Aiming at the challenges in planning for the synchronized execution of multiple geometrically heterogeneous tasks by swarms in uncertain environments, this work proposes a distributed task planning method. This method is intended to overcome the inefficiencies in traditional methods caused by fragmented task modeling, high sensitivity to environmental uncertainty, and the risk of collaborative deadlocks. First, a Conditional Value-at-Risk (CVaR) model is introduced to dynamically estimate task execution time under uncertainty. Second, multi-geometry tasks (area, line, point) are uniformly transformed into equivalent point-task sets via grid-based decomposition and key-point discretization. Finally, a cooperative task dependency graph is embedded into a distributed auction consensus mechanism with cycle detection to prevent deadlocks, and an adaptive cooperative obstacle avoidance strategy based on task progress is integrated to ensure synchronized and collision-free motion. Simulation results demonstrate our method’s superiority over benchmarks in completion time, success rate, and efficiency. Its practicality is validated through real-world UAV tests, while scalability experiments confirm its robustness in large-scale scenarios. This work provides a solid theoretical foundation and technical support for efficient coordination of unmanned swarms in complex and dynamic environments.

In order to address the high complexity and low efficiency of amphibious propulsion systems, this paper proposes a novel variable wheel-propeller integrated mechanism for amphibious robots. By adjusting the blade pitch angle, it enables multiple motion modes, including rapid and stable movement on flat ground, obstacle crossing, and omnidirectional movement on water surface. This study establishes a kinematic model for the propeller blades and conducts multi-objective optimization of the structural parameters by considering both the land obstacle-crossing performance and underwater propulsion performance. Based on the optimized structural parameters, a virtual simulation prototype is constructed. Simulation results indicate that when water surface movement, with a driving torque of 3N.m, robot achieves a maximum linear velocity of 1.25m/s and a maximum angular self-rotation velocity of 3.5rad/s. Moreover, varying the blade pitch angle can alter the thrust direction, enabling omnidirectional mobility on water surface. During land movement, with a rotation speed of 60rpm, the highest obstacle-crossing height is 184mm. This wheel-propeller integrated mechanism exhibits robust comprehensive motion performance and environmental adaptability, with convenient motion modes switching.

Addressing intermittent client availability is critical for the real-world deployment of federated learning algorithms. Most prior work either overlooks the potential non-stationarity in the dynamics of client unavailability or requires substantial memory/computation overhead. We study federated learning in the presence of heterogeneous and non-stationary client availability, which may occur when the deployment environments are uncertain, or the clients are mobile. The impacts of heterogeneity and non-stationarity on client unavailability can be significant, as we illustrate using FedAvg, the most widely adopted federated learning algorithm. We propose FedAWE, which includes novel algorithmic structures that (i) compensate for missed computations due to unavailability with only $O(1)$ additional memory and computation with respect to standard FedAvg, and (ii) evenly diffuse local updates within the federated learning system through implicit gossiping, despite being agnostic to non-stationary dynamics. We show that FedAWE converges to a stationary point of even non-convex objectives while achieving the desired linear speedup property. We corroborate our analysis with numerical experiments over diversified client unavailability dynamics on real-world data sets.

It is an open question what semantic representations transformer-based language models can encode and whether they have access to more abstract aspects of semantic meaning. Here, we propose a diagnostic dataset to investigate how well language models understand the degree semantics of adjectives. In the dataset, referred as the Adjective Scale Probe (ASP), we semi-automatically generate 8 tests of Natural Language Inference (NLI) questions to test 8 key capabilities of adjective interpretation. We apply the ASP dataset to evaluate the performance of 3 language models, i.e., BERT, DeBERTa, and T0. It is found that language models perform below the majority baseline for most tests of the ASP, even when the models have been fine-tuned to achieve high performance on the large-scale MNLI dataset. But after we fine-tune the pre-trained models on a subset of the ASP, DeBERTa can achieve high performance on the untrained adjectives and untrained tests, suggesting that DeBERTa may have captured degree semantic information of adjectives through pre-training but it needs specific training data to learn how to apply such information to the current tasks. In sum, the ASP provides an easy-to-use method to test fine-grained formal semantic properties of adjectives, and reveals language models' abilities to access formal semantic information.

This article develops a general yet effective hyperspectral image (HSI) classification framework that is trained with small data. To this end, two identically structured but differently initialized classifiers, which are referred to as two base classifiers, are trained in an iterative manner. Each iteration consists of three steps, that is: 1) the two base classifiers that are trained separately on guide data; 2) unclassified data that are processed by the two trained base classifiers; and 3) the classification results with high confidence that are explored as new guide data. In the first iteration, the guide data comprising the original small training data are the same for the two base classifiers. From the second iteration, the guide data for the two base classifiers start becoming different. Specifically, in each iteration, the guide data for training one base classifier keep being augmented by high confidence classification results provided by the other base classifier. It is in such an iterative manner that the two classifiers continuously provide different new guide data for each other, and thus increasingly augment labeled data from the original small training set to a reasonably larger amount of samples in a HSI. We refer to such a training strategy as mutual guide. We develop a mutual guide implementation scheme by exploiting extreme learning machines (ELMs) as base classifiers. Extensive experiments on four public HSI datasets, i.e., Indian Pines (IP), Kennedy Space Center (KSC), University of Pavia (UP), and Salinas (SA), validate the classification effectiveness of our mutual guide framework with small training data.

In order to extend the unsupervised nonlinear dimensionality reduction method Isomap for use in supervised learning, a new supervised manifold learning method namely discriminant Isomap (D-Isomap) is proposed, in which the geometrical structure of each class data is preserved by keeping geodesic distances between data points of the same class and the discriminant capacity is enhanced by maximizing the distances between data points of different classes. A new objective function is defined for this purpose and the corresponding optimization problem is solved by using the SMACOF algorithm. The effectiveness of D-Isomap is examined by extensive simulations on artificial and real-world data sets, including MNIST, USPS, and UCI. In both visualization and classification experiments, D-Isomap achieves comparable or better performance than the widely used dimensionality reduction algorithms.

In this paper, we consider a parallel distributed detection network consisting of a fusion center and N sensors. We assume that the observations at different sensors are conditionally dependent, and optimize the system performance under the Neyman- Pearson criterion. Unlike previous papers dealing with the optimal N-P detection problem, we allow the sensor decision rules to be randomized, and obtain the necessary conditions for optimal fusion rule and sensor decision rules without making any assumptions on the joint density functions of sensor observations. The optimality conditions are obtained using an important property of points on the overall ROC curve that is established in the paper. And, a sufficient condition that guarantees the optimal sensor decision rules to be deterministic is also presented.

In this paper, the problem of distributed detection under Neyman-Pearson criterion is considered. We assume that the observations of different sensors are conditionally dependent. First, an important property of the overall ROCs is investigated. Based on this property, necessary conditions for optimal fusion rule and sensor decision rules are then obtained. In the derivation of our optimality conditions, no assumption regarding the convexity of the overall ROC is assumed. Instead, we assume the differentiability of the overall ROCs. The method used here is straightforward, and the result obtained is clear and simple. Some relations between our results and the Lagrange method exist, and the implication of our results to the validity of Lagrange method is also investigated

In this paper, we consider a serial distributed detection network consisting of N sensors. We assume that the observations at different sensors are conditionally dependent, and optimize the system performance under Bayesian criterion. The necessary conditions for optimal sensor decision rules are obtained, and from which it is seen that the optimal sensor decision rules are coupled and are generally not likelihood ratio tests. In cases of independent sensor observations, the optimal sensor decision rules can be reduced to likehood ratio tests, and in such cases the problem of finding optimal detection scheme reduces to the problem of finding optimal threshold values of the sensors

The purpose of decision fusion in a distributed detection system is to achieve a performance that is better than that of local detectors (or sensors). We consider a distributed Bayesian detection system consisting of n sensors and a fusion center, in which the decision rules of the sensors have been given and the decisions of different sensors are conditionally independent. We assume that the decision rules of the sensors can be optimum or suboptimum, and that the probabilities of detection and false alarm of the sensors can be different. Theoretical analysis on the performance of this fusion system is carried out. Conditions for the fusion system to achieve a global risk that is smaller than local risks are obtained.

The performance of a distributed Neyman-Pearson detection system is considered. We assume that the decision rules of the sensors are given and that decisions from different sensors are mutually independent conditioned on both hypotheses. The purpose of decision fusion is to improve the performance of the overall system, and we are interested to know under what conditions can a better performance be achieved at fusion center, and under what conditions cannot. We assume that the probabilities of detection and false alarm of the sensors can be different. By comparing the probability of detection at fusion center with that of each of the sensors, with the probability of false alarm at fusion center constrained equal to that of the sensor, we give conditions for a better performance to be achieved at fusion center.