Robotic manipulation tasks frequently share foundational structures. Meta-reinforcement learning aims to develop generalizable policies that leverage these shared structures. However, existing methods usually struggle to efficiently encode this task-specific knowledge into their policy networks: hierarchical policies depend on task-agnostic action-level skills or intrinsic rewards, while context-based paradigms suffer from Markov Decision Process ambiguity. To address these challenges, we propose a hierarchical neural skill-based meta-reinforcement learning framework. This framework includes neural skill generation, neural skill decoding, and policy network construction. A Transformer-based neural skill generation unit sequentially generates hierarchical neural skills conditioned on a task description. The decoding mechanism translates these neural skills into network parameters for a policy. Using these decoded parameters, the policy network is constructed layer by layer to process the state information. Unlike previous work, our method treats skills as abstractions of layer-wise network parameters, allowing task-specific knowledge embedded in neural skills to directly configure the policy network. Experimental results demonstrate that our method has enhanced flexibility and efficiency.

Indoor radio maps with frequency domain data are difficult to reconstruct when only limited measurements at a few locations are available. Naive convolutional neural networks suffer from flawed structures in the frequency domain when predicting these radio maps, resulting in overly smoothed predictions. We propose a Spatial Frequency Interleaving Residual Autoencoder (SFIRA) architecture to tackle this problem, along with a Procedural Radio Map Generation (PRMG) method to address the data deficiency of indoor radio maps. Experimental results indicate that the proposed architecture achieves lower Normalized Root Mean Square Error (NRMSE). Visualization of the reconstruction further suggests that the proposed methods are effective.

Task planning refers to autonomously organizing actions in response to instruction signals, especially language signals. Previous reinforcement learning and imitation learning methods always require a large amount of task-related data (data interacting with the environment or expert demonstrations) for policy training. In this paper, we propose a method named Language Signal Parse Tree (LSPT for short) for task planning. LSPT is a training-free method and consists of a language signal parser and a precondition action generator. The language signal parser translates goal predicates (neural language instructions) into a predicate parse tree. This tree contains essential information about position and state changes, which is subsequently employed to generate key actions. The precondition action generator leverages commonsense to produce additional pre-actions that complement each key action, thereby increasing the likelihood of successful execution of all actions. The final plan used to accomplish the task incorporates both key actions and pre-actions. Experiments on a virtual indoor scene show that the proposed method outperforms the state-of-the-art method LID by 24.8% on In-Distribution dataset, 43.2% on Novel Scenes dataset, and 38.8% on Novel Tasks dataset.

Mesh denoising is a fundamental task in geometry processing, and recent studies have demonstrated the remarkable superiority of deep learning-based methods in this field. However, existing works commonly rely on neural networks without explicit designs for noise and geometry which are actually fundamental factors in mesh denoising. In this paper, by jointly considering noise intensity and geometric characteristics, a novel Filtering Coefficient Learner (FCL for short) for mesh denoising is developed, which delicately generates coefficients to filter face normals. Specifically, FCL produces filtering coefficients consisting of a noise-aware component and a geometry-aware component. The first component is inversely proportional to the noise intensity of each face, resulting in smaller coefficients for faces with stronger noise. For the effective assessment of the noise intensity, a noise intensity estimation module is designed, which predicts the angle between paired noisy-clean normals based on a mean filtering angle. The second component is derived based on two types of geometric features, namely the category feature and face-wise features. The category feature provides a global description of the input patch, while the face-wise features complement the perception of local textures. Extensive experiments have validated the superior performance of FCL over SOTA works in both noise removal and feature preservation.

Given different instructions, large vision-language models (LVLMs) exhibit different degrees of object hallucinations, posing a significant challenge to the evaluation of object hallucinations. Overcoming this challenge, existing object hallucination evaluation methods average the results obtained from a set of instructions. However, these methods fail to provide consistent evaluation across instruction sets that generate image descriptions of significantly different lengths. In this paper, we present the first systematic investigation of the effect of instructions on object hallucinations in LVLMs, with a specific focus on the role played by image description lengths. A valuable finding is that instructions indirectly affect hallucinations through the length of image descriptions. The longer the image description, the higher the object hallucination degree. Accordingly, we fit an informative length-hallucination curve, upon which a fine-grained evaluation framework named LeHaCE is introduced for evaluating object hallucinations at any given image description length. LeHaCE evaluates the object hallucination degree at a uniform image description length to mitigate the effect of description lengths, promoting stability and fairness. Moreover, LeHaCE incorporates the curve slope as an innovative hallucination evaluation metric, reflecting the extent to which the object hallucination degree is affected by the image description length, achieving a more comprehensive evaluation. Experimental results demonstrate that LeHaCE provides a more stable, fair, and comprehensive evaluation of object hallucinations in LVLMs compared to existing methods.

In response to the escalating data volumes and the pressing need for reduced network latency in Mobile Edge Computing (MEC), this paper delves into the sphere of edge computing. Many MEC platforms are turning to container-based virtualization and leveraging image layering to cut down transmission costs. Amid this shift, applications are growing more complex, composed of multiple tasks and demanding diverse execution environments. However, existing research has mainly concentrated on task scheduling within edge systems, sidelining the vital aspect of preparing runtime environments on MEC servers. To bridge this gap, our paper tackles task scheduling complexities, including data and layer dependencies. It offers an integrated solution that optimizes task scheduling and layer loading across MEC servers, all with the goal of minimizing the total computation completion time. To address this NP-hard problem, we present a heuristic task-scheduling algorithm rooted in the genetic algorithm (GA) and complement it with an in-depth exploration of layer-loading policies. Our experiments conclusively demonstrate the substantial reduction in total computation completion time achievable through this approach.

The rapid expansion of Mobile Edge Computing (MEC), driven by the escalating data volume and the demand for minimal network latency, underscores the need for efficient data processing. To address the growing complexity of neural networks and applications, segmentation into smaller components (e.g., neural network layers, subnetworks, and subtasks) for parallel computation across diverse nodes is common. However, effective data transmission between these segments necessitates optimized task scheduling among edge servers. Many platforms leverage container-based OS-level virtualization to enhance edge computing efficiency, leveraging container image layers to cut storage and transmission costs. However, previous research predominantly emphasizes task scheduling, overlooking runtime environment preparation on edge servers and potential collaboration among edge nodes. This paper introduces an innovative approach that adeptly manages task data and image layer dependencies collaboratively. It formulates an NP-hard problem: minimizing total computation completion time by jointly determining downlink transmission rate allocation, task-offloading strategies, and layer-loading schemes, allowing for thoughtful decoupling and iterative refinement. The Gray Wolf Optimizer and Cellular Automata are introduced for dynamic task scheduling, complemented by a low-complexity algorithm inspired by the Nawas-Enscore-Ham method. For layer downloading, the paper explores a partial-layer loading policy, considering storage constraints, and establishes a full-layer loading strategy with the Peer-to-Peer mechanism, significantly reducing computational complexity. Rigorous experimental results underscore the remarkable efficacy of these approaches in curtailing total computation completion time, positioning them as benchmarks for comparison against alternative solutions.

Due to high flexibility and ease of deployment, Unmanned Aerial Vehicle (UAV)-enabled mobile edge computing (MEC) has recently emerged to provide services for users to meet the demands of computing-intensive tasks at edge. However, the MEC-server mounted UAV is inappropriate for heavy-computation tasks owing to the limitation of energy supply and hardware cost, which may make for excessively high CPU temperature. To tackle this issue, this paper considers a UAV-and-basestation (BS) hybrid-assisted MEC network, where a hover-fly-hover mode is adopted to facilitate the provisioning of MEC services with the help of BS. Furthermore, a temperature control constraint is introduced to ensure the reliability of CPU at UAV. We aim to minimize the cruise duration of UAV with thermal-aware constraint by jointly optimizing UAV hovering trajectory, task scheduling strategy, and computation-and-communication resource allocation strategy. Although the formulated problem is non-convex, we decouple it into three subproblems and solve them in an iterative manner. Simulation results demonstrate that the proposed algorithm can not only satisfy the CPU temperature constraint but also help save the cruise duration.

Inspired by the behavior law of human social groups, a new swarm intelligence algorithm named the dual-population social group optimization (DPSGO) algorithm is proposed in this article. Based on the primitive social group optimization (SGO) algorithm, dual-population grouping technology, reverse learning technology, immigration migration technology, and Gaussian mutation are introduced to further simulate the behavior law of actual human social groups. Experimental results and performance comparison show that the DPSGO algorithm has a better searchability and convergence rate. In addition, aiming at the socially hot issue of aviation safety, the simulation and experimental results show that the temperature measurement error can be reduced to less than 7.5 °C by using the DPSGO algorithm combined with reflected radiation correction to process the aeroengine multispectral radiation temperature measurement data. This article is of great significance to the design and optimization of swarm intelligence algorithms by using the behavior law of human social groups and provides valuable guidance for enhancing the safety monitoring of aeroengines.

Visible surface area estimation for irregular objects, one of the most fundamental and challenging topics in mathematics, supports a wide range of applications. The existing techniques usually estimate the visible surface area via mathematical modeling from 3D point clouds. However, the 3D scanner is expensive, and the corresponding evaluation method is too complex. In this paper, we propose a novel problem setting, deep learning for visible surface area estimation, which is the first trial to estimate the visible surface area for irregular objects from monocular images. Technically, we first build a novel visible surface area estimation dataset including 9099 real annotations. Then, we design a learning-based architecture to predict the visible surface area, including two core modules (i.e., the classification module and the area-bins module). The classification module is presented to predict the visible surface area distribution interval and assist network training for more accurate visible surface area estimation. Meanwhile, the area-bins module using the transformer encoder is proposed to distinguish the difference in visible surface area between irregular objects of the same category. The experimental results demonstrate that our approach can effectively estimate the visible surface area for irregular objects with various categories and sizes. We hope that this work will attract further research into this newly identified, yet crucial research direction. Our source code and data are available at \textcolormagenta \urlhttps://github.com/liuxu0303/VSAnet .