Science China Information Sciences最新文献

This paper investigates the minimum-time trajectory planning problem of an autonomous vehicle. To deal with unknown and uncertain dynamics of the vehicle, the trajectory planning problem is modeled as a Markov decision process with a continuous action space. To solve it, we propose a continuous advantage learning (CAL) algorithm based on the advantage-value equation, and adopt a stochastic policy in the form of multivariate Gaussian distribution to encourage exploration. A shared actor-critic architecture is designed to simultaneously approximate the stochastic policy and the value function, which greatly reduces the computation burden compared to general actor-critic methods. Moreover, the shared actor-critic is updated with a loss function built as mean square consistency error of the advantage-value equation, and the update step is performed several times at each time step to improve data efficiency. Simulations validate the effectiveness of the proposed CAL algorithm and its better performance than the soft actor-critic algorithm.

In reinforcement learning (RL), training a policy from scratch with online experiences can be inefficient because of the difficulties in exploration. Recently, offline RL provides a promising solution by giving an initialized offline policy, which can be refined through online interactions. However, existing approaches primarily perform offline and online learning in the same task, without considering the task generalization problem in offline-to-online adaptation. In real-world applications, it is common that we only have an offline dataset from a specific task while aiming for fast online-adaptation for several tasks. To address this problem, our work builds upon the investigation of successor representations for task generalization in online RL and extends the framework to incorporate offline-to-online learning. We demonstrate that the conventional paradigm using successor features cannot effectively utilize offline data and improve the performance for the new task by online fine-tuning. To mitigate this, we introduce a novel methodology that leverages offline data to acquire an ensemble of successor representations and subsequently constructs ensemble Q functions. This approach enables robust representation learning from datasets with different coverage and facilitates fast adaption of Q functions towards new tasks during the online fine-tuning phase. Extensive empirical evaluations provide compelling evidence showcasing the superior performance of our method in generalizing to diverse or even unseen tasks.

A novel feature extraction method that can tolerate imbalanced user data is proposed. A cost-sensitive SVM assigns different misclassification costs to faults with different severity levels to optimize the cause diagnosis process. The simulation results demonstrate the effectiveness and superiority of the proposed algorithm.

Video portrait segmentation (VPS), aiming at segmenting prominent foreground portraits from video frames, has received much attention in recent years. However, the simplicity of existing VPS datasets leads to a limitation on extensive research of the task. In this work, we propose a new intricate large-scale multi-scene video portrait segmentation dataset MVPS consisting of 101 video clips in 7 scenario categories, in which 10843 sampled frames are finely annotated at the pixel level. The dataset has diverse scenes and complicated background environments, which is the most complex dataset in VPS to our best knowledge. Through the observation of a large number of videos with portraits during dataset construction, we find that due to the joint structure of the human body, the motion of portraits is part-associated, which leads to the different parts being relatively independent in motion. That is, the motion of different parts of the portraits is imbalanced. Towards this imbalance, an intuitive and reasonable idea is that different motion states in portraits can be better exploited by decoupling the portraits into parts. To achieve this, we propose a part-decoupling network (PDNet) for VPS. Specifically, an inter-frame part-discriminated attention (IPDA) module is proposed which unsupervisedly segments portrait into parts and utilizes different attentiveness on discriminative features specified to each different part. In this way, appropriate attention can be imposed on portrait parts with imbalanced motion to extract part-discriminated correlations, so that the portraits can be segmented more accurately. Experimental results demonstrate that our method achieves leading performance with the comparison to state-of-the-art methods.

Recent advancements in attribute localization have showcased its potential in discovering the intrinsic semantic knowledge for visual feature representations, thereby facilitating significant visual-semantic interactions essential for zero-shot learning (ZSL). However, the majority of existing attribute localization methods heavily rely on classification constraints, resulting in accurate localization of only a few attributes while neglecting the rest important attributes associated with other classes. This limitation hinders the discovery of the intrinsic semantic relationships between attributes and visual features across all classes. To address this problem, we propose a novel attribute localization refinement (ALR) module designed to enhance the model’s ability to accurately localize all attributes. Essentially, we enhance weak discriminant attributes by grouping them and introduce weighted attribute regression to standardize the mapping values of semantic attributes. This module can be flexibly combined with existing attribute localization methods. Our experiments show that when combined with the ALR module, the localization errors in existing methods are corrected, and state-of-the-art classification performance is achieved.

In this study, an adaptive neural network (NN) control is proposed for nonlinear two-degree-of-freedom (2-DOF) helicopter systems considering the input constraints and global prescribed performance. First, radial basis function NN (RBFNN) is employed to estimate the unknown dynamics of the helicopter system. Second, a smooth nonaffine function is exploited to approximate and address nonlinear constraint functions. Subsequently, a new prescribed function is proposed, and an original constrained error is transformed into an equivalent unconstrained error using the error transformation and barrier function transformation methods. The analysis of the established Lyapunov function proves that the controlled system is globally uniformly bounded. Finally, the simulation and experimental results on a constructed Quanser’s test platform verify the rationality and feasibility of the proposed control.

We consider Shor’s quantum factoring algorithm in the setting of noisy quantum gates. Under a generic model of random noise for (controlled) rotation gates, we prove that the algorithm does not factor integers of the form pq when the noise exceeds a vanishingly small level in terms of n—the number of bits of the integer to be factored, where p and q are from a well-defined set of primes of positive density. We further prove that with probability 1 − o(1) over random prime pairs (p, q), Shor’s factoring algorithm does not factor numbers of the form pq, with the same level of random noise present.

We propose an intelligent control strategy that synthesizes optimal guidance and SAC, guidance and NFZs avoidance missions are realized with high precision and low energy loss. The training efficiency is enhanced by introducing a prediction method to calculate the terminal states and adding process rewards. By improving the training process, the learned strategy has a strong generalization on problems of dynamic NFZs, indicating higher applicability and flexibility in flight missions.

Research on flexible strain sensors has advanced rapidly in recent years, with particular attention being devoted to two-dimensional (2D) semiconductor materials owing to their exceptional mechanical and electrical properties that are conducive to sophisticated sensing performance. However, resistive strain sensors based on 2D semiconductor materials typically exhibit positive gauge factors (GF), while materials for strain sensors with a negative GF remain elusive. We have identified a trend of reduction in the band gap of the emerging 2D semiconductor material tellurium (Te) under strain in simulations reported in past research, and have observed a negative GF in the Te-based strain sensor. In this study, we combined Te with a flexible polyethylene terephthalate (PET) substrate to manufacture a flexible strain sensor with a significantly negative GF. The results of tests revealed that the Te-based strain sensor achieved an impressive maximum sensitivity of −139.7 within a small range of bending-induced strain (< 1%). Furthermore, it exhibited excellent linearity and good cyclic stability, and was successfully applied to monitor limb movements. The work here verifies the significant potential for the use of Te-based strain sensors in next-generation flexible electronics.

Locally repairable codes (LRCs), which can recover any symbol of a codeword by reading only a small number of other symbols, have been widely used in real-world distributed storage systems, such as Microsoft Azure Storage and Ceph Storage Cluster. Since binary linear LRCs can significantly reduce coding and decoding complexity, constructions of binary LRCs are of particular interest. The aim of this paper is to construct dimensional optimal binary LRCs with disjoint local repair groups. We introduce a method to connect intersection subspaces with binary LRCs and construct dimensional optimal binary linear LRCs with locality 2^b (b ≽ 3) and minimum distance d ≽ 6 by employing intersection subspaces deduced from the direct sum. This method will sufficiently increase the number of possible repair groups of dimensional optimal LRCs, thus efficiently expanding the range of the construction parameters while keeping the largest code rates compared with all known binary linear LRCs with minimum distance d ≽ 6 and locality 2^b.