills. It considers stimulation target determination and various targeting approaches. The translation of neuromodulation interventions to clinical practice is explored, emphasizing generalization and functional communication. The review also covers real-time fMRI neurofeedback, discussing current evidence for efficacy and essential implementation parameters. Finally, we address future directions for neuromodulation research in aphasia.

CONCLUSIONS

This comprehensive review aims to serve as a resource for a broad audience of researchers and clinicians interested in incorporating neuromodulation for advancing aphasia care.

Pavlovian Disgust Conditioning and Generalization: Specificity and Associations With Individual Differences

Although studies have identified differences between fear and disgust conditioning, much less is known about the generalization of conditioned disgust. This is an important gap in the literature given that overgeneralization of conditioned disgust to neutral stimuli may have clinical implications. To address this knowledge gap, female participants (n = 80) completed a Pavlovian conditioning procedure in which one neutral food item (conditioned stimulus; CS+) was followed by disgusting videos of individuals vomiting (unconditioned stimulus; US) and another neutral food item (CS-) was not reinforced with the disgusting video. Following this acquisition phase, there was an extinction phase in which both CSs were presented unreinforced. Importantly, participants also evaluated generalization stimuli (GS+, GS-) that resembled, but were distinct from, the CS after each conditioning phase. As predicted, the CS+ was rated as significantly more disgusting and fear inducing than the CS- after acquisition and this pattern persisted after extinction. However, disgust ratings of the CS+ after acquisition were significantly larger than fear ratings. Participants also rated the GS+ as significantly more disgusting, but not fear inducing, than the GS- after acquisition. However, this effect was not observed after extinction. Disgust proneness did predict a greater increase in disgust and fear ratings of the CS+ relative to the CS- after acquisition and extinction. In contrast, trait anxiety predicted only higher fear ratings to the CS+ relative to the CS- after acquisition and extinction. Disgust proneness nor trait anxiety predicted the greater increase in disgust to the GS+ relative to the GS- after acquisition. These findings suggest that while conditioned disgust can generalize, individual difference variables that predict generalization remain unclear. The implications of these findings for disorders of disgust are discussed.

Enhanced Mental Reinstatement of Exposure to Improve Extinction Generalization: A Study on Claustrophobia and MRI Fear

Fear of enclosed spaces prevents many people from receiving magnetic resonance imaging (MRI) scans. Although exposure therapy can effectively treat such fears, reductions in fear during exposure often do not generalize beyond the context in which they took place. This study tested a strategy designed to increase generalization, which involved revisiting the memory of a prior exposure to enhance retrieval of extinction learning. Forty-five participants with claustrophobia that included fear of MRI scans underwent a series of exposures lying inside a narrow cabinet. One week later, participants were randomly assigned to enhanced mental reinstatement (EMR) or control procedures. Prior to entering a mock MRI scanner, EMR participants recalled the memory of exposure training and listened to an audio recording of themselves describing what they learned, whereas control participants recalled a neutral memory. Compared to the control condition, EMR led to significantly reduced heart rate reactivity in the mock MRI scanner, but not self-reported fear or avoidance. There were no differences between conditions in claustrophobia symptoms or MRI fear at 1-month follow-up. Results suggest some benefits of mental reinstatement for improving generalization of gains following exposure training for claustrophobia, with measures of subjective fear and physiological arousal showing discordant outcomes.

Using Telehealth to Program Generalization of Caregiver Behavior

Stokes and Baer, Journal of Applied Behavior Analysis, 10(2), 349-367 (1977) provided guidelines to assist practitioners with programming for the generalization of behavior change. Despite the suggestions provided in their seminal paper, generalization remains an often overlooked area within behavior analytic research and practice. In addition, few studies have described explicit strategies to program for the generalization of caregiver behaviors that are consistent with interventions to reduce child challenging behavior. In the current discussion, we describe how telehealth provides a potential avenue for practitioners to focus on generalization. Telehealth helps practitioners access behavior-change agents, materials, and contexts that they may not directly contact in educational and clinical environments. Using telehealth to target these areas early on, and throughout treatment for child challenging behavior, may facilitate more rapid treatment success and maintenance. We provide a case example to demonstrate the use of telehealth to program the generalization of a mother's treatment plan implementation to reduce the severe challenging behavior of an adolescent. We report clinically and socially significant outcomes related to caregiver fidelity and challenging behavior reduction.

Accounting for the valley of recovery during post-stroke rehabilitation training via a model-based analysis of macaque manual dexterity

Background

True recovery, in which a stroke patient regains the same precise motor skills observed in prestroke conditions, is the fundamental goal of rehabilitation training. However, a transient drop in task performance during rehabilitation training after stroke, observed in human clinical outcome as well as in both macaque and squirrel monkey retrieval data, might prevent smooth transitions during recovery. This drop, i.e., recovery valley, often occurs during the transition from compensatory skill to precision skill. Here, we sought computational mechanisms behind such transitions and recovery. Analogous to motor skill learning, we considered that the motor recovery process is composed of spontaneous recovery and training-induced recovery. Specifically, we hypothesized that the interaction of these multiple skill update processes might determine profiles of the recovery valley.

Methods

A computational model of motor recovery was developed based on a state-space model of motor learning that incorporates a retention factor and interaction terms for training-induced recovery and spontaneous recovery. The model was fit to previously reported macaque motor recovery data where the monkey practiced precision grip skills after a lesion in the sensorimotor area in the cortex. Multiple computational models and the effects of each parameter were examined by model comparisons based on information criteria and sensitivity analyses of each parameter.

Result

Both training-induced and spontaneous recoveries were necessary to explain the behavioral data. Since these two factors contributed following logarithmic function, the training-induced recovery were effective only after spontaneous biological recovery had developed. In the training-induced recovery component, the practice of the compensation also contributed to recovery of the precision grip skill as if there is a significant generalization effect of learning between these two skills. In addition, a retention factor was critical to explain the recovery profiles.

Conclusions

We found that spontaneous recovery, training-induced recovery, retention factors, and interaction terms are crucial to explain recovery and recovery valley profiles. This simulation-based examination of the model parameters provides suggestions for effective rehabilitation methods to prevent the recovery valley, such as plasticity-promoting medications, brain stimulation, and robotic rehabilitation technologies.

Transfer Learning Strategy in Neural Network Application for Underwater Visible Light Communication System

Post-equalization using neural network (NN) is a promising technique that models and offsets the nonlinear distortion in visible light communication (VLC) channels, which is recognized as an essential component in the incoming 6G era. NN post-equalizer is good at modeling complex channel effects without previously knowing the law of physics during the transmission. However, the trained NN might be weak in generalization, and thus consumes considerable computation in retraining new models for different channel conditions. In this paper, we studied transfer learning strategy, growing DNN models from a well-trained 'stem model' instead of exhaustively training multiple models from randomly initialized states. It extracts the main feature of the channel first whose signal power balances the signal-to-noise ratio and the nonlinearity, and later focuses on the detailed difference in other channel conditions. Compared with the exhaustive training strategy, stem-originated DNN models achieve 64% of the working range with five times the training efficiency at most or more than 95% of the working range with 150% higher efficiency. This finding is beneficial to improving the feasibility of DNN application in real-world UVLC systems.

Do Spiders Ride on the Fear of Scorpions? A Cross-Cultural Eye Tracking Study

Deep fear of spiders is common in many countries, yet its origin remains unexplained. In this study, we tested a hypothesis based on recent studies suggesting that fear of spiders might stem from a generalized fear of chelicerates or fear of scorpions. To this end, we conducted an eye tracking experiment using a spontaneous gaze preference paradigm, with spiders and scorpions (previously neglected but crucial stimuli) as threatening stimuli and grasshoppers as control stimuli. In total, 67 participants from Somaliland and 67 participants from the Czech Republic were recruited and presented with a sequence of paired images. Both Somali and Czech people looked longer (total duration of the gaze) and more often (number of fixations) on the threatening stimuli (spiders and scorpions) when presented with a control (grasshopper). When both threatening stimuli were presented together, Somali participants focused significantly more on the scorpion, whereas in Czech participants, the effect was less pronounced, and in Czech women it was not significant. This supports the hypothesis that fear of spiders originated as a generalized fear of scorpions. Moreover, the importance of spiders as fear-eliciting stimuli may be enhanced in the absence of scorpions in the environment.

Episodes of experience and generative intelligence

How do humans, including toddlers, take knowledge from past experiences and apply this knowledge in new ways? Current approaches to human and artificial intelligence (AI) fail to offer satisfactory explanations. We suggest the explanation will be found in the coherence statistics of the individual time-extended episodes of human experience and the cognitive processes those statistics engage.

Memory decay and generalization following distinct motor learning mechanisms

Motor skill learning is considered to arise out of contributions from multiple learning mechanisms, including error-based learning (EBL), use-dependent learning (UDL), and reinforcement learning (RL). These learning mechanisms exhibit dissociable roles and engage different neural circuits during skill acquisition. However, it remains largely unknown how a newly formed motor memory acquired through each learning mechanism decays over time and whether distinct learning mechanisms produce different generalization patterns. Here, we used variants of reaching paradigms that dissociated these learning mechanisms to examine the time course of memory decay following each learning and the generalization patterns of each learning. We found that motor memories acquired through these learning mechanisms decayed as a function of time. Notably, 15 min, 6 h, and 24 h after acquisition, the memory of EBL decayed much greater than that of RL. The memory acquired through UDL faded away within a few minutes. Motor memories formed through EBL and RL for given movement directions generalized to untrained movement directions, with the generalization of EBL being greater than that of RL. In contrast, motor memory of UDL could not generalize to untrained movement directions. These results suggest that distinct learning mechanisms exhibit different patterns of memory decay and generalization.NEW & NOTEWORTHY Motor skill learning is likely to involve error-based learning, use-dependent plasticity, and operant reinforcement. Here, we showed that these dissociable learning mechanisms exhibited distinct patterns of memory decay and generalization. With a better understanding of the characteristics of these learning mechanisms, it becomes possible to regulate each learning process separately to improve neurological rehabilitation.

Childhood Maltreatment in Females Is Associated with Enhanced Fear Acquisition and an Overgeneralization of Fear

Childhood maltreatment may alter fear neurocircuitry, which results in pathological anxiety and depression. One alteration of fear-related behaviors that has been observed in several psychiatric populations is an overgeneralization of fear. Thus, we examined the association between childhood maltreatment and fear generalization in a non-clinical sample of young adults. Two hundred and ninety-one participants underwent differential fear conditioning in a fear-potentiated startle paradigm. One visual stimulus (CS+), but not another (CS-), was associated with an aversive airblast to the throat (US) during acquisition. The next day, participants were tested for their fear responses to the CS+, CS-, and several generalization stimuli (GS) without the presence of the US. Participants also completed questionnaires that assessed symptoms of childhood maltreatment, anxiety, depression, and post-traumatic stress disorder (PTSD). Participants reporting high childhood maltreatment (n = 71; 23 males, 48 females) exhibited significantly greater anxiety, depression, and symptoms of PTSD than participants reporting low childhood maltreatment (n = 220; 133 males, 87 females). Females reporting high childhood maltreatment demonstrated significantly enhanced fear learning and greater fear generalization, based on their fear-potentiated startle responses. Our findings suggest that childhood maltreatment may sex-dependently influence the development of fear neurocircuitry and result in greater fear generalization in maltreated females.

Adult neurogenesis acts as a neural regularizer

New neurons are continuously generated in the subgranular zone of the dentate gyrus throughout adulthood. These new neurons gradually integrate into hippocampal circuits, forming new naive synapses. Viewed from this perspective, these new neurons may represent a significant source of "wiring" noise in hippocampal networks. In machine learning, such noise injection is commonly used as a regularization technique. Regularization techniques help prevent overfitting training data and allow models to generalize learning to new, unseen data. Using a computational modeling approach, here we ask whether a neurogenesis-like process similarly acts as a regularizer, facilitating generalization in a category learning task. In a convolutional neural network (CNN) trained on the CIFAR-10 object recognition dataset, we modeled neurogenesis as a replacement/turnover mechanism, where weights for a randomly chosen small subset of hidden layer neurons were reinitialized to new values as the model learned to categorize 10 different classes of objects. We found that neurogenesis enhanced generalization on unseen test data compared to networks with no neurogenesis. Moreover, neurogenic networks either outperformed or performed similarly to networks with conventional noise injection (i.e., dropout, weight decay, and neural noise). These results suggest that neurogenesis can enhance generalization in hippocampal learning through noise injection, expanding on the roles that neurogenesis may have in cognition.

Generalization of Deep Learning in Digital Pathology: Experience in Breast Cancer Metastasis Detection

Poor generalizability is a major barrier to clinical implementation of artificial intelligence in digital pathology. The aim of this study was to test the generalizability of a pretrained deep learning model to a new diagnostic setting and to a small change in surgical indication. A deep learning model for breast cancer metastases detection in sentinel lymph nodes, trained on CAMELYON multicenter data, was used as a base model, and achieved an AUC of 0.969 (95% CI 0.926-0.998) and FROC of 0.838 (95% CI 0.757-0.913) on CAMELYON16 test data. On local sentinel node data, the base model performance dropped to AUC 0.929 (95% CI 0.800-0.998) and FROC 0.744 (95% CI 0.566-0.912). On data with a change in surgical indication (axillary dissections) the base model performance indicated an even larger drop with a FROC of 0.503 (95%CI 0.201-0.911). The model was retrained with addition of local data, resulting in about a 4% increase for both AUC and FROC for sentinel nodes, and an increase of 11% in AUC and 49% in FROC for axillary nodes. Pathologist qualitative evaluation of the retrained model´s output showed no missed positive slides. False positives, false negatives and one previously undetected micro-metastasis were observed. The study highlights the generalization challenge even when using a multicenter trained model, and that a small change in indication can considerably impact the model´s performance.