Deep neural networks (DNNs) have achieved state-of-the-art performance in many important domains, including medical diagnosis, security, and autonomous driving. In domains where safety is highly ...critical, an erroneous decision can result in serious consequences. While a perfect prediction accuracy is not always achievable, recent work on Bayesian deep networks shows that it is possible to know when DNNs are more likely to make mistakes. Knowing what DNNs do not know is desirable to increase the safety of deep learning technology in sensitive applications; Bayesian neural networks attempt to address this challenge. Traditional approaches are computationally intractable and do not scale well to large, complex neural network architectures. In this paper, we develop a theoretical framework to approximate Bayesian inference for DNNs by imposing a Bernoulli distribution on the model weights. This method called Monte Carlo DropConnect (MC-DropConnect) gives us a tool to represent the model uncertainty with little change in the overall model structure or computational cost. We extensively validate the proposed algorithm on multiple network architectures and datasets for classification and semantic segmentation tasks. We also propose new metrics to quantify uncertainty estimates. This enables an objective comparison between MC-DropConnect and prior approaches. Our empirical results demonstrate that the proposed framework yields significant improvement in both prediction accuracy and uncertainty estimation quality compared to the state of the art.
Detecting roads in automatic driving environments poses a challenge due to issues such as boundary fuzziness, occlusion, and glare from light. We believe that two factors are instrumental in ...addressing these challenges and enhancing detection performance: global context dependency and effective feature representation that prioritizes important feature channels. To tackle these issues, we introduce DTRoadseg, a novel duplex Transformer-based heterogeneous feature fusion network designed for road segmentation. DTRoadseg leverages a duplex encoder architecture to extract heterogeneous features from both RGB images and point-cloud depth images. Subsequently, we introduce a multi-source Heterogeneous Feature Reinforcement Block (HFRB) for fusion of the encoded features, comprising a Heterogeneous Feature Fusion Module (HFFM) and a Reinforcement Fusion Module (RFM). The HFFM leverages the self-attention mechanisms of Transformers to achieve effective fusion through token interactions, while the RFM focuses on emphasizing informative features while downplaying less important ones, thereby reinforcing feature fusion. Finally, a Transformer decoder is utilized to produce the final semantic prediction. Furthermore, we employ a boundary loss function to optimize the segmentation structure area, reduce false detection areas, and improve model accuracy. Extensive experiments are carried out on the KITTI road dataset. The results demonstrate that, compared with state-of-the-art methods, DTRoadseg exhibits superior performance, achieving an average accuracy of 97.01%, a Recall of 96.35%, and runs at a speed of 0.09 s per picture.
Pupillometry has been one of the most widely used response systems in psychophysiology. Changes in pupil size can reflect diverse cognitive and emotional states, ranging from arousal, interest and ...effort to social decisions, but they are also widely used in clinical practice to assess patients’ brain functioning. As a result, research involving pupil size measurements has been reported in practically all psychology, psychiatry, and psychophysiological research journals, and now it has found its way into the primatology literature as well as into more practical applications, such as using pupil size as a measure of fatigue or a safety index during driving. The different systems used for recording pupil size are almost as variable as its applications, and all yield, as with many measurement techniques, a substantial amount of noise in addition to the real pupillometry data. Before analyzing pupil size, it is therefore of crucial importance first to detect this noise and deal with it appropriately, even prior to (if need be) resampling and baseline-correcting the data. In this article we first provide a short review of the literature on pupil size measurements, then we highlight the most important sources of noise and show how these can be detected. Finally, we provide step-by-step guidelines that will help those interested in pupil size to preprocess their data correctly. These guidelines are accompanied by an open source MATLAB script (available at
https://github.com/ElioS-S/pupil-size
). Given that pupil diameter is easily measured by standard eyetracking technologies and can provide fundamental insights into cognitive and emotional processes, it is hoped that this article will further motivate scholars from different disciplines to study pupil size.
Abstract Achieving fast and precise initialization of qubits is a critical requirement for the successful operation of quantum computers. The combination of engineered environments with all-microwave ...techniques has recently emerged as a promising approach for the reset of superconducting quantum devices. In this work, we experimentally demonstrate the utilization of a single-junction quantum-circuit refrigerator (QCR) for an expeditious removal of several excitations from a transmon qubit. The QCR is indirectly coupled to the transmon through a resonator in the dispersive regime, constituting a carefully engineered environmental spectrum for the transmon. Using single-shot readout, we observe excitation stabilization times down to roughly 500 ns, a 20-fold speedup with QCR and a simultaneous two-tone drive addressing the e – f and f 0– g 1 transitions of the system. Our results are obtained at a 48-mK fridge temperature and without postselection, fully capturing the advantage of the protocol for the short-time dynamics and the drive-induced detrimental asymptotic behavior in the presence of relatively hot other baths of the transmon. We validate our results with a detailed Liouvillian model truncated up to the three-excitation subspace, from which we estimate the performance of the protocol in optimized scenarios, such as cold transmon baths and fine-tuned driving frequencies. These results pave the way for optimized reset of quantum-electric devices using engineered environments and for dissipation-engineered state preparation.
The role of cosmic rays generated by supernovae and young stars has very recently begun to receive significant attention in studies of galaxy formation and evolution due to the realization that ...cosmic rays can efficiently accelerate galactic winds. Microscopic cosmic-ray transport processes are fundamental for determining the efficiency of cosmic-ray wind driving. Previous studies modeled cosmic-ray transport either via a constant diffusion coefficient or via streaming proportional to the Alfvén speed. However, in predominantly cold, neutral gas, cosmic rays can propagate faster than in the ionized medium, and the effective transport can be substantially larger; i.e., cosmic rays can decouple from the gas. We perform three-dimensional magnetohydrodynamical simulations of patches of galactic disks including the effects of cosmic rays. Our simulations include the decoupling of cosmic rays in the cold, neutral interstellar medium. We find that, compared to the ordinary diffusive cosmic-ray transport case, accounting for the decoupling leads to significantly different wind properties, such as the gas density and temperature, significantly broader spatial distribution of cosmic rays, and higher wind speed. These results have implications for X-ray, γ-ray, and radio emission, and for the magnetization and pollution of the circumgalactic medium by cosmic rays.
Perceptual experiences may arise from neuronal activity patterns in mammalian neocortex. We probed mouse neocortex during visual discrimination using a red-shifted channelrhodopsin (ChRmine, ...discovered through structure-guided genome mining) alongside multiplexed multiphoton-holography (MultiSLM), achieving control of individually specified neurons spanning large cortical volumes with millisecond precision. Stimulating a critical number of stimulus-orientation-selective neurons drove widespread recruitment of functionally related neurons, a process enhanced by (but not requiring) orientation-discrimination task learning. Optogenetic targeting of orientation-selective ensembles elicited correct behavioral discrimination. Cortical layer-specific dynamics were apparent, as emergent neuronal activity asymmetrically propagated from layer 2/3 to layer 5, and smaller layer 5 ensembles were as effective as larger layer 2/3 ensembles in eliciting orientation discrimination behavior. Population dynamics emerging after optogenetic stimulation both correctly predicted behavior and resembled natural internal representations of visual stimuli at cellular resolution over volumes of cortex.
The formulation of high-efficient energy management strategy (EMS) for hybrid electric vehicles (HEVs) becomes the most crucial task owing to the variation of electrified powertrain topology and ...uncertainty of driving scenarios. In this study, a deep reinforcement learning (DRL) algorithm, namely TD3, is leveraged to derivate intelligent EMS for HEV. A heuristic rule-based local controller (LC) is embedded within the DRL loop to eliminate irrational torque allocation with considering the characteristics of powertrain components. In order to resolve the influence of environmental disturbance, a hybrid experience replay (HER) method is proposed based on a mixed experience buffer (MEB) consisting of offline computed optimal experience and online learned experience. The results indicate that improved TD3 based EMS obtained the best fuel optimality, fastest convergence speed and highest robustness in comparison to typical value-based and policy-based DRL EMSs under various driving cycles. LC leads to a boosting effect on the convergence speed of TD3-based EMS wherein a “warm” start of exploring is exhibited. Meanwhile, by incorporating HER coupled with MEB, the impact of environmental disturbance including load mass and road gradient, as an increase of input observations, can be negligible to the performance of TD3-based EMS.
Display omitted
•A novel DRL algorithm TD3 is leveraged to formulate intelligent HEV EMS.•A heuristic rule-based local controller is embedded in the DRL loop to eliminate irrational exploration.•A hybrid experience replay method is proposed through mixed experience buffer consisting of environmental disturbances.•Comparison analysis is systematically conducted among DDPG, double DQN, dueling DQN with proposed DRL-based EMS.
Solar eruptions are the main driver of space-weather disturbances at Earth. Extreme events are of particular interest, not only because of the scientific challenges they pose, but also because of ...their possible societal consequences. Here we present a magnetohydrodynamic (MHD) simulation of the 2000 July 14 "Bastille Day" eruption, which produced a very strong geomagnetic storm. After constructing a "thermodynamic" MHD model of the corona and solar wind, we insert a magnetically stable flux rope along the polarity inversion line of the eruption's source region and initiate the eruption by boundary flows. More than 1033 erg of magnetic energy is released in the eruption within a few minutes, driving a flare, an extreme-ultraviolet wave, and a coronal mass ejection (CME) that travels in the outer corona at 1500 km s−1, close to the observed speed. We then propagate the CME to Earth, using a heliospheric MHD code. Our simulation thus provides the opportunity to test how well in situ observations of extreme events are matched if the eruption is initiated from a stable magnetic equilibrium state. We find that the flux-rope center is very similar in character to the observed magnetic cloud, but arrives 8.5 hr later and 15° too far to the north, with field strengths that are too weak by a factor of 1.6. The front of the flux rope is highly distorted, exhibiting localized magnetic field concentrations as it passes 1 au. We discuss these properties with regard to the development of space-weather predictions based on MHD simulations of solar eruptions.
Using a general model of opinion dynamics, we conduct a systematic investigation of key mechanisms driving elite polarization in the United States. We demonstrate that the self-reinforcing nature of ...elite-level processes can explain this polarization, with voter preferences accounting for its asymmetric nature. Our analysis suggests that subtle differences in the frequency and amplitude with which public opinion shifts left and right over time may have a differential effect on the self-reinforcing processes of elites, causing Republicans to polarize more quickly than Democrats. We find that as self-reinforcement approaches a critical threshold, polarization speeds up. Republicans appear to have crossed that threshold while Democrats are currently approaching it.