Locus coeruleus neurons encode the subjective difficulty of triggering and executing actions

The brain stem noradrenergic nucleus locus coeruleus (LC) is involved in various costly processes: arousal, stress, and attention. Recent work has pointed toward an implication in physical effort, and indirect evidence suggests that the LC could be also involved in cognitive effort. To assess the dynamic relation between LC activity, effort production, and difficulty, we recorded the activity of 193 LC single units in 5 monkeys performing 2 discounting tasks (a delay discounting task and a force discounting task), as well as a simpler target detection task where conditions were matched for difficulty and only differed in terms of sensory-motor processes. First, LC neurons displayed a transient activation both when monkeys initiated an action and when exerting force. Second, the magnitude of the activation scaled with the associated difficulty, and, potentially, the corresponding amount of effort produced, both for decision and force production. Indeed, at action initiation in both discounting tasks, LC activation increased in conditions associated with lower average engagement rate, i.e., those requiring more cognitive control to trigger the response. Decision-related activation also scaled with response time (RT), over and above task parameters, in line with the idea that it reflects the amount of resources (here time) spent on the decision process. During force production, LC activation only scaled with the amount of force produced in the force discounting task, but not in the control target detection task, where subjective difficulty was equivalent across conditions. Our data show that LC neurons dynamically track the amount of effort produced to face both cognitive and physical challenges with a subsecond precision. This works provides key insight into effort processing and the contribution of the noradrenergic system, which is affected in several pathologies where effort is impaired, including Parkinson disease and depression.

Hyperspectral Target Detection with an Auxiliary Generative Adversarial Network

In recent years, the deep neural network has shown a strong presence in classification tasks and its effectiveness has been well proved. However, the framework of DNN usually requires a large number of samples. Compared to the training sets in classification tasks, the training sets for the target detection of hyperspectral images may only include a few target spectra which are quite limited and precious. The insufficient labeled samples make the DNN-based hyperspectral target detection task a challenging problem. To address this problem, we propose a hyperspectral target detection approach with an auxiliary generative adversarial network. Specifically, the training set is first expanded by generating simulated target spectra and background spectra using the generative adversarial network. Then, a classifier which is highly associated with the discriminator of the generative adversarial network is trained based on the real and the generated spectra. Finally, in order to further suppress the background, guided filters are utilized to improve the smoothness and robustness of the detection results. Experiments conducted on real hyperspectral images show the proposed approach is able to perform more efficiently and accurately compared to other target detection approaches.

Visual Perception of Apparent Motion Abides by Minimization Principles of Geometry

Apparent motion is a robust perceptual phenomenon in which observers perceive a stimulus traversing the vacant visual space between two flashed stimuli. Although it is known that the “filling-in” of apparent motion favors the simplest and most economical path, the interpolative computations remain poorly understood. Here, we tested whether the perception of apparent motion is best characterized by Newtonian physics or kinematic geometry. Participants completed a target detection task while Pacmen- shaped objects were presented in succession to create the perception of apparent motion. We found that target detection was impaired when apparent motion, as predicted by kinematic geometry, not Newtonian physics, obstructed the target’s location. Our findings shed light on the computations employed by the visual system, suggesting specifically that the “filling-in” perception of apparent motion may be dominated by kinematic geometry, not Newtonian physics.

A Comparison of Auditory and Visual Representations of System Confidence to Support Trust Specificity, Attention Management, and Joint Performance in Human-Machine Teams

Breakdowns in human-robot teaming can result from trust miscalibration, i.e., a poor mapping of trust to a system’s capabilities, resulting in misuse or disuse of the technology. Trust miscalibration also negatively affects operators’ top-down attention allocation and monitoring of the system. This experiment assessed the efficacy of visual and auditory representations of a system’s confidence in its own abilities for supporting trust specificity, attention management and joint performance in the context of a UAV-supported target detection task. In contrast to earlier studies, neither visual nor auditory confidence information improved detection accuracy. Visual representations of confidence led to slower response times than auditory representations, likely due to resource competition with the visual target detection task. Finally, slower response times were observed when a UAV incorrectly detected a target. Results from this study can inform the design of visual and auditory representations of system confidence in human-machine teams with high attention demands.

Talker variability facilitates the statistical learning of speech sounds

Natural speech contains many sources of acoustic variability both within and between talkers, which challenges speech recognition in some contexts but may facilitate language understanding in novel listening situations. Despite this ubiquitous variability, most previous studies that have examined the ability to extract sound patterns in speech—known as statistical learning—have used highly controlled, artificial, monotonic streams of syllables. Thus, it is unknown whether variability in speech may help or hinder statistical learning – an important question to resolve if statistical learning does indeed play a role in the segmentation of naturally spoken language, as widely theorized. Here, we assessed whether the use of naturally produced, variable speech sounds produced by multiple talkers benefits or impairs statistical learning, including the ability to generalize patterns to a novel talker. During training, participants listened to approximately 12 minutes of continuous speech made up of repeating trisyllabic words, spoken either by a single talker (single-talker condition) or four talkers speaking for three minutes each (multiple-talker condition). Post-training, all participants completed three assessments of learning: (1) an explicit familiarity rating task, (2) an explicit forced-choice recognition task, and (3) an implicit syllable target detection task. Results indicated that participants in both training conditions showed evidence of statistical learning across all assessments, providing an important demonstration that statistical learning is robust to additional variability in the speech signal. Further, in both the forced-choice recognition task and target detection task, participants in the multiple-talker condition showed evidence of facilitated statistical learning, particularly when listening to a novel talker. In the familiarity rating task, performance was comparable between conditions; however, participants trained with multiple talkers were less likely to conflate word familiarity with talker voice familiarity. Overall, these results suggest that training with multiple talkers can improve aspects of statistical learning across multiple measures of learning.

LFM: A Lightweight LCD Algorithm Based on Feature Matching between Similar Key Frames

Loop Closure Detection (LCD) is an important technique to improve the accuracy of Simultaneous Localization and Mapping (SLAM). In this paper, we propose an LCD algorithm based on binary classification for feature matching between similar images with deep learning, which greatly improves the accuracy of LCD algorithm. Meanwhile, a novel lightweight convolutional neural network (CNN) is proposed and applied to the target detection task of key frames. On this basis, the key frames are binary classified according to their labels. Finally, similar frames are input into the improved lightweight feature matching network based on Transformer to judge whether the current position is loop closure. The experimental results show that, compared with the traditional method, LFM-LCD has higher accuracy and recall rate in the LCD task of indoor SLAM while ensuring the number of parameters and calculation amount. The research in this paper provides a new direction for LCD of robotic SLAM, which will be further improved with the development of deep learning.

Lifting Scheme-Based Sparse Density Feature Extraction for Remote Sensing Target Detection

The design of backbones is of great significance for enhancing the location and classification precision in the remote sensing target detection task. Recently, various approaches have been proposed on altering the feature extraction density in the backbones to enlarge the receptive field, make features prominent, and reduce computational complexity, such as dilated convolution and deformable convolution. Among them, one of the most widely used methods is strided convolution, but it loses the information about adjacent feature points which leads to the omission of some useful features and the decrease of detection precision. This paper proposes a novel sparse density feature extraction method based on the relationship between the lifting scheme and convolution, which improves the detection precision while keeping the computational complexity almost the same as the strided convolution. Experimental results on remote sensing target detection indicate that our proposed method improves both detection performance and network efficiency.

What canonical online and offline measures of statistical learning can and cannot tell us

Statistical learning (SL) allows individuals to rapidly detect regularities in the sensory environment. We replicated previous findings showing that adult participants become sensitive to the implicit structure in a continuous speech stream of repeating tri-syllabic pseudowords within minutes, as measured by standard tests in the SL literature: a target detection task and a 2AFC word recognition task. Consistent with previous findings, we found only a weak correlation between these two measures of learning, leading us to question whether there is overlap between the information captured by these two tasks. Representational similarity analysis on reaction times measured during the target detection task revealed that reaction time data reflect sensitivity to transitional probability, triplet position, word grouping, and duplet pairings of syllables. However, individual performance on the word recognition task was not predicted by similarity measures derived for any of these four features. We conclude that online detection tasks provide richer and multi-faceted information about the SL process, as compared with 2AFC recognition tasks, and may be preferable for gaining insight into the dynamic aspects of SL.