Offline Reinforcement Learning (RL) aims to learn policies by using static datasets. The extrapolation error in out-of-distribution (OOD) samples can cause off-policy RL algorithms to perform poorly on offline datasets. Hence, it is critical to avoid visiting OOD states and taking OOD actions in offline RL. Several recent methods have used uncertainty estimation to distinguish OOD samples. However, errors in the uncertainty estimation make the purely uncertainty-based method unstable and require additional components to ensure sufficient pessimism . In this study, we propose a Bounded Uncertainty based Pessimistic policy iteration algorithm (BUP). The BUP pessimistically estimates the value function via bounded uncertainty, and the uncertainty bound is achieved by constraining the actor from taking highly uncertain actions. The suboptimality bound of BUP is theoretically guaranteed in linear Markov Decision Processes (MDPs), and experiments on D4RL datasets show that BUP matches the state-of-the-art performance. Moreover, BUP is simple to implement with low computational cost and does not require any additional components.

Offline reinforcement learning (RL) aims to learn policy from the passively collected offline dataset. Applying existing RL methods on the static dataset straightforwardly will raise distribution shift, causing these unconstrained RL methods to fail. To cope with the distribution shift problem, a common practice in offline RL is to constrain the policy explicitly or implicitly close to behavioral policy. However, the available dataset usually contains sub-optimal or inferior actions, constraining the policy near all these actions will make the policy inevitably learn inferior behaviors, limiting the performance of the algorithm. Based on this observation, we propose a weighted policy constraints (wPC) method that only constrains the learned policy to desirable behaviors, making room for policy improvement on other parts. Our algorithm outperforms existing state-of-the-art offline RL algorithms on the D4RL offline gym datasets. Moreover, the proposed algorithm is simple to implement with few hyper-parameters, making the proposed wPC algorithm a robust offline RL method with low computational complexity.

Digital entrepreneurial success has gained attention because the development of digitalization has brought multiple opportunities for entrepreneurs along with many challenges. While prior research has primarily examined the determinants of success at the founder and organizational levels, the impact of venture groups has been neglected, resulting in a lack of exploration of the group effect. In this study, we apply social learning theory to investigate how lead entrepreneur's digital leadership influences digital entrepreneurial success. By analyzing data from 88 digital entrepreneurial teams, we found that lead entrepreneur's digital leadership is positively related to entrepreneurial team's digital entrepreneurial success. Likewise, lead entrepreneur's digital leadership positively correlates with digital entrepreneurial success through technology absorptive capacity and technological innovation capability. Moreover, we argue that leader group prototypicality not only strengthens the impact of lead entrepreneur's digital leadership on the entrepreneurial team's technology absorptive capacity but also amplifies its influence on the entrepreneurial team's technological innovation capability. Our results provide contributions to both the entrepreneurship and leadership literature.

It is quite challenging for conventional estimators to extract the radial wind velocities from the raw spectrum data of coherent Doppler wind Lidar (CDWL) at very low signal-to-noise ratios (SNR). This work proposes a new wind velocity estimation method based on the constraints from neighboring wind profiles, which can effectively and robustly extend the reliable detection range of CDWL. Considering the spatial continuity of the wind field, priori information from the velocity probability distributions in previous range bins is utilized to reshape the contaminated spectrum of the current range bin. The influences of different preceding range bins are weighted by their overlapping ratios to the contaminated range bin. Iterations are carried out when computing the variance of the Gaussian-shaped probability, which can preserve the wind field details while removing the parameter dependence in practical applications. The method is verified theoretically in simulated atmosphere echoes as well as experimentally in our self-developed dual-frequency pulsed CDWL system. The first results show that reasonable estimates can be obtained in places far beyond the previous detection boundary provided by conventional estimators.

Compared with traditional biomagnetic field detection devices, such as superconducting quantum interference devices (SQUIDs) and atomic magnetometers, only giant magnetoimpedance (GMI) sensors can be applied for unshielded human brain biomagnetic detection, and they have the potential for application in next-generation wearable equipment for brain-computer interfaces (BCIs). Achieving a better GMI sensor without magnetic shielding requires the stimulation of the GMI effect to be maximized and environmental noise interference to be minimized. Moreover, the GMI effect stimulated in an amorphous filament is closely related to its working point, which is sensitive to both the external magnetic field and the drive current of the filament. In this paper, we propose a new noisereducing GMI gradiometer with a dual-loop self-adapting structure. Noise reduction is realized by a direction-flexible differential probe, and the dual-loop structure optimizes and stabilizes the working point by automatically controlling the external magnetic field and drive current. This dual-loop structure is fully program controlled by a micro control unit (MCU), which not only simplifies the traditional constantparameter sensor circuit, saving the time required to adjust the circuit component parameters, but also improves the sensor performance and environmental adaptation. In the performance test, within 2 min of self-adaptation, our sensor showed a better sensitivity and signal-to-noise ratio (SNR) than those of the traditional designs and achieved a background noise of 12 pT/√Hz at 10 Hz and 7pT/√Hz at 200 Hz. To the best of our knowledge, our sensor is the first to realize self-adaptation of both the external magnetic field and the drive current.