Representational Capacity Research Articles

Enhanced marine fish small sample image recognition with RVFL in Faster R-CNN model

Addressing the challenge of insufficient underwater marine life images leading to diminished accuracy in existing target recognition models, this paper introduces Random Vector Functional Link (RVFL), a suitable model for small sample training, to reduce the dependency of traditional visual recognition models on data volume by replacing the Softmax classifier in the Faster Region-Based Convolutional Neural Network (R-CNN) model. The RVFL network assigns random weights to enhance the model's adaptability to diverse data distributions, thereby improving its generalization performance. Furthermore, the RVFL model incorporates multiple nonlinear activation functions, thereby enhancing the neural network's representational capacity for high-dimensional data. The application of the proposed model to the task of marine fish small sample target recognition demonstrates its effectiveness in mitigating the low accuracy issue caused by data scarcity. It significantly improved recognition accuracy and offered new insights into addressing underwater small sample image recognition challenges.

Quaternion Convolutional Neural Networks: Current Advances and Future Directions

Since their first applications, Convolutional Neural Networks (CNNs) have solved problems that have advanced the state-of-the-art in several domains. CNNs represent information using real numbers. Despite encouraging results, theoretical analysis shows that representations such as hyper-complex numbers can achieve richer representational capacities than real numbers, and that Hamilton products can capture intrinsic interchannel relationships. Moreover, in the last few years, experimental research has shown that Quaternion-valued CNNs (QCNNs) can achieve similar performance with fewer parameters than their real-valued counterparts. This paper condenses research in the development of QCNNs from its very beginnings. We propose a conceptual organization of current trends and analyze the main building blocks used in the design of QCNN models. Based on this conceptual organization, we propose future directions of research.

DFS-DETR: Detailed-Feature-Sensitive Detector for Small Object Detection in Aerial Images Using Transformer

Object detection in aerial images plays a crucial role across diverse domains such as agriculture, environmental monitoring, and security. Aerial images present several challenges, including dense small objects, intricate backgrounds, and occlusions, necessitating robust detection algorithms. This paper addresses the critical need for accurate and efficient object detection in aerial images using a Transformer-based approach enhanced with specialized methodologies, termed DFS-DETR. The core framework leverages RT-DETR-R18, integrating the Cross Stage Partial Reparam Dilation-wise Residual Module (CSP-RDRM) to optimize feature extraction. Additionally, the introduction of the Detail-Sensitive Pyramid Network (DSPN) enhances sensitivity to local features, complemented by the Dynamic Scale Sequence Feature-Fusion Module (DSSFFM) for comprehensive multi-scale information integration. Moreover, Multi-Attention Add (MAA) is utilized to refine feature processing, which enhances the model’s capacity for understanding and representation by integrating various attention mechanisms. To improve bounding box regression, the model employs MPDIoU with normalized Wasserstein distance, which accelerates convergence. Evaluation across the VisDrone2019, AI-TOD, and NWPU VHR-10 datasets demonstrates significant improvements in the mean average precision (mAP) values: 24.1%, 24.0%, and 65.0%, respectively, surpassing RT-DETR-R18 by 2.3%, 4.8%, and 7.0%, respectively. Furthermore, the proposed method achieves real-time inference speeds. This approach can be deployed on drones to perform real-time ground detection.

Small Object Detection in UAV Images Based on YOLOv8n

With the rapid development of unmanned aerial vehicle (UAV) technology, there is an urgent need for high-performance aerial object detection algorithms that are tailored for deployment on drones with limited computing capabilities. This paper proposes a series of improvements to the state-of-the-art YOLOv8 object detector to enhance its detection accuracy and speed for small, partially occluded objects in complex environments. Specifically, we introduce multi-scale feature fusion through additional detection layers, employ conditionally parameterized convolutions to increase representational capacity, and import a dynamic non-monotonic loss function named Wise-IoU to enable more effective regression of bounding boxes. Experiments conducted on the large-scale UAV benchmark dataset VisDrone demonstrate that the improved model achieves state-of-the-art accuracy of 37.6% mAP with 3 M parameters, outperforming other lightweight YOLO detectors and two-stage detectors like Faster R-CNN. The improved model also reaches 40 FPS on an embedded edge device, validating its efficiency and suitability for real-time UAV applications. Through comprehensive quantitative experiments and visual results, this work provides valuable insights and techniques to tailor object detection algorithms for robust and efficient deployment on UAVs with limited onboard computing power.

TFMKC: Tuning-Free Multiple Kernel Clustering Coupled With Diverse Partition Fusion.

Clustering is a popular research pipeline in unsupervised learning to find potential groupings. As a representative paradigm in multiple kernel clustering (MKC), late fusion-based models learn a consistent partition across multiple base kernels. Despite their promising performance, a common concern is the limited representation capacity caused by the inflexible fusion mechanism. Concretely, the representations are constrained by truncated- k Eigen-decomposition (EVD) without fully exploiting potential information. An intuitive idea to alleviate this concern is to generate a set of augmented partitions and then select the optimal partition by fine-tuning. However, this is overlimited by: 1) introducing undesired hyperparameters and dataset-related consequences; 2) neglecting rich information across diverse partitions; and 3) expensive parameter-tuning costs. To address these problems, we propose transforming the challenging problem of directly determining the optimal partition (optimal parameter) into a diverse partition fusion (parameter ensemble) problem. We design a novel flexible fusion mechanism called tuning-free multiple kernel clustering coupled with diverse partition fusion (TFMKC) by reweighting diverse partitions through optimization, achieving an optimal consensus partition by integrating diverse and complementary information rather than traditional fine-tuning, and distinguishing our work from existing methods. Extensive experiments verify that TFMKC achieves competitive effectiveness and efficiency over comparison baselines. The code can be accessed at https://github.com/ZJP/TFMKC.

Brain-consistent architecture for imagination.

Imagination represents a pivotal capability of human intelligence. To develop human-like artificial intelligence, uncovering the computational architecture pertinent to imaginative capabilities through reverse engineering the brain's computational functions is essential. The existing Structure-Constrained Interface Decomposition (SCID) method, leverages the anatomical structure of the brain to extract computational architecture. However, its efficacy is limited to narrow brain regions, making it unsuitable for realizing the function of imagination, which involves diverse brain areas such as the neocortex, basal ganglia, thalamus, and hippocampus. In this study, we proposed the Function-Oriented SCID method, an advancement over the existing SCID method, comprising four steps designed for reverse engineering broader brain areas. This method was applied to the brain's imaginative capabilities to design a hypothetical computational architecture. The implementation began with defining the human imaginative ability that we aspire to simulate. Subsequently, six critical requirements necessary for actualizing the defined imagination were identified. Constraints were established considering the unique representational capacity and the singularity of the neocortex's modes, a distributed memory structure responsible for executing imaginative functions. In line with these constraints, we developed five distinct functions to fulfill the requirements. We allocated specific components for each function, followed by an architectural proposal aligning each component with a corresponding brain organ. In the proposed architecture, the distributed memory component, associated with the neocortex, realizes the representation and execution function; the imaginary zone maker component, associated with the claustrum, accomplishes the dynamic-zone partitioning function; the routing conductor component, linked with the complex of thalamus and basal ganglia, performs the manipulation function; the mode memory component, related to the specific agranular neocortical area executes the mode maintenance function; and the recorder component, affiliated with the hippocampal formation, handles the history management function. Thus, we have provided a fundamental cognitive architecture of the brain that comprehensively covers the brain's imaginative capacities.

Indoor Scene Reconstruction with Fine-Grained Details Using Hybrid Representation and Normal Prior Enhancement.

The reconstruction of indoor scenes from multi-view RGB images is challenging due to the coexistence of flat and texture-less regions alongside delicate and fine-grained regions. Recent methods leverage neural radiance fields aided by predicted surface normal priors to recover the scene geometry. These methods excel in producing complete and smooth results for floor and wall areas. However, they struggle to capture complex surfaces with high-frequency structures due to the inadequate neural representation and the inaccurately predicted normal priors. This work aims to reconstruct high-fidelity surfaces with fine-grained details by addressing the above limitations. To improve the capacity of the implicit representation, we propose a hybrid architecture to represent low-frequency and high-frequency regions separately. To enhance the normal priors, we introduce a simple yet effective image sharpening and denoising technique, coupled with a network that estimates the pixel-wise uncertainty of the predicted surface normal vectors. Identifying such uncertainty can prevent our model from being misled by unreliable surface normal supervisions that hinder the accurate reconstruction of intricate geometries. Experiments on the benchmark datasets show that our method outperforms existing methods in terms of reconstruction quality. Furthermore, the proposed method also generalizes well to real-world indoor scenarios captured by our hand-held mobile phones. Our code is publicly available at: https://github.com/yec22/Fine-Grained-Indoor-Recon.

A streamlined framework for BEV-based 3D object detection with prior masking

In the field of autonomous driving, perception tasks based on Bird's-Eye-View (BEV) have attracted considerable research attention due to their numerous benefits. Despite recent advancements in performance, efficiency remains a challenge for real-world implementation. In this study, we propose an efficient and effective framework that constructs a spatio-temporal BEV feature from multi-camera inputs and leverages it for 3D object detection. Specifically, the success of our network is primarily attributed to the design of the lifting strategy and a tailored BEV encoder. The lifting strategy is tasked with the conversion of 2D features into 3D representations. In the absence of depth information in the images, we innovatively introduce a prior mask for the BEV feature, which can assess the significance of the feature along the camera ray at a low cost. Moreover, we design a lightweight BEV encoder, which significantly boosts the capacity of this physical-interpretation representation. In the encoder, we investigate the spatial relationships of the BEV feature and retain rich residual information from upstream. To further enhance performance, we establish a 2D object detection auxiliary head to delve into insights offered by 2D object detection and leverage the 4D information to explore the cues within the sequence. Benefiting from all these designs, our network can capture abundant semantic information from 3D scenes and strikes a balanced trade-off between efficiency and performance.

DEAF-Net: Detail-Enhanced Attention Feature Fusion Network for Retinal Vessel Segmentation.

Retinal vessel segmentation is crucial for the diagnosis of ophthalmic and cardiovascular diseases. However, retinal vessels are densely and irregularly distributed, with many capillaries blending into the background, and exhibit low contrast. Moreover, the encoder-decoder-based network for retinal vessel segmentation suffers from irreversible loss of detailed features due to multiple encoding and decoding, leading to incorrect segmentation of the vessels. Meanwhile, the single-dimensional attention mechanisms possess limitations, neglecting the importance of multidimensional features. To solve these issues, in this paper, we propose a detail-enhanced attention feature fusion network (DEAF-Net) for retinal vessel segmentation. First, the detail-enhanced residual block (DERB) module is proposed to strengthen the capacity for detailed representation, ensuring that intricate features are efficiently maintained during the segmentation of delicate vessels. Second, the multidimensional collaborative attention encoder (MCAE) module is proposed to optimize the extraction of multidimensional information. Then, the dynamic decoder (DYD) module is introduced to preserve spatial information during the decoding process and reduce the information loss caused by upsampling operations. Finally, the proposed detail-enhanced feature fusion (DEFF) module composed of DERB, MCAE and DYD modules fuses feature maps from both encoding and decoding and achieves effective aggregation of multi-scale contextual information. The experiments conducted on the datasets of DRIVE, CHASEDB1, and STARE, achieving Sen of 0.8305, 0.8784, and 0.8654, and AUC of 0.9886, 0.9913, and 0.9911 on DRIVE, CHASEDB1, and STARE, respectively, demonstrate the performance of our proposed network, particularly in the segmentation of fine retinal vessels.

Probing the capacity of a spatiotemporal deep learning model for short-term PM2.5 forecasts in a coastal urban area

Accurate forecast of fine particulate matter (PM2.5) is crucial for city air pollution control, yet remains challenging due to the complex urban atmospheric chemical and physical processes. Recently deep learning has been routinely applied for better urban PM2.5 forecasts. However, their capacity to represent the spatiotemporal urban atmospheric processes remains underexplored, especially compared with traditional approaches such as chemistry-transport models (CTMs) and shallow statistical methods other than deep learning. Here we probe such urban-scale representation capacity of a spatiotemporal deep learning (STDL) model for 24-hour short-term PM2.5 forecasts at six urban stations in Rizhao, a coastal city in China. Compared with two operational CTMs and three statistical models, the STDL model shows its superiority with improvements in all five evaluation metrics, notably in root mean square error (RMSE) for forecasts at lead times within 12 h with reductions of 49.8 % and 47.8 % respectively. This demonstrates the STDL model's capacity to represent nonlinear small-scale phenomena such as street-level emissions and urban meteorology that are in general not well represented in either CTMs or shallow statistical models. This gain of small-scale representation in forecast performance decreases at increasing lead times, leading to similar RMSEs to the statistical methods (linear shallow representations) at about 12 h and to the CTMs (mesoscale representations) at 24 h. The STDL model performs especially well in winter, when complex urban physical and chemical processes dominate the frequent severe air pollution, and in moisture conditions fostering hygroscopic growth of particles. The DL-based PM2.5 forecasts align with observed trends under various humidity and wind conditions. Such investigation into the potential and limitations of deep learning representation for urban PM2.5 forecasting could hopefully inspire further fusion of distinct representations from CTMs and deep networks to break the conventional limits of short-term PM2.5 forecasts.

Dual Contrastive Learning Network for Graph Clustering.

Graph representation is an important part of graph clustering. Recently, contrastive learning, which maximizes the mutual information between augmented graph views that share the same semantics, has become a popular and powerful paradigm for graph representation. However, in the process of patch contrasting, existing literature tends to learn all features into similar variables, i.e., representation collapse, leading to less discriminative graph representations. To tackle this problem, we propose a novel self-supervised learning method called dual contrastive learning network (DCLN), which aims to reduce the redundant information of learned latent variables in a dual manner. Specifically, the dual curriculum contrastive module (DCCM) is proposed, which approximates the node similarity matrix and feature similarity matrix to a high-order adjacency matrix and an identity matrix, respectively. By doing this, the informative information in high-order neighbors could be well collected and preserved while the irrelevant redundant features among representations could be eliminated, hence improving the discriminative capacity of the graph representation. Moreover, to alleviate the problem of sample imbalance during the contrastive process, we design a curriculum learning strategy, which enables the network to simultaneously learn reliable information from two levels. Extensive experiments on six benchmark datasets have demonstrated the effectiveness and superiority of the proposed algorithm compared with state-of-the-art methods.

Spatial memory and frames of reference: How deeply do we rely on the body and the environment?

How do we mentally represent the world out there? Psychology, philosophy and neuroscience have given two classical answers: as a living space where we act and perceive, dependent on our bodies; as an enduring physical space with its feature, independent of our bodily interactions. The first would be based on egocentric frames of reference anchored to the body, while the second on allocentric frames of reference centred on the environment itself or on objects. This raises some questions concerning how deep the reliance on the body and the environment is when using these reference frames, and whether they are affected differently by the duration of time and the scale (small or large) of space. To answer these questions, I have brought empirical evidence of the effect of motor interference, blindness, environmental characteristics and temporal factors on egocentric and allocentric spatial representational capacity. The results suggest that egocentric representations are deeply rooted in the body, with its sensory and motor properties, and are closely linked to acting now in small-scale or peripersonal space. Allocentric representations are more influenced by environmental than by bodily characteristics, by visual than by motor properties, and seem particularly related to large-scale or extrapersonal space. In line with neurophysiological evidence and a Kantian perspective, it appears that we are endowed with an internal spatial representation system ready to structure environmental information for our purposes. To what extent this system is innate and pervasive in cognition and what is its relationship to the neural 'positioning' substrate discovered by O'Keefe and colleagues requires further scientific investigation.

On the Complementarity between CMMN and iStar in Complex Domain Modeling

Case Management Modeling and Notation (CMMN) and iStar are two distinct, multi-purposed modeling techniques that may be used to represent organizational challenges at separate levels of abstraction. CMMN, a flexible process-oriented technique, aims to extract knowledge that enhances the representational capacity of activity flows for a specific case. Conversely, the iStar framework adopts a goal-oriented modeling approach, effectively capturing the interplay of social actors and their influence on the attainment of organizational objectives. While prior studies have explored methods for integrating these techniques to attain a more comprehensive understanding of the organizational landscape, these remain primarily associated with a distinct level of the organizational matrix (i.e., CMMN for operations and tactics, and iStar for more strategic aspects). As such, any effort to evaluate their semantic proximity appears fragmented, as it deals only with the partial association of specific notations and elements. This article describes the conduct of a dual-purposed literature review to identify specific criteria that might accommodate a more holistic assessment of the two modeling techniques; these criteria are then employed as a guiding framework to construct a set of propositions that articulate the areas of complementarity and/or divergence between these two techniques, as identified in previous research. These propositions are subsequently subjected to validation by domain experts, leveraging a real-world case study in the educational domain. The results show that there can be areas of semantic convergence between the two techniques, suggesting their parallel use to effectively model complex domain problems. Overall, the present study aims to crystallize an approach for conducting complex modeling comparisons that transcends technical considerations.

AssistDistil for Medical Image Segmentation

Deep learning models have demonstrated significant effectiveness in addressing intricate object segmentation and image classification tasks. Nevertheless, their widespread use is impeded by high computational demands, limiting their applicability on resource-constrained devices and in contexts like medical image segmentation. This paper proposes AssistDistil, a semi-knowledge distillation technique designed to facilitate the transfer of knowledge from a larger teacher network to a more compact student model. During the inference process, the student model works in conjunction with the teacher model by condensing the teacher model’s latent information into its own latent representation, thereby boosting its representational capacity. The effectiveness of the proposed approach is demonstrated for multiple case studies in medical image segmentation task of eye segmentation, skin lesion segmentation, and chest X-ray segmentation. Experimental results on the IIITD Cataract Surgery, HAM10000, PH2, Shenzhen and Montgomery chest X-ray datasets demonstrate the efficacy of the proposed approach both in terms of accuracy and computational cost. For example, in comparison to the AUNet-based teacher model, the proposed approach achieves a similar mIOU with only 0.5% of the model size. In the future, we plan to explore knowledge distillation approaches to improve the distillation process in case of large model capacity gap between teacher and student networks. With fewer parameters, we intend for the student model to attain performance comparable to that of the teacher model without additional assistance.

A Efficient and Accurate UAV Detection Method Based on YOLOv5s

Due to the limited computational resources of portable devices, target detection models for drone detection face challenges in real-time deployment. To enhance the detection efficiency of low, slow, and small unmanned aerial vehicles (UAVs), this study introduces an efficient drone detection model based on YOLOv5s (EDU-YOLO), incorporating lightweight feature extraction and balanced feature fusion modules. The model employs the ShuffleNetV2 network and coordinate attention mechanisms to construct a lightweight backbone network, significantly reducing the number of model parameters. It also utilizes a bidirectional feature pyramid network and ghost convolutions to build a balanced neck network, enriching the model’s representational capacity. Additionally, a new loss function, EloU, replaces CIoU to improve the model’s positioning accuracy and accelerate network convergence. Experimental results indicate that, compared to the YOLOv5s algorithm, our model only experiences a minimal decrease in mAP by 1.1%, while reducing GFLOPs from 16.0 to 2.2 and increasing FPS from 153 to 188. This provides a substantial foundation for networked optoelectronic detection of UAVs and similar slow-moving aerial targets, expanding the defensive perimeter and enabling earlier warnings.

Nested compressed co-representations of multiple sequential experiences during sleep.

Animals encounter and remember multiple experiences daily. During sleep, hippocampal neuronal ensembles replay past experiences and preplay future ones. Although most previous studies investigated p/replay of a single experience, it remains unclear how the hippocampus represents many experiences without major interference during sleep. By monitoring hippocampal neuronal ensembles as rats encountered 15 distinct linear track experiences, we uncovered principles for efficient multi-experience compressed p/replay representation. First, we found a serial position effect whereby the earliest and the most recent experiences had the strongest representations. Second, distinct experiences were co-represented in a multiplexed, flickering manner during nested p/replay events, which greatly enhanced the network's representational capacity. Third, spatially contiguous and disjunct track pairs were bound together into contiguous conjunctive representations during sleep. Finally, sequences spanning day-long multi-track experiences were p/replayed at hyper-compressed ratios during sleep. These coding schemes efficiently parallelize, bind and compress multiple sequential representations with reduced interference and enhanced capacity during sleep.

Single Image Super-Resolution via Wide-Activation Feature Distillation Network.

Feature extraction plays a pivotal role in the context of single image super-resolution. Nonetheless, relying on a single feature extraction method often undermines the full potential of feature representation, hampering the model's overall performance. To tackle this issue, this study introduces the wide-activation feature distillation network (WFDN), which realizes single image super-resolution through dual-path learning. Initially, a dual-path parallel network structure is employed, utilizing a residual network as the backbone and incorporating global residual connections to enhance feature exploitation and expedite network convergence. Subsequently, a feature distillation block is adopted, characterized by fast training speed and a low parameter count. Simultaneously, a wide-activation mechanism is integrated to further enhance the representational capacity of high-frequency features. Lastly, a gated fusion mechanism is introduced to weight the fusion of feature information extracted from the dual branches. This mechanism enhances reconstruction performance while mitigating information redundancy. Extensive experiments demonstrate that the proposed algorithm achieves stable and superior results compared to the state-of-the-art methods, as evidenced by quantitative evaluation metrics tests conducted on four benchmark datasets. Furthermore, our WFDN excels in reconstructing images with richer detailed textures, more realistic lines, and clearer structures, affirming its exceptional superiority and robustness.

The Evolution of Symbolic Thought: At the Intersection of Schizophrenia Psychopathology, Ethnoarchaeology, and Neuroscience.

The human capacity for symbolic representation arises, evolutionarily and developmentally, from the exploitation of a widespread sensorimotor network, along a fundamental continuity between embodied and symbolic modes of experience. In this regard, the fine balancing between constrained sensorimotor connections (responsible for self-embodiment processing) and more untethered neural associations (responsible for abstract and symbolic processing) is context dependent and plastically neuromodulated, thus intersubjectively constructed within a specific socio-cultural milieu. Instead, in the schizophrenia spectrum this system falls off catastrophically, due to an unbalance toward too unconstrained sensorimotor connectivity, leading to a profound distortion of self/world relation with a symbolic activity detached from its embodied ground. For this very reason, however, schizophrenia psychopathology may contribute to unveil, in a distorted or magnified way, ubiquitous structural features of human symbolic activity, beneath the various, historically determined cultural systems. In this respect, a comparative approach, linking psychopathology and ethnoarchaeology, allows highlight the following invariant formal characteristics of symbolic processing: (1) Emergence of salient perceptive fragments, which stand out from the perceptual field. (2) Spreading of a multiplicity of new significances with suspension of common-sense meaning. (3) Dynamic and passive character through which meaning proliferation is experienced. This study emphasizes the importance of fine-grained psychopathology to elucidate, within a cross-disciplinary framework, the evolutionarily and developmental pathways that shape the basic structures of human symbolization.

Network Traffic Anomaly Detection Based on Spatiotemporal Feature Extraction and Channel Attention

To overcome the challenges of feature selection in traditional machine learning and enhance the accuracy of deep learning methods for anomaly traffic detection, we propose a novel method called DCGCANet. This model integrates dilated convolution, a GRU, and a Channel Attention Network, effectively combining dilated convolutional structures with GRUs to extract both temporal and spatial features for identifying anomalous patterns in network traffic. The one-dimensional dilated convolution (DC-1D) structure is designed to expand the receptive field, allowing for comprehensive traffic feature extraction while minimizing information loss typically caused by pooling operations. The DC structure captures spatial dependencies in the data, while the GRU processes time series data to capture dynamic traffic changes. Furthermore, the channel attention (CA) module assigns importance-based weights to features in different channels, enhancing the model’s representational capacity and improving its ability to detect abnormal traffic. DCGCANet achieved an accuracy rate of 99.6% on the CIC-IDS-2017 dataset, outperforming other algorithms. Additionally, the model attained precision, recall, and F1 score rates of 99%. The generalization capability of DCGCANet was validated on a subset of CIC-IDS-2017, demonstrating superior detection performance and robust generalization potential.

Models of bee responses to land use and land cover changes in agricultural landscapes - a review and research agenda.

Predictive modelling tools can be used to support the design of agricultural landscapes to promote pollinator biodiversity and pollination services. Despite the proliferation of such modelling tools in recent decades, there remains a gap in synthesising their main characteristics and representation capacities. Here, we reviewed 42 studies that developed non-correlative models to explore the impact of land use and land cover changes on bee populations, and synthesised information about the modelled systems, modelling approaches, and key model characteristics like spatiotemporal extent and resolution. Various modelling approaches are employed to predict the biodiversity of bees and the pollination services they provide, with a prevalence of models focusing on wild populations compared to managed ones. Of these models, landscape indicators and distance decay models are relatively simple, with few parameters. They allow mapping bee visitation probabilities using basic land cover data and considering bee foraging ranges. Conversely, mechanistic or agent-based models delineate, with varying degrees of complexity, a multitude of processes that characterise, among others, the foraging behaviour and population dynamics of bees. The reviewed models collectively encompass 38 ecological, agronomic, and economic processes, producing various outputs including bee abundance, habitat visitation rate, and crop yield. To advance the development of predictive modelling tools aimed at fostering pollinator biodiversity and pollination services in agricultural landscapes, we highlight future avenues for increasing biophysical realism in models predicting the impact of land use and land cover changes on bees. Additionally, we address the challenges associated with balancing model complexity and practical usability.