Artificial Intelligence Review最新文献

Reinforcement learning (RL), as an emerging interdisciplinary field formed by the integration of artificial intelligence and control science, is currently demonstrating a cross-disciplinary development trend led by artificial intelligence and has become a research hotspot in the field of optimal control. This paper systematically reviews the development context of RL, focusing on the intrinsic connection between single-agent reinforcement learning (SARL) and multi-agent reinforcement learning (MARL). Firstly, starting from the formation and development of RL, it elaborates on the similarities and differences between RL and other learning paradigms in machine learning, and briefly introduces the main branches of current RL. Then, with the basic knowledge and core ideas of SARL as the basic framework, and expanding to multi-agent system (MAS) collaborative control, it explores the coherence characteristics of the two in theoretical frameworks and algorithm design. On this basis, this paper reconfigures SARL algorithms into dynamic programming, value function decomposition and policy gradient (PG) type, and abstracts MARL algorithms into four paradigms: behavior analysis, centralized learning, communication learning and collaborative learning, thus establishing an algorithm mapping relationship from single-agent to multi-agent scenarios. This innovative framework provides a new perspective for understanding the evolutionary correlation of the two methods, and also discusses the challenges and solution ideas of MARL in solving large-scale MAS problems. This paper aims to provide a reference for researchers in this field, and to promote the development of cooperative control and optimization methods for MAS as well as the advancement of related application research.

Wetlands are critical ecosystems supporting biodiversity and providing essential environmental services. With increasing threats to wetlands, efficient classification techniques are essential for effective conservation. Several researchers have contributed to wetland classification across reputable journals. However, some challenges (data scarcity, noisy labels, and model generalisability) still exist. Therefore, this paper presents a comprehensive survey of recent advancements in machine learning (ML) and deep learning (DL) for wetland classification, focusing on developments from 2018 to 2025. Key methodologies, including convolutional neural networks, transformers, and generative adversarial networks, are critically reviewed, highlighting their strengths, limitations, and applications in remote sensing. Unlike previous reviews, this work emphasises underexplored techniques such as few-shot learning and Mamba networks, offering practical recommendations for handling limited training data and improving model generalisability. The study also identifies promising research directions, such as test-time training and hybrid loss functions, to address challenges in wetland classification. This survey aims to guide researchers and practitioners in advancing state-of-the-art wetland classification through ML and DL technologies.

Effective communication is essential for human touch, and it’s especially crucial for those who rely on sign language because they have speech or hearing difficulties. Communication between deaf and normal people is restricted by the inability of current technology to automatically create flexible Bangla Sign Language (BdSL) animations from Bangla text or voice. This work introduces a revolutionary computer graphics-based system that takes voice or text input in Bangla and uses it to create BdSL animations. The system uses a HamNoSys to Gesture Markup Language (SiGML) conversion to dynamically translate input into gestures. With the use of 94 classes in the dataset that cover Bangla numbers, alphabets, and word motions, the model can generate any word or phrase in Bangla. Additionally, the system creates gestures for inputs that aren’t in the dataset by spelling out the letters. The system converts Bangla text into three-dimensional animated BdSL movements by parsing it based on Linguistic principles. According to performance evaluations, each input has a processing cost of 79.57 ms, and the average accuracy for text and voice input is 97.50% and 94.75%, respectively. The method ensures fluency and naturalness by taking into consideration crucial elements of sign language, such as hand shape, palm orientation, and non-manual messages. By reducing communication barriers between the sign and non-sign populations, this study significantly advances accessibility. Visit this GitHub repository link for further details on the implementation of the suggested system and the SiGML dataset: https://gitlab.com/devarifkhan/bdsl-3d-animation

The decline of bee and wind-based pollination systems in greenhouses due to controlled environments and limited access has boosted the importance of finding alternative pollination methods. Robot-based pollination systems have emerged as a promising solution, ensuring adequate crop yield even in challenging pollination scenarios. This paper presents a comprehensive review of the current robotic-based pollinators employed in agriculture. The review categorizes pollinator technologies into major categories such as air-jet, water-jet, linear actuator, ultrasonic wave, and air-liquid spray, each suitable for specific crop pollination requirements. However, these technologies are often tailored to particular crops, limiting their versatility. The advancement of science and technology has led to the integration of automated pollination technology, encompassing information technology, automatic perception, detection, control, and operation. This integration not only reduces labor shortage problem but also fosters the ongoing progress of modern agriculture by refining technology, enhancing automation, and promoting intelligence in agricultural practices. Finally, the challenges encountered in the design of robot-based pollinators are addressed, and a forward-looking perspective is taken towards future developments, aiming to contribute to the sustainable advancement of this technology.

Transfer Learning (TL) has gained significant traction in machine learning, especially in deep learning contexts. However, its integration with Nature-Inspired Algorithms (NIAs) remains fragmented, with limited understanding of strategies, challenges, and outcomes. This paper presents the first systematic review focused exclusively on the use of TL in NIAs, excluding deep learning approaches. Major challenges include dealing with domain/task similarity, avoiding negative transfer, selecting what and when to transfer, and adapting TL mechanisms to population-based search paradigms. To address these issues, we conducted a structured analysis of 47 primary studies, categorizing them by TL strategies, learning paradigms, and algorithmic goals. Our findings reveal recurring patterns, highlight open research gaps, and propose future directions for developing robust TL-based NIAs. This review provides a foundation for researchers interested in designing adaptive, efficient, and knowledge-guided metaheuristics for complex optimization tasks.

The challenge of unanswerable questions in Machine Reading Comprehension (MRC) has drawn considerable attention, as current MRC systems are typically designed under the assumption that every question has a valid answer within the provided context. However, these systems often encounter real-world situations where no valid answer is available. This paper provides a comprehensive review of existing methods for addressing unanswerable questions in MRC systems, categorizing them into model-agnostic and model-specific approaches. It explores key strategies, examines relevant datasets, and evaluates commonly used metrics. This work aims to provide a comprehensive understanding of current techniques and identify critical gaps in the field, offering insights and key challenges to direct future research toward developing more robust MRC systems capable of handling unanswerable questions.

We develop a deep reinforcement learning framework for tactical decision making in an autonomous truck, specifically for Adaptive Cruise Control (ACC) and lane change maneuvers in a highway scenario. Our results demonstrate that it is beneficial to separate high-level decision-making processes and low-level control actions between the reinforcement learning agent and the low-level controllers based on physical models. In the following, we study optimizing the performance with a realistic and multi-objective reward function based on Total Cost of Operation (TCOP) of the truck using different approaches; by adding weights to reward components, by normalizing the reward components and by using curriculum learning techniques.

This survey explores the evolution and application of U-Net in generative AI, highlighting its success across various modalities, including image, text, audio, video, 3D, and pose/action generation. Initially designed for biomedical segmentation, U-Net has been adapted and enhanced with architectural innovations such as normalization techniques, self and cross-attention mechanisms, and residual connections. These advancements have made U-Net a powerful backbone for modern generative models in diffusion-based frameworks, GANs, and autoregressive architectures. The survey comprehensively reviews U-Net’s modality-specific applications, from high-resolution image synthesis and text-to-image generation to speech enhancement, video generation, 3D reconstruction, and pose/action generation. Despite its widespread success, U-Net faces challenges in computational efficiency, contextual understanding, and scalability for multimodal tasks. Future directions focus on optimizing U-Net for lightweight and real-time applications, enhancing its contextual awareness, and improving its integration with emerging architectures like transformers and diffusion models.

This paper explores advancements in Natural Language Processing (NLP) for legal lay summarization by systematically analyzing existing methodologies, datasets, and research findings. We review current literature, highlighting key challenges such as data scarcity and the complexity of legal language. A primary contribution of this study is the development of LegalEase, a specialized dataset designed to improve model training for summarizing legal documents in layman’s terms. Our findings demonstrate that subdomain-specific datasets within the legal domain outperform general legal datasets in enhancing NLP model performance for generating accurate and comprehensible legal summaries. The insights and methodologies presented provide a foundation for future research in legal lay summarization.

The cold email industry faces significant challenges in ensuring successful message delivery, with SMTP errors and bounces being common occurrences. Predicting these errors can help optimize email delivery and improve senders’ reputation. In this study, the abilities of Gated Recurrent Units (GRU), Convolutional Neural Network (CNN), and Temporal Convolutional Network (TCN) were examined to predict SMTP errors and bounces in the context of a cold email dataset. The study reveals that both the GRU networks and the CNNs have achieved over 70% accuracy in predicting SMTP errors and bounces. Exploiting all three architectures, GRU, MLP, and TCN, can substantially enhance the management and optimization of the cold email sending process, leading to improved email delivery rates and better sender reputation.