• Fuses gradient-pixel features to amplify subtle pathological distinctions. • Jointly captures tissue-context and cellular details via dual-branch modeling. • Integrates multi-scale features for class uniformity-variability representation. • SOTA over datasets against costy foundational models Pathological diagnosis is crucial for patient care, and Region of Interest (ROI) analysis serves as a key pathological method for extracting local cellular details to guide precise clinical decision-making. While most of the current foundation models have shown promise in ROI pathological image classification, existing approaches often fall short in addressing the unique characteristics of pathology ROI data from three aspects simultaneously: (1) inter-class similarity, (2) complex global patterns, and (3) multi-scale granularity. To address them, we propose CellMixer, a novel framework designed to extract and integrate local-global ROI pathological image representations. The key innovation lies in the synergistic integration of three corresponding domain-aware components: (1) To amplify subtle morphological distinctions, we designed a data augmentation (GradMix), which selectively fuses gradient maps and pixel-level features to enhances low-level feature sensitivity, directly improving discrimination of visually similar classes; (2) To capture both localized patterns and global tissue structures across ROI regions, we proposed Dual-branch VMamba Block (DVB), which enhances long-range dependency modeling and simultaneously extracts cell-level fine-grained features; (3) To fuse local and global features to concurrently represent intra-class homogeneity and inter-class heterogeneity across scales, a novel feature fusion strategy (Insert-Merge (InM)). Extensive experiments on 8 public pathology ROI datasets demonstrate that CellMixer consistently outperforms existing methods, proving task-specific model, even with limited data, yields superior visual representations to generic foundation models.

Context: The detection of code defects is foundational to modern software development and maintenance, playing a critical role in ensuring software quality and security. However, as software systems grow in scale and complexity, the limitations of traditional static analysis and conventional machine learning techniques have become increasingly evident. These methods rely heavily on intricate, manual feature engineering and fail to capture dynamic runtime behavior, resulting in suboptimal accuracy and elevated error rates. Objective: To address these deficiencies, we propose UF-CDDFM, a unified framework for code defect detection that integrates multi-modal inputs, active learning, and state-of-the-art few-shot learning techniques. We aim to improve detection performance, reduce feature selection complexity and sample bias through active learning, and maintain practical efficiency in real-world development contexts. Methods: UF-CDDFM employs parallel encoding of source code, code annotations, and abstract syntax trees (ASTs) using large language models (LLMs) alongside multilayer perceptrons (MLPs) to derive robust, high-fidelity representations of code. To streamline feature selection and mitigate sample bias, an active learning component is introduced for automated identification of high-quality features. Addressing the pervasive challenge of data scarcity, we incorporate two complementary few-shot learning strategies-MAML for small-scale datasets and LEO for larger-scale settings to enhance overall generalization capability. Results: Empirical evaluations demonstrate that UF-CDDFM consistently outperforms existing methods, establishing new state-of-the-art detection rates: 72.04% for defect detection and 95.23% for clone detection. Crucially, these gains are achieved within resource-constrained computational environments, which highlights the practicality of the method. Conclusion: By fusing multi-modal code representations, active learning, and adaptive few-shot learning techniques, UF-CDDFM delivers significant improvements in detection accuracy and computational efficiency. This work offers a new paradigm for robust, scalable, and practical code defect and clone detection in modern software engineering.

Pathological diagnosis is highly reliant on image analysis, where Regions of Interest (ROIs) serve as the primary basis for diagnostic evidence, while whole-slide image (WSI)-level tasks primarily capture aggregated patterns. To extract these critical morphological features, ROI-level Foundation Models (FMs) based on Vision Transformers (ViTs) and large-scale self-supervised learning (SSL) have been widely adopted. However, three core limitations remain in their application to ROI analysis: (1) cross-magnification domain shift, as fixed-scale pretraining hinders adaptation to diverse clinical settings; (2) inadequate local-global relationship modeling, wherein the ViT backbone of FMs suffers from high computational overhead and imprecise local characterization; (3) insufficient fine-grained sensitivity, as traditional self-attention mechanisms tend to overlook subtle diagnostic cues. To address these challenges, we propose SSMamba, a hybrid SSL framework that enables effective fine-grained feature learning without relying on large external datasets. This framework incorporates three domain-adaptive components: Mamba Masked Image Modeling (MAMIM) for mitigating domain shift, a Directional Multi-scale (DMS) module for balanced local-global modeling, and a Local Perception Residual (LPR) module for enhanced fine-grained sensitivity. Employing a two-stage pipeline, SSL pretraining on target ROI datasets followed by supervised fine-tuning (SFT), SSMamba outperforms 11 state-of-the-art (SOTA) pathological FMs on 10 public ROI datasets and surpasses 8 SOTA methods on 6 public WSI datasets. These results validate the superiority of task-specific architectural designs for pathological image analysis.

• Parallel branches resolve conflicts: masked autoencoder & edge-aware contrastive path. • Random coordinate attention fuses cross-branch features via gradient-modulated keys. • Gradient histogram alignment balances objectives, boosting classification and detection. • Validated on ImageNet, COCO, VOC, Birdsnap, Leaf Disease datasets shows SOTA performance.

As the demand for logistics continues to surge, traditional ground delivery methods fall short, prompting the advent of low-altitude unmanned aerial vehicle (UAV) delivery as a revolutionary solution for rapid and adaptable logistics services. This paper introduces the Adaptive-Formation Multi-UAV Task Allocation with Load-Dependent Constraints (AF-MUTALC), an approach that transcends the Capacitated Vehicle Routing Problem (CVRP) framework with the distinctive operational constraints of UAVs, addressing the burgeoning need for efficient low-altitude delivery systems. Our proposed model integrates the impact of payload on UAV performance, encompassing constraints such as limited power, endurance, payload capacity, and the intricate interplay between flight dynamics and energy consumption. Additionally, the model contemplates the inherent uncertainties in transit times during actual UAV operations. To effectively manage this multi-objective optimization challenge, we employ an Enhanced Multi-Objective Evolutionary Algorithm (MOEA) that combines the strengths of Non-dominated Sorting Genetic Algorithm II (NSGA-II) with a refined local search strategy. This approach aims to optimize resource utilization by minimizing the UAV fleet size and reducing the maximum flight time of a single UAV, thus ensuring an optimal balance between operational efficiency and resource allocation. Our algorithm has undergone rigorous validation through simulations, showcasing its ability to surpass conventional methodologies in delivering high-quality solutions. Furthermore, corroborated by empirical trials in real-world settings, we attest to the feasibility of the algorithm for real-life implementation.

Recent advancements in large language models (LLMs) have demonstrated their potential to significantly impact finance trading, particularly through sentiment analysis. The cryptocurrency market, known for its volatility and unpredictability, often renders price-based trading approaches inadequate. This necessitates the adoption of more sophisticated techniques such as market sentiment analysis, which can benefit from the insights provided by LLMs. This study introduces an innovative method that integrates sentiment analysis derived from five distinct LLMs with deep reinforcement learning to devise a cryptocurrency trading strategy. Recognizing that LLM outputs cannot be guaranteed to be infallibly accurate, which contributing to the LLM hallucinations, this paper details the implementation of a stringent outlier detection and removal process. By adopting a “Trust-The-Majority” strategy, the research aims to ensure that trading decisions are informed by reliable sentiment data. In addition, sentiment scores are traditionally timestamped to the publication of news or social media posts. To more accurately reflect the actual impact of such information on market sentiment, this study applies the Ebbinghaus Forgetting Curve to model the waning influence of information over time. This allows for a more nuanced understanding of how news affects market dynamics. The enhanced sentiment scores, in conjunction with traditional market data such as OHLCV (Open, High, Low, Close, Volume), are utilized by a deep reinforcement learning model to make trading decisions. Experimental results demonstrate that the proposed multi-LLM sentiment-driven framework improves trading performance in the fast-paced cryptocurrency market. The methodology outlined in this paper offers a solid foundation for incorporating real-time market sentiment analysis into financial applications.

Bounding box regression (BBR) is a critical component in object detection, significantly influencing the accuracy of object localization. However, existing Intersection over Union (IoU)-based loss functions encounter two primary challenges: (i) The penalty factor configuration often results in the expansion of anchor boxes during the regression, which in turn slows the convergence rate of the loss. (ii) There is a spatial imbalance caused by the disproportionate influence of anchor boxes with minimal overlap with the ground truth boxes. To resolve these two challenges, this paper proposes a novel loss function termed Fast-IoU, designed to swiftly and precisely measure the overlap area and aspect ratio in BBR. Building upon this, a dynamic non-monotonic focusing mechanism is integrated to evaluate the quality of anchor boxes in a non-linear manner. Fast-IoU can enhance the capability to focus on anchor boxes of medium quality. By incorporating Fast-IoU into popular object detectors such as YOLOv7, YOLOv8 and YOLOv10, we achieved an increase in average precision and improved performance compared to their original loss functions on the MS COCO datasets, thus validating the effectiveness of ourproposed improvement strategies.

Deep Reinforcement Learning (DRL)-based Dynamic Voltage Frequency Scaling (DVFS) has shown great promise for energy conservation in embedded systems. While many works were devoted to validating its efficacy or improving its performance, few discuss the feasibility of the DRL agent deployment for embedded computing. State-of-the-art approaches focus on the miniaturization of agents’ inferential networks, such as pruning and quantization, to minimize their energy and resource consumption. However, this spatial-based paradigm still proves inadequate for resource-stringent systems. In this paper, we address the feasibility from a temporal perspective, where FiDRL, a flexible invocation-based DRL model is proposed to judiciously invoke itself to minimize the overall system energy consumption, given that the DRL agent incurs non-negligible energy overhead during invocations. Our approach is three-fold: (1) FiDRL that extends DRL by incorporating the agent's invocation interval into the action space to achieve invocation flexibility; (2) a FiDRL-based DVFS approach for both inter- and intra-task scheduling that minimizes the overall execution energy consumption; and (3) a FiDRL-based DVFS platform design and an on/off-chip hybrid algorithm specialized for training the DRL agent for embedded systems. Experiment results show that FiDRL achieves 55.1% agent invocation cost reduction, under 23.3% overall energy reduction, compared to state-of-the-art approaches.

The shift towards transportation electrification, marked by the rising use of electric vehicles (EVs) and the development of fast charging stations (FCS), plays a crucial role in transport decarbonization initiatives. To optimize the rollout of FCS and set appropriate charging service fees (CSF)-a process referred to as the coupled FCS multi-stage bi-level operation problem (FCS-MBOP)-is essential for improving both investment and operational efficiency within the integrated power distribution and transportation network (CPTN). For operators, it's not only necessary to adapt to short-term fluctuations within the environment but also to swiftly respond to changes in the FCS layout resulting from various long-term investment decisions. To address this complexity, we introduce a dual-timescale evolutionary assist deep reinforcement learning framework, which includes two specialized agents with distinct functions: an investment agent (planner) and an operational agent (operator). The planner focuses on annual investments, evolving long-term strategies that weigh social benefits against investment costs through the use of a genetic algorithm (GA). In contrast, the operator acts on an hourly basis, fine-tuning CSF to alleviate traffic congestion and minimize the social costs, while taking into account the planner's feasible investment decisions. Leveraging the integrated capabilities of a graph neural network (GNN), long-short-term memory (LSTM), and attention mechanisms, our framework's agents are adept at extracting both temporal and spatial features and facilitating the transfer of experiences across different investment stages. Empirical evidence underscores the effectiveness of our approach, showcasing its ability to surpass conventional methodologies in delivering high-quality solutions.

Portfolio optimization is a strategic approach aiming at achieving an optimal balance between risk and returns through the judicious allocation of limited capital across various assets. In recent years, there has been a growing interest in leveraging Deep Reinforcement Learning (DRL) to tackle the complexities of portfolio optimization. Despite its potential, a notable limitation of DRL algorithms is their inherent difficulty in integrating conflicted objectives with the reward functions throughout the learning process. Typically, DRL's reward function prioritizes the maximization of returns or other performance indicators, often overlooking the integration of risk aspects. Furthermore, the standard DRL framework struggles to incorporate practical constraints, such as cardinality and bounding, into the decision process. Without these constraints, the investment strategies developed might be unrealistic and unmanageable. To this end, in this paper, we propose an adaptive and safe DRL framework, which can dynamically optimize the portfolio weights while strictly respecting practical constraints. In our method, any infeasible action (i.e., one that violates the constraints) decided by the RL agent will be mapped to a feasible region using a safety layer. The extended Markowitz Mean-Variance (M-V) model is explicitly encoded in the safety layer to ensure the feasibility of the actions from the alternative views. In addition, we utilize Projection-based Interior-point Policy Optimization (IPO) to resolve multiple objectives and constraints in the examined problem. Extensive results on real-world datasets show that our method is effective in strictly respecting constraints under dynamic market environments, in contrast to prevailing data- driven trading strategies and conventional model-based static solutions.

Gesture recognition in few-shot scenarios presents a significant challenge due to the scarcity of labeled data. In this work, we propose a novel end-to-end framework tailored for few-shot gesture recognition, addressing the limitations of current methods. A novel feature map generating method is proposed to leverage a greater number of dimensions in capturing gesture feature information and simplify the structure of network. Our approach also maximizes the utility of a limited set of real training samples by generating new virtual samples in two domains based on data augmentation, and employs a feature fusion strategy for comprehensive gesture characteristic extraction by using both Convolutional Neural Network (CNN) and Histogram of Oriented Gradients (HOG) to extract features. Extensive experimental results validate the efficacy of our proposed method, achieving a final accuracy of 85.26%, exhibiting a remarkable 35% improvement over the baseline, thereby confirming the effectiveness of our work in the challenging few-shot gesture recognition task.

Sequential decision making is an important part of robotic problems that is receiving unprecedented attention from both academia and industry. Recently, Deep Reinforcement Learning (DRL) has shown its promising capabilities in decision making problems. However, traditional DRL algorithms directly operate in the space of low-level actions, when it is applied in the domain of robotics, it can easily result in an exponential growth of computational complexity and suffer from the “curse of dimensionality”, becoming less efficient as the dimensionality of the environment increases. To address this issue, a novel DRL hyper-heuristic approach is proposed in this paper. The proposed approach is tailored to align with a problem taken from a real-world competition by taking advantage of well-developed low-level heuristic actions in order to narrow the search space and speed up the convergence. This fundamental contribution is a significant step forward from earlier approaches that directly exploit the entire low-level action domain. A state augmentation scheme and a novel reward design are utilized to further improve the performance of the proposed method. Moreover, a Real-to-Sim based training framework is developed to reduce the cost of acquiring real-time data and improve the robustness of agent’s decision-making model. Numerous experimental results demonstrate our proposed method can achieve notable performance gains compared to both competitive DRL baselines and heuristic approaches of the same problem in both known environment and previously unseen scenarios.

The online bin packing problem is a well-known optimization challenge that finds application in a wide range of real-world scenarios. In the paper, we propose a novel algorithm called FuzzyPatternPack(FPP), which leverages fuzzy inference and pattern-based predictions of the distribution of item sizes in online bin packing. In comparison to traditional heuristics like BestFit(BF) and FirstFit(FF), as well as the more recent PatternPack(PaP) and ProfilePacking(PrP) algorithm based on online predictions, FPP demonstrates competitive and superior performance in solving various benchmark problems. Particularly, it excels in addressing problems with evolving distributions, making it a promising solution for real-world applications where the item sizes may change over time. This research unveils the promising potential of employing fuzzy logic to effectively address uncertainty in scheduling and planning problems.

Transportation electrification, involving large-scale integration of electric vehicles (EV) and fast charging stations (FCS), constitutes one of the key enablers toward decarbonization. Coordination of EV charging routes and demand through suitably designed price signals constitutes an imperative step in secure and economic operation of the coupled transportation network (TN) and power distribution network (PDN). In this work, we model the noncooperative pricing game of self-interested FCSs, taking into account the complex interactions between the EV users and the coupled operation of TN and PND. The uncertainties stemming from the EV users' cost elasticity and their travel energy requirements are encapsulated in the modeling of the TN, while the power flows in the PDN are coordinated considering the penetration of renewable energy sources. A modified multiagent proximal policy optimization method is developed to solve the pricing game. It employs an attention mechanism to selectively incorporate agents' representative information for estimating the Q-values. As such, it not only mitigates the nonstationary effect without exploding the input of the centralized critic but also safeguards the business confidentiality of FCSs. Moreover, a sequential updating scheme is used to ensure policy monotonic improvement and a Bayesian inference technique is adopted to enhance the robustness of the pricing strategy. Case studies on a large-scale test CTPN system reveal that the proposed method facilitates sufficient competition among FCSs, which is able to drive down the average charging prices for EV users. It also smooths out the spatial distribution of EV charging demands, which reduces the traffic congestions in the TN while enhancing the wind absorption and cost efficiency of the PDN.

A competitive traveling salesmen problem is a variant of traveling salesman problem in that multiple agents compete with each other in visiting a number of cities. The agent who is the first one to visit a city will receive a reward. Each agent aims to collect as more rewards as possible with the minimum traveling distance. There is still not effective algorithms for this complicated decision making problem. We investigate an improved ant colony approach for the competitive traveling sales-men problem which adopts a time dominance mechanism and a revised pheromone depositing method to improve the quality of solutions with less computational complexity. Simulation results show that the proposed algorithm outperforms the state of art algorithms.

Loss functions are essential to bounding box regression which plays a significant role in deep learning based object detection. Despite the effectiveness of the popular Intersection over Union (IoU) based losses, there is still an imbalance problem of high- and low-quality predicted bounding boxes, impeding the accuracy and convergence speed during bound box regression. Specifically, we observe that the huge amount of predicted bounding boxes having small overlapping regions with ground truth box overwhelms the amount of predicted bounding boxes having large overlapping regions. In this paper, we propose a simple weighting factor that is able to reshape the existing IoU-based losses according to a geometric relationship of bounding boxes. In this way, we are able to effectively down-weight the contribution of low-quality predicted boxes and focus training on high-quality ones. Extensive experiments have been carried out on popular IoU-based losses with various object detection techniques. By simply incorporating the proposed weighting factor, we are able to achieve notable performance gains on the popular MS COCO dataset.

Remote measurement of SIF has opened a new perspective to assess plant actual photosynthesis at larger, ecologically relevant scales. However, understanding the underling mechanisms between SIF and GPP remains challenging before SIF used as a robust constraint for estimating GPP. In this study, GOME-2 SIF was found to be consistently related to MODIS GPP. We also noticed the SIF-GPP relationship was ecosystem-specific and influenced by land surface temperature. The former was due to some structural and physiological characteristics related to each ecosystem. The latter can be attributed to the biochemical process influenced by temperature conditions. Model simulations also indicated the SIF-GPP relationship was complex and affected by some factors like chlorophyll content and LAI. Our study contributes to a better understanding of the information inherent in remotely sensed SIF and its functional relationship to GPP.

The scale effects in earth science, which are related to various aspects in remote sensing monitoring, have become an international prosperous research area. As spatial heterogeneity of the earth system limits the transferring between different scale, it is necessary to study these spatial heterogeneity factors, and analyze their impact on NPP scale effect. Then we can introduce an approach to perform spatial scale calibration based on a correction factor for scale effect, and perform it to NPP. This study presented an approach driven by remotely sensed data and meteorological data to estimate GPP and NPP over regional scales. By using multi-scale data and different scaling strategies, NPP of Heihe River Basin in 2012 with various scales were derived. With a focus on differences among land cover types, we introduced and tested a kind of spatial scale calibration method, to get close to the real value of the net primary productivity.

Portfolio optimization is one of the most important problems in the finance field. The traditional mean-variance model has its drawbacks since it fails to take the market uncertainty into account. In this work, we investigate a two-stage stochastic portfolio optimization model with a comprehensive set of real world trading constraints in order to capture the market uncertainties in terms of future asset prices. A hybrid approach, which integrates genetic algorithm (GA) and a linear programming (LP) solver is proposed in order to solve the model, where GA is used to search for the assets selection heuristically and the LP solver solves the corresponding sub-problems of weight allocation optimally. Scenarios are generated to capture uncertain prices of assets for five benchmark market instances. The computational results indicate that the proposed hybrid algorithm can obtain very promising solutions. Possible future research directions are also discussed.

Portfolio optimization is an important problem based on the modern portfolio theory (MPT) in the finance field. The idea is to maximize the portfolio expected return as well as minimizing portfolio risk at the same time. In this work, we propose a combinatorial algorithm for the portfolio optimization problem with the cardinality and bounding constraints. The proposed algorithm hybridizes a metaheuristic approach (particle swarm optimization, PSO) and a mathematical programming method where PSO is used to deal with the cardinality constraints and the math programming method is used to deal with the rest of the model. Computational results are given for the benchmark datasets from the OR-library and they indicate that it is a useful strategy for this problem. We also present the solutions obtained by the CPLEX mixed integer program solver for these instances and they can be used as the criteria for the comparison of algorithms for the same problem in the future.