Large Language Models (LLMs) have demonstrated notable capabilities across financial tasks, including financial report summarization, earnings call transcript analysis, and asset classification. However, their real-world effectiveness in managing complex fund investment remains inadequately assessed. A fundamental limitation of existing benchmarks for evaluating LLM-driven trading strategies is their reliance on historical back-testing, inadvertently enabling LLMs to "time travel"—leveraging future information embedded in their training corpora, thus resulting in possible information leakage and overly optimistic performance estimates. To address this issue, we introduce DeepFund, a live fund benchmark tool designed to rigorously evaluate LLM in real-time market conditions. Utilizing a multi-agent architecture, DeepFund connects directly with real-time stock market data—specifically data published after each model’s pretraining cutoff—to ensure fair and leakage-free evaluations. Empirical tests on nine flagship LLMs from leading global institutions across multiple investment dimensions—including ticker-level analysis, investment decision-making, portfolio management, and risk control—reveal significant practical challenges. Notably, even cutting-edge models such as DeepSeek-V3 and Claude-3.7-Sonnet incur net trading losses within DeepFund real-time evaluation environment, underscoring the present limitations of LLMs for active fund management. Our code is available at https://github.com/HKUSTDial/DeepFund.

Automated design of metaheuristic algorithms offers an attractive avenue to reduce human effort and gain enhanced performance beyond human intuition. Current automated methods design algorithms within a fixed structure and operate from scratch. This poses a clear gap toward fully discovering potentials over the metaheuristic family and fertilizing from prior design experience. To bridge the gap, this article proposes an autoregressive learning-based designer for automated design of metaheuristic algorithms. Our designer formulates metaheuristic algorithm design as a sequence generation task, and harnesses an autoregressive generative network to handle the task. This offers two advances. First, through autoregressive inference, the designer generates algorithms with diverse lengths and structures, enabling to fully discover potentials over the metaheuristic family. Second, prior design knowledge learned and accumulated in neurons of the designer can be retrieved for designing algorithms for future problems, paving the way to continual design of algorithms for open-ended problem solving. Extensive experiments on numeral benchmarks and real-world problems reveal that the proposed designer generates algorithms that outperform all human-created baselines on 24 out of 25 test problems. The generated algorithms display various structures and behaviors, reasonably fitting for different problem-solving contexts. Code is available athttps://github.com/auto4opt/ALDes.

In this paper, we present an open-source pure-Python library called PyPop7 for black-box optimization (BBO). As population-based methods (e.g., evolutionary algorithms, swarm intelligence, and pattern search) become increasingly popular for BBO, the design goal of PyPop7 is to provide a unified API and elegant implementations for them, particularly in challenging high-dimensional scenarios. Since these population-based methods easily suffer from the notorious curse of dimensionality owing to random sampling as one of core operations for most of them, recently various improvements and enhancements have been proposed to alleviate this issue more or less mainly via exploiting possible problem structures: such as, decomposition of search distribution or space, low-memory approximation, low-rank metric learning, variance reduction, ensemble of random subspaces, model self-adaptation, and fitness smoothing. These novel sampling strategies could better exploit different problem structures in high-dimensional search space and therefore they often result in faster rates of convergence and/or better qualities of solution for large-scale BBO. Now PyPop7 has covered many of these important advances on a set of well-established BBO algorithm families and also provided an open-access interface to adding the latest or missed black-box optimizers for further functionality extensions. Its well-designed source code (under GPL-3.0 license) and full-fledged online documents (under CC-BY 4.0 license) have been freely available at https://github.com/Evolutionary-Intelligence/pypop and https://pypop.readthedocs.io, respectively.

In the post-Moore era, main performance gains of black-box optimizers are increasingly depending on parallelism, especially for large-scale optimization (LSO). Here we propose to parallelize the well-established covariance matrix adaptation evolution strategy (CMA-ES) and in particular its one latest LSO variant called limited-memory CMA-ES (LM-CMA). To achieve efficiency while approximating its powerful invariance property, we present a multilevel learning-based meta-framework for distributed LM-CMA. Owing to its hierarchically organized structure, Meta-ES is well-suited to implement our distributed meta-framework, wherein the outer-ES controls strategy parameters while all parallel inner-ESs run the serial LM-CMA with different settings. For the distribution mean update of the outer-ES, both the elitist and multi-recombination strategy are used in parallel to avoid stagnation and regression, respectively. To exploit spatiotemporal information, the global step-size adaptation combines Meta-ES with the parallel cumulative step-size adaptation. After each isolation time, our meta-framework employs both the structure and parameter learning strategy to combine aligned evolution paths for CMA reconstruction. Experiments on a set of large-scale benchmarking functions with memory-intensive evaluations, arguably reflecting many data-driven optimization problems, validate the benefits (e.g., effectiveness w.r.t. solution quality, and adaptability w.r.t. second-order learning) and costs of our meta-framework.

Bacterial foraging optimization (BFO) has been exploited for function optimization, owing to its innovative ideas gleaned from the microbiological system. This paper first discusses its three crucial limitations: high computational cost, difficulty in parameter settings, and premature convergence. To alleviate the above problems, simplified BFO with quorum sensing (QS) is proposed. First, a novel computational framework is provided to reduce the computational complexity, leading to a simplified version. Second, the concept of “QS,” bacterial reciprocal behavior, is integrated into the simplified version by utilizing a new position updating equation coupled with a dynamic communication topology. Each bacterium adjusts its search trajectory based on both biased random walk and promising search directions provided by its communicatees. The communicatees are selected via a dynamic communication topology, where a rank-based communication strategy and two information mutation schemes are used for global exploration of the search space. Finally, a parameter automation strategy is introduced to promote the exploitation of promising regions. Further, the effectiveness and efficiency of the proposed algorithm are empirically confirmed on 30 benchmark functions, by comparing it with the four variants of BFO and four other advanced algorithms.

Recently, graph neural networks (GNNs) have been shown to be an effective tool for learning the node representations of the networks and have achieved good performance on the semi-supervised node classification task. However, most existing GNNs methods fail to take networks’ temporal information into account, therefore, cannot be well applied to dynamic network applications such as the continuous-time link prediction task. To address this problem, we propose a Temporal Dependent Graph Neural Network (TDGNN), a simple yet effective dynamic network representation learning framework which incorporates the network temporal information into GNNs. TDGNN introduces a novel Temporal Aggregator (TDAgg) to aggregate the neighbor nodes’ features and edges’ temporal information to obtain the target node representations. Specifically, it assigns the neighbor nodes aggregation weights using an exponential distribution to bias different edges’ temporal information. The performance of the proposed method has been validated on six real-world dynamic network datasets for the continuous-time link prediction task. The experimental results show that the proposed method outperforms several state-of-the-art baselines.

Decomposition-based optimizers have shown very promising computational and convergence performance on many large-scale real-parameter optimization problems. Among them, a class of recently proposed cooperative coevolutionary algorithms (CCEAs) and a type of conventional block coordinate descent algorithms (BCDAs) are arguably the two most representative frameworks applied to the minimization of non-differentiable and differentiable objective function, respectively. This paper explores the connections between CCEAs and BCDAs, which can help gain deeper understandings of CCEAs. First, we propose a unified analytical framework for both CCEAs and BCDAs to capture the common game-theoretic nature by combining their respective theoretical advances. Second, many real-world objective functions are non-additively separable, where all decision variables interact with each other in a direct or indirect fashion. However, most of the state-of-the-art decomposition strategies for CCEAs can only capture the simple additive separability and cannot recognize the non-additive separability, but which has been widely studied in the BCDAs context. The performance of CCEAs on such functions is yet to be fully understood since intuitively CCEAs seem to be not suitable for them. We use the proposed framework to confirm and extend CCEAs' applicability to a special class of non-additively separable functions. Finally, based on the proposed framework, we provide two practical suggestions as well as a suite of new test functions to help design practically better CCEAs for large-scale optimization.

Inspired by the ideas from the mutual cooperation of symbiosis in natural ecosystem, this paper proposes a new variant of PSO, named Symbiosis-based Alternative Learning Multi-swarm Particle Swarm Optimization (SALMPSO). A learning probability to select one exemplar out of the center positions, the local best position, and the historical best position including the experience of internal and external multiple swarms, is used to keep the diversity of the population. Two different levels of social interaction within and between multiple swarms are proposed. In the search process, particles not only exchange social experience with others that are from their own sub-swarms, but also are influenced by the experience of particles from other fellow sub-swarms. According to the different exemplars and learning strategy, this model is instantiated as four variants of SALMPSO and a set of 15 test functions are conducted to compare with some variants of PSO including 10, 30 and 50 dimensions, respectively. Experimental results demonstrate that the alternative learning strategy in each SALMPSO version can exhibit better performance in terms of the convergence speed and optimal values on most multimodal functions in our simulation.

The Integrated Yard Truck Scheduling and Storage Allocation Problem (YTS-SAP) is one of the major optimization problems in container port which minimizes the total delay for all containers. To deal with this NP-hard scheduling problem, standard particle swarm optimization (SPSO) and a local version PSO (LPSO) are developed to obtain the optimal solutions. In addition, a simple and effective `problem mapping' mechanism is used to convert particle position vector into scheduling solution. To evaluate the performance of the proposed approaches, experiments are conducted on different scale instances to compare the results obtained by GA. The experimental studies show that PSOs outperform GA in terms of computation time and solution quality.

BACKGROUND: Recent microbiologic studies have shown that quorum sensing mechanisms, which serve as one of the fundamental requirements for bacterial survival, exist widely in bacterial intra- and inter-species cell-cell communication. Many simulation models, inspired by the social behavior of natural organisms, are presented to provide new approaches for solving realistic optimization problems. Most of these simulation models follow population-based modelling approaches, where all the individuals are updated according to the same rules. Therefore, it is difficult to maintain the diversity of the population. RESULTS: In this paper, we present a computational model termed LCM-QS, which simulates the bacterial quorum-sensing (QS) mechanism using an individual-based modelling approach under the framework of Agent-Environment-Rule (AER) scheme, i.e. bacterial lifecycle model (LCM). LCM-QS model can be classified into three main sub-models: chemotaxis with QS sub-model, reproduction and elimination sub-model and migration sub-model. The proposed model is used to not only imitate the bacterial evolution process at the single-cell level, but also concentrate on the study of bacterial macroscopic behaviour. Comparative experiments under four different scenarios have been conducted in an artificial 3-D environment with nutrients and noxious distribution. Detailed study on bacterial chemotatic processes with quorum sensing and without quorum sensing are compared. By using quorum sensing mechanisms, artificial bacteria working together can find the nutrient concentration (or global optimum) quickly in the artificial environment. CONCLUSIONS: Biomimicry of quorum sensing mechanisms using the lifecycle model allows the artificial bacteria endowed with the communication abilities, which are essential to obtain more valuable information to guide their search cooperatively towards the preferred nutrient concentrations. It can also provide an inspiration for designing new swarm intelligence optimization algorithms, which can be used for solving the real-world problems.