Large language models (LLMs) have achieved striking performance across a broad range of reasoning benchmarks, yet the quality of their outputs remains acutely sensitive to prompt design. A meticulously engineered prompt can coax an LLM into correctly answering even highly complex questions, prompting a surge of research into techniques that boost prompt efficacy. Manual chain-of-thought (CoT) prompting and automated prompt generation have emerged as leading strategies. However, CoT exemplars must be painstakingly tailored to each task, making the process labor-intensive, while prompts optimized for a single task often fail to generalize. We introduce a simple yet powerful alternative: by treating the LLM itself as a multitask optimizer, we enable iterative self-refinement of prompts through natural language task descriptions and few-shot in-context learning (ICL). Recognizing that individual prompts exhibit task-dependent sensitivity, we further ensemble the top-performing prompts at inference time. Empirical evaluation with several state-of-the-art LLMs shows that our method substantially surpasses prior baselines, delivering gains of up to 6.0% on mathematical reasoning tasks and 10.2% on commonsense reasoning benchmarks.

Designing specific optimizers for specific optimization tasks has been an important but difficult task due to the No Free Lunch theorem. Nowadays, large language models have been widely applied to automatic algorithm design, but much of the current work still focuses on the design of heuristic rules. In this work, we propose a framework for the evolution of optimizer (Fireworks Algorithm) based on large language models, thereby achieving the customization of specific Fireworks Algorithms for different specific problems. In the experimental part, we use GFWA as the starting point for evolution and achieve significant improvement in optimization on both CEC2013 and CEC2017, while on many engineering problems with very complex constraints, the LLM-driven Fireworks Algorithm generates solutions that not only satisfy all constraints but also have higher quality compared to previous work, which underscores the great potential of the large language model in optimizing the optimizer . It is worth noting that this framework diminishes the reliance on expert knowledge, which was previously emphasized.

This paper proposes the Large Language Model Assisted Operator Design Framework (LLM-AODF), a novel approach for the automatic generation and optimization of operators within the Fireworks Algorithm (FWA). The framework involves inputting the source code of the fireworks algorithm into a large language model to generate new mutation operators, followed by multiple rounds of iterative optimization of the operator design. By leveraging large language models, LLMAODF can fully automate the improvement of mutation operators in the FWA. The effectiveness of the framework is validated through experiments using the CEC2013 black-box optimization test suite. Experimental results indicate that mutation operators generated by different large language models can improve the performance of the FWA to different extents. An in-depth analysis of the mutation operators generated by various models revealed that large language models can produce complex and diverse mutation strategies through multiple iterations, thereby enhancing the algorithm’s exploration and exploitation capabilities. The framework can be easily extended to different algorithms and various problems, offering a novel approach for the automated design of algorithms.

Stock price forecasting remains a critical challenge due to market non-stationarity and the influence of multiple factors. Existing studies apply frequency domain analysis methods to mitigate the impacts of non-stationarity by decoupling high- and low-frequency variation patterns. However, these approaches primarily focus on single series decomposition while neglecting cross frequency interactions among different stocks. Moreover, as a key indicator of overall market trends, current methods inadequately utilize market index information. In this paper, we propose STEAM, a Spatio-Temporal Wavelet Enhanced Attention Mamba model. We introduce Discrete Wavelet Transform (DWT) to disentangle multi-frequency temporal features and propose Wavelet Enhanced Attention (WEA) to capture cross frequency spatial dependencies, effectively leveraging both local and global inter-stock relationships. To extract the synergistic spatio-temporal dependencies in stock data, AMamba module is designed that integrates WEA into the Mamba-2 architecture. Additionally, to further enhance the model's perception of macro-market conditions, we incorporate market index as a prefix, guiding predictions with holistic market information in both spatial and temporal dependencies learning. Extensive experiments across multiple national stock markets demonstrate that STEAM achieves state-of-the-art forecasting performance.

Market regimes are a critical factor influencing stock price fluctuations. We observe that regime characteristics can be reflected in the dynamic variations in the strength of multiple inter-stock relationships. However, existing methodologies predominantly rely on a single graph constructed using prior knowledge or directly infer a singular type of relationship from time series data. These approaches fail to account for the existence of multiple types of relationships and their dynamic variations in strength. To address this limitation, we propose a novel framework, the Dual-Path Adaptive-Correlation Spatial-Temporal Inverted Transformer (DPA-STIFormer), which decouples time series data to learn diverse types of relationships and introduces a gated mechanism to adaptively fuse them, thereby accommodating different market regimes. Experiments conducted on four stock market datasets demonstrate state-of-the-art performance, with an average improvement of over 5%, validating the model’s superior capability in uncovering latent temporal-correlation patterns.

The integration of temporal features and correlations across time series has emerged as an effective strategy in time series prediction. Spatial-Temporal Graph Neural Networks (STGNNs) have demonstrated strong performance on many temporal-correlation forecasting problems. However, their effectiveness and robustness are less satisfactory when applied to tasks that lack periodicity, such as stock market prediction, and they are constrained by memory limitations, rendering them unsuitable for problems with a large number of nodes. To address these challenges, we propose the Temporal-Correlation Graph Pre-trained Network (TCGPN), which utilizes a temporal-correlation fusion encoder for mixed representation and incorporates pre-training methods with carefully designed temporal and correlation tasks. The structure is independent of node number and order, enabling improved outcomes through various data augmentation techniques, and reduces memory consumption during training via multiple sampling strategies. Experiments conducted on real stock market datasets, CSI300, CSI500, NASDAQ and NYSE, demonstrate that fine-tuning a simple MLP in downstream tasks achieves state-of-the-art results, validating TCGPN’s ability to capture robust temporal correlation patterns.

Contrastive language image pretraining (CLIP) has received widespread attention since its learned representations can be transferred well to various downstream tasks. During the training process of the CLIP model, the InfoNCE objective aligns positive image-text pairs and separates negative ones. We show an underlying representation grouping effect during this process: the InfoNCE objective indirectly groups semantically similar representations together via randomly emerged within-modal anchors. Based on this understanding, in this article, prototypical contrastive language image pretraining (ProtoCLIP) is introduced to enhance such grouping by boosting its efficiency and increasing its robustness against the modality gap. Specifically, ProtoCLIP sets up prototype-level discrimination between image and text spaces, which efficiently transfers higher level structural knowledge. Furthermore, prototypical back translation (PBT) is proposed to decouple representation grouping from representation alignment, resulting in effective learning of meaningful representations under a large modality gap. The PBT also enables us to introduce additional external teachers with richer prior language knowledge. ProtoCLIP is trained with an online episodic training strategy, which means it can be scaled up to unlimited amounts of data. We trained our ProtoCLIP on conceptual captions (CCs) and achieved an +5.81% ImageNet linear probing improvement and an +2.01% ImageNet zero-shot classification improvement. On the larger YFCC-15M dataset, ProtoCLIP matches the performance of CLIP with 33% of training time.

The potential intelligence behind advanced machining systems (AMSs) offers positive contributions toward process improvement. Imitation learning (IL) offers an appealing approach to accessing this intelligence by observing demonstrations from skilled technologists. However, existing IL algorithms that implement single policy strategies have yet to consider realistic scenarios for complex AMS tasks, where the available demonstrations may have come from various experts. Moreover, most IL assumes that the expert's policy is optimal, preventing the learning from fulfilling the previously ignored green missions. This article introduces a novel three-phase policy search algorithm based on IL, enabling the learning of heterogeneous expert policies while balancing energy preferences. The first phase equips the agent with machining basics through upper-level policy learning, generating an imitation policy distribution with various decision-making principles. The second phase enhances energy conservation capabilities by employing Pareto-improvement learning and fine-tuning the agent's policies to a Pareto-policy manifold. The third phase produces outcomes and amplifies the efficacy of human feedback by utilizing ensemble policies. The experimental results indicate that the proposed method outperforms meta-heuristics, exhibiting superior solution quality and faster computation times compared to four diverse baseline methods, each with diverse samples.

The Knapsack Problem (KP) is a renowned combinatorial optimization challenge, recognized for its NP-hard complexity. This characteristic renders large-scale KP instances resistant to resolution through exact algorithms. In our research, we introduce an enhanced version of the discrete fireworks algorithm, specifically tailored to address various large-scale KP instances. This advancement involves a re-encoding of the original fireworks algorithm and its adaptation into a discrete format, thereby equipping it to effectively tackle discrete KP challenges. Furthermore, we have integrated a probability mechanism and a historical information learning mechanism into the algorithm. These innovations enable the algorithm to assimilate and utilize historical data, significantly augmenting its capability to explore solutions. This enhancement not only accelerates the algorithm's convergence but also improves its proficiency in identifying optimal solutions. Additionally, we have developed a dual-population strategy to optimize the balance between exploration and exploitation performances of the algorithm. Comparative experimental analyses demonstrate that our modified algorithm exhibits rapid convergence, successfully avoids entrapment in local optima, and efficiently resolves large-scale KPs within a reasonable time, outperforming other evolutionary algorithms.

Parameter-efficient tuning methods such as LoRA could achieve comparable performance to model tuning by tuning a small portion of the parameters. However, substantial computational resources are still required, as this process involves calculating gradients and performing back-propagation throughout the model. Much effort has recently been devoted to utilizing the derivative-free optimization methods to eschew the computation of gradients and showcase an augmented level of robustness in few-shot settings. In this paper, we prepend the low-rank modules into each self-attention layer of the model and employ two derivative-free optimization methods to optimize these low-rank modules at each layer alternately. Extensive results on various tasks and language models demonstrate that our proposed method achieves substantial improvement and exhibits clear advantages in memory usage and convergence speed compared to existing gradient-based parameter-efficient tuning and derivative-free optimization methods in few-shot settings.

Evolutionary and swarm intelligence-based black-box optimization algorithms have been widely utilized to solve an increasing number of machine learning problems in recent years. Despite the effectiveness of these algorithms, the low sample efficiency has become one of the main obstacles due to their sampling-based nature. Thus, it is of great interest to improve the sample efficiency of traditional black-box optimization algorithms while keeping their merits. In this paper, the population-based fireworks algorithm (FWA) is equipped with policy gradient (PG) information, leading to the development of a novel and effective algorithm called FWA-RL. The main idea is to enhance the explosion operator with policy gradient-guided explosion and conduct firework cooperation using distillation-based cooperation. Experimental studies show the proposed algorithm can outperform state-of-the-art pure reinforcement learning (RL) algorithms and other hybrid evolutionary reinforcement learning algorithms (EARL) on the standard MuJoCo benchmark suite for continuous control. With efficient parallel implementation, FWA-RL may serve as a competitive algorithm for solving more even complex problems.

Multi-agent Reinforcement Learning (MARL) has drawn wide attention since a bunch of real-world complex scenes can be abstracted as Multi-Agent Systems. In order to solve the non-local training objective problem in shared reward environments, value-decomposition-based methods were proposed. Most of them introduce priori Individual-Global-Max (IGM) and value-decomposition constraints. Some attempts tune the value-decomposition constraints to achieve a better performance. However, IGM constraint, the fundamental assumption of value-decomposition methods, is adopted in most value-decomposition methods, which may lead to poor exploration in some situations. To deal with this problem, a novel algorithm called Mutual-guided Multi-agent Actor-Critic (MugAC) is proposed in this paper. MugAC, inspired by the core idea of evolutionary computation, imposes a joint-action pool, from which a joint-action is selected by the critic to interact with the environment and as a training objective of the actor. The training paradigm of MugAC provides an off-policy training for actor-critic, making the sample efficiency higher than that of traditional actor-critic methods in MARL. We evaluate our method against the state-of-the-art methods in StarCraft micromanagement. Experimental results show that MugAC outperforms other methods in various scenarios of widely adopted StarCraft Multi-Agent Challenge (SMAC).

In this article, we investigate how multiple agents learn to coordinate to form efficient exploration in reinforcement learning. Though straightforward, independent exploration of the joint action space of multiple agents will become exponentially more difficult as the number of agents increases. To tackle this problem, we propose feudal latent-space exploration (FLE) for multi-agent reinforcement learning (MARL). FLE introduces a feudal commander to learn a low-dimensional global latent structure that instructs multiple agents to explore coordinately. Under this framework, the multi-agent policy gradient (PG) is adopted to optimize both the agent policy and latent structure end-to-end. We demonstrate the effectiveness of this method in two multi-agent environments that need explicit coordination. Experimental results validate that FLE outperforms baseline MARL approaches that use independent exploration strategy in terms of mean rewards, efficiency, and the expressiveness of coordination policies.

In real-world applications, a specific class of multiobjective optimization problems, such as the cloud service allocation problem (CSAOPs), possess the characteristic of variable-length and mixed variables, termed as variable multiobjective optimization problems (VMMOPs). Unfortunately, little research has been reported to solve them. To fill the gap, we propose a tailored enhanced decomposition-based algorithm to handle the VMMOPs. Specifically, a variable-length coding structure is designed to flexibly represent the solutions of VMMOPs. In order to facilitate the solution generation, a simple dimensionality incremental learning strategy is developed to choose representative solutions for the training of two learning models. The one is the fast-clustering-based histogram model, which is built for the sampling of solutions in the continuous decision space, while the other one is the incremental learning-based histogram model, designed to sample solutions in discrete decision space. Following the traditional constructor of the DTLZ test suite and the features of CSAOPs, we present a test suite of VMMOPs for the verification of the performance of the methods in handling VMMOPs. Experimental results on a number of benchmark problems and two real CSAOPs have shown the effectiveness and competitiveness of the proposed method in handling VMMOPs.

Previous class imbalance learning methods are mostly grounded on the assumption that all training data have been labeled, however, is impractical in many real-world applications. The limited amount of labeled instances may produce a classifier with poor generalization. To address the issue, a transfer weighted extreme learning machine (TWELM) classifier is proposed, with the purpose of extracting knowledge from other domains to improve the classification performance of a classifier in a limited labeled target domain. To be specific, a well-tuned weighted extreme learning machine classifier is first learned from source data that has been completely labeled. Subsequently, another extreme learning machine classifier is obtained from the limited labeled target domain data to preserve the target domain structural knowledge and the decision boundary information. Finally, the target classifier is optimized by minimizing the outputs of the two classifiers on unlabeled target data. Experimental results on real-world data sets show that TWELM outperforms existing algorithms on classification accuracy and computation cost.

In multiagent reinforcement learning (MARL), it is crucial for each agent to model the relation with its neighbors. Existing approaches usually resort to concatenate the features of multiple neighbors, fixing the size and the identity of the inputs. But these settings are inflexible and unscalable. In this article, we propose an attentive relational encoder (ARE), which is a novel scalable feedforward neural module, to attentionally aggregate an arbitrary-sized neighboring feature set for state representation in the decentralized MARL. The ARE actively selects the relevant information from the neighboring agents and is permutation invariant, computationally efficient, and flexible to interactive multiagent systems. Our method consistently outperforms the latest competing decentralized MARL methods in several multiagent tasks. In particular, it shows strong cooperative performance in challenging StarCraft micromanagement tasks and achieves over a 96% winning rate against the most difficult noncheating built-in artificial intelligence bots.

Fireworks algorithm (FWA) as an efficient and robust swarm intelligence algorithm can successfully deal with complex multi-modal problems. In this paper, a novel new explosion operator called exponentially decaying explosion is proposed to enhance the local search ability of fireworks algorithm based on the principle of utilizing more information. The proposed method takes the idea of guided mutation a step further and dismantled the explosion process into an exponentially decaying series of guided explosion. The FWA variant with this explosion operator is called exponentially decaying fireworks algorithm (EDFWA). Theoretical analysis proved the superiority of EDFWA in terms of information utilization ratio compared with GFWA. Experimental results showed that EDFWA not only surpassed LoTFWA in low dimensional situations, but also exhibited powerful searching capability on 1000 dimensional high dimensional problems compared with multiple representative optimizers specially designed for large-scale problems.

Fireworks Algorithm (FWA) is a special swarm intelligent optimization algorithm, which controls multiple subgroups of the population to search collaboratively. Instead of assigning fireworks to different local areas, we propose the multiscale collaborative firework algorithm (MSCFWA) which helps fireworks to search at coordinated scales. Since the collaboration of search scales is accomplished by restarting or adjusting fireworks whose local search are not making meaningful progress, fireworks in MSCFWA are able to exploit different local areas independently or cooperate in the same local area with different search scales. Experimental results show that the proposed strategy stably improved the overall optimization performance of fireworks algorithm on the benchmark functions of the CEC'13 competition significantly. It also shows outstanding efficiency compared with typical swarm intelligence optimization algorithms and evolutionary algorithms.

Due to the strong connections among stocks, the information valuable for forecasting is not only included in individual stocks, but also included in the stocks related to them. These inter-correlations can provide invaluable information to be further leveraged to improve the overall forecasting performances. However, most previous works focus on the forecasting task of one single stock, which easily ignore the valuable information in others. Therefore, in this paper, we propose a jointly forecasting approach to process the time series of multiple related stocks simultaneously, using multi-task learning framework. In particular, this framework processes multiple forecasting tasks of different stocks simultaneously by sharing the information extracted based on latent inter-correlations. Meanwhile, each stock has their private encoding networks to keep their own information. Moreover, to dynamically balance private and shared information, we propose an attention based method, called Shared-private Attention, to optimally combine the shared and private information of stocks, which is inspired by the idea of Capital Asset Pricing Model (CAPM). Experimental results on the datasets of both stock and other domains demonstrate the proposed method can outperform other methods in forecasting performance.

Automatic grammatical error detection system is useful for language learners to identify whether the texts written by themselves have errors. Researches have paid more attention on different models to deal with this task, various approaches have been proposed and better results have been achieved compare with rules base methods. It is known that artificially generated incorrect texts can further improve the performance of grammatical error correction and pairwise training is necessary for many recommendation algorithms. We incorporating these two techniques together to solve the error detection task with pre-trained words embeddings from BERT in this paper. It is the first work that adopt pairwise training with pairs of samples to detect grammatical errors since all previous work were training models with batches of samples piontwisely. Pairwise training is useful for models to capture the differences within the pair of samples, which are intuitive useful for model to distinguish errors. Extensive experiments have been carried out to prove the effectiveness of pairwise training mechanism. The experimental results shown that the proposed method can achieve the state of the art performance on four different standard benchmarks. With the help of data augmentation and filtering, the value of F0:5can be further improved. The overall improvements among the four test set are around 2.5% which demonstrate the generality of pairwise training for datasets from differen domains.

This paper proposes an ensemble decision approach which combines global and local features of e-mails together to detect spam effectively. In the proposed method, a special feature construction method named term space partition (TSP) is utilized to divide the whole term space into several subspaces and adopt different feature construction strategies on each of them, respectively. This method can make each term play a distinct and important role when conducting detection. This method is utilized and extended by introducing the sliding window technique to extract local features from e-mails. The global classifier and local classifiers are constructed on a global feature vector set and local feature vector sets, respectively, and together make the ensemble decision by adopting the voting technique. The principles of the TSP-based approach and mechanism of the ensemble decision method are presented in detail. Five different and standard benchmark corpora are applied to experiments for performance evaluation of this proposed method. Comprehensive experimental results show that the proposed method brings significant performance improvement and better robustness on the basis of the TSP-based approach. In addition, the proposed method outperforms the current prevalent and state-of-the-art approaches, especially when a comprehensive consideration of performance, efficiency, and robustness is taken. This endows it with flexible capability and adaptivity in the real-world applications.

Semisupervised text classification has attracted much attention from the research community. In this paper, a novel model, the semisupervised sequential variational autoencoder (SSVAE), is proposed to tackle this problem. By treating the categorical label of unlabeled data as a discrete latent variable, the proposed model maximizes the variational evidence lower bound of the data likelihood, which implicitly derives the underlying label distribution for the unlabeled data. Analytical work indicates that the autoregressive nature of the sequential model is the crucial issue that renders the vanilla model ineffective. To remedy this, two types of decoders are investigated in the SSVAE model and verified. In addition, a reweighting approach is proposed to circumvent the credit assignment problem that occurs during the reconstruction procedure, which can further improve performance for sparse text data. Experimental results show that our method significantly improves the classification accuracy compared with other modern methods.

We propose a distance-based exclusive strategy to extend fireworks algorithm as a niche method to find out multiple global/local optima. This strategy forms sub-groups consisting of a firework individual and its generated spark individuals, each sub-group is guaranteed not to search overlapped areas each other. Finally, firework individuals are expected to find different global/local optima. The proposed strategy checks the distances between a firework and other fireworks which fitness is better than that of the firework. If the distance between two firework individuals is shorter than the sum of their searching radius, i.e. amplitude of firework explosions, these two firework individuals are considered to search overlapped area. Thus, the poor firework is removed and replaced by its opposite point to track multiple optima. To evaluate the performance of our proposed strategy, enhanced fireworks algorithm (EFWA) is used as a baseline algorithm and combined with our proposal. We design a controlled experiment, and run EFWA and (EFWA + our proposal) on 8 benchmark functions from CEC 2015 test suite, that is dedicated to single objective multi-niche optimization. The experimental results confirmed that the proposed strategy can find multiple different optima in one trial run.

Although learning-based methods such as reinforcement learning have been applied to multi-agent systems design successfully, it is still difficult to learn efficient coordinated policies for agents in partially observed environment settings. Centralized learners contain much more information, but add more complexity, while independent learners suffer from partial observation. To address these problems, we propose a directed multi-agent actor-critic algorithm to directly learn the coordinated policy from experience. The directed critic model can obtain all information including global information and actions, which provides effective learning signals for distributed learning actors. We take Multi-Agent Catching Game as the test scenario, where the task is to coordinate multiple moving paddles to catch balls dropping from the top of the screen. We perform several experimental evaluations and show that our method leads to superior results in learning performance, coordination effect and scalability, compared with both centralized and independent learning approach.

In most metaheuristic algorithms, the updating process fails to make use of information available from individuals in previous iterations. If this useful information could be exploited fully and used in the later optimization process, the quality of the succeeding solutions would be improved significantly. This paper presents our method for reusing the valuable information available from previous individuals to guide later search. In our approach, previous useful information was fed back to the updating process. We proposed six information feedback models. In these models, individuals from previous iterations were selected in either a fixed or random manner. Their useful information was incorporated into the updating process. Accordingly, an individual at the current iteration was updated based on the basic algorithm plus some selected previous individuals by using a simple fitness weighting method. By incorporating six different information feedback models into ten metaheuristic algorithms, this approach provided a number of variants of the basic algorithms. We demonstrated experimentally that the variants outperformed the basic algorithms significantly on 14 standard test functions and 10 CEC 2011 real world problems, thereby, establishing the value of the information feedback models.

Fireworks algorithm (FWA) is a newly proposed swarm intelligence optimization method. It simulates the fireworks explosion process to search for the best location of sparks and has demonstrated good performance in many continuous optimization problems. In this paper, we apply FWA to the travel salesman problem (TSP), a classical discrete optimization problem. We propose a discrete fireworks algorithm for TSP by combining the general framework of FWA and current ideas for solving the TSP. We call it DFWA-TSP. In DFWA-TSP, 2-opt and 3-opt edge exchange heuristic are implemented as the basic explosion operation in FWA. An adaptive strategy is designed to decide the explosion amplitude. A particular mutation method based on insertion is also used to cover the shortage of edge exchange and a new selection method based on the quality of fireworks is adopted to pick up good fireworks efficiently. Various experiments on both TSPLIB and synthetic data have been made to compare the performance of our algorithm with current heuristic methods for TSP, such as genetic algorithm and ant colony system algorithm. We conclude that our algorithm out-performs these algorithms, especially on large-scale cases.

Conditional text generation is a fundamental task in natural language generation. Traditional conditional generative models build conditional probability distributions over the given labels. However, categorical label information is usually very abstract, e.g., sentiment, and it is difficult to be disentangled from the content. Therefore, instead of generating text by modeling conditional probability distribution, we propose a novel text generation method TextDream through searching in the semantic space. Specifically, in this method, a random text seed is initially given and the new text is generated by local search operation. The generation procedure is guided by a fitness function, typically a classification model. Text with higher fitness will be preserved. This procedure loops until the qualified solution is found. Experimental results show that our method is able to generate more diverse text compared with advanced conditional generative models.

Fireworks algorithm(FWA), which is proposed for global optimization of complex function, becomes a hot spot in optimization field recently, caused by its competitive performance. Boundary handling for FWA, which maps the out-of-bound sparks into feasible space, is critical for its convergence efficiency. However, random mapping rule, which is widely used for boundary handling, always caused computing resource waste, especially for high-dimensional optimization. In this paper, we propose three novel mapping rules to speed up large scale optimization of FWA. Meanwhile, to evaluate the effectiveness of the new rules, we compare them by representative nine benchmark functions on different dimensionality scale. Experimental results indicate that the mirror rule which we proposed, achieve superior performance for most optimization functions.

The seven papers in this special section were presented at the ICSI 2016 Conference. These articles are primarily dealing with either novel bioinspired swarm intelligence algorithms and their improvements aswell as some practical applications inmulti-objective optimization, network community detection, curve fitting, and swarm robotics, etc.

Real-world optimization problems are usually multimodal which require optimization algorithms to keep a balance between exploration and exploitation. Therefore, multimodal optimization is one of the main opportunities as well as one of the main challenges for evolutionary algorithms. In this paper, a loser-out tournament-based fireworks algorithm (LoTFWA) is proposed for solving multimodal optimization problems. The search manner of the conventional fireworks algorithm (FWA) is based on the cooperation of several fireworks. While in the LoTFWA, we propose competition as a new manner of interaction, in which the fireworks are compared with each other not only according to their current status but also according to their progress rate. If the fitness of a certain firework cannot catch up with the best one with its current progress rate, it is considered a loser in the competition. The losers will be eliminated and reinitialized because it is vain to continue their search processes. Reinitializing these fireworks would greatly reduce the probability of being trapped in local minima for the algorithm. Experimental results show that the proposed algorithm is very powerful in optimizing multimodal functions. It not only outperforms previous versions of the FWA, but also outperforms several famous evolutionary algorithms.

With the rapid popularity of the Internet, a large amount of new malware is produced every day, while the traditional signature based malware detection algorithm is unable to detect such unseen malware. In recent years, many machine learning based algorithms have been proposed to detect new malware, and several of these algorithms are able to achieve quite good detection performance when supplied with plenty of training data. However, most of these algorithms just focus on how to improve the classification performance, while the robustness is not taken into consideration. This paper performs a detailed analysis on the robustness of four well-known machine learning based malware detection approaches, i.e. the DLL and API feature, the string feature, PE-Miner and the byte level N-Gram feature. We proposed two pretense approaches under which malware is able to pretend to be benign and bypass the detection algorithms. Experimental results show that the performances of these detection algorithms decline greatly under the pretense approaches. The lack of robustness makes these algorithms unable to be used in real world applications. In future works of machine learning based malware detection, researchers have to take the problem of robustness seriously.

Presents the introductory editorial for this issue of the publication.

This paper presents a cooperative framework for fireworks algorithm (CoFFWA). A detailed analysis of existing fireworks algorithm (FWA) and its recently developed variants has revealed that ( i) the current selection strategy has the drawback that the contribution of the firework with the best fitness (denoted as core firework) overwhelms the contributions of all other fireworks (non-core fireworks) in the explosion operator, ( ii) the Gaussian mutation operator is not as effective as it is designed to be. To overcome these limitations, the CoFFWA is proposed, which significantly improves the exploitation capability by using an independent selection method and also increases the exploration capability by incorporating a crowdness-avoiding cooperative strategy among the fireworks. Experimental results on the CEC2013 benchmark functions indicate that CoFFWA outperforms the state-of-the-art FWA variants, artificial bee colony, differential evolution, and the standard particle swarm optimization SPSO2007/SPSO2011 in terms of convergence performance.

The fireworks algorithm (FWA) is a competitive swarm intelligence algorithm which has been shown to be very useful in many applications. In this paper, a novel guiding spark (GS) is introduced to further improve its performance by enhancing the information utilization in the FWA. The idea is to use the objective function's information acquired by explosion sparks to construct a guiding vector (GV) with promising direction and adaptive length, and to generate an elite solution called a GS by adding the GV to the position of the firework. The FWA with GS is called the guided FWA (GFWA). Experimental results show that the GS contributes greatly to both exploration and exploitation of the GFWA. The GFWA outperforms previous versions of the FWA and other swarm and evolutionary algorithms on a large variety of test functions and it is also a useful method for large scale optimization. The principle of the GS is very simple but efficient, which can be easily transplanted to other population-based algorithms.

The fireworks algorithm (FWA) is a newly proposed nature-inspired swarm intelligence algorithm. In this paper, two novel mechanisms are proposed to enhance the exploration capability by means of interaction among fireworks in the FWA. The dynamic resource allocation allows the algorithm to allocate the resource (the number of sparks) adaptively according to the search results and fitness ranking of the fireworks. The fitness-based crowdedness-avoiding strategy is proposed to improve the diversity of the fireworks population by sharing the fitness information among the fireworks. Experimental results on a large variety of test functions indicate that the proposed measures significantly improve the performance of the FWA, especially on complex objective functions.

Since fireworks algorithm (FWA) debuted in 2010, a dozen proposals of improvement for FWA had been published in an effort to enhance, refine and optimize accuracy while minimizing calculation speed and volume. In this paper, we introduce the covariance mutation operator into cooperative framework fireworks algorithm (CoFFWA) to create cooperative framework fireworks algorithm with covariance mutation (CoFFWA-CM) in order to solve the Congress on Evolutionary Computation (CEC 2016) competition functions on single objective optimization. The experimental results are calculated on all 10-, 30-, 50- and 100-dimensional functions.

In the field of deep neural networks, several generative methods have been proposed to address the challenges from generative and discriminative tasks, e.g., natural language process, image caption and image generation. In this paper, a conditional recurrent variational autoencoder is proposed for multi-digit image synthesis. This model is capable of generating multi-digit images from the given number sequences and retaining the generalisation ability to recover different types of background. Our method is evaluated on SVHN dataset and the experimental results show it succeeds to generate multi-digit images with various styles according to the given sequential inputs. The generated images can also be easily identified by both human beings and convolutional neural networks for digit classification.

The nearest neighbor (NN) classifier suffers from high time complexity when classifying a test instance since the need of searching the whole training set. Prototype generation is a widely used approach to reduce the classification time, which generates a small set of prototypes to classify a test instance instead of using the whole training set. In this paper, particle swarm optimization is applied to prototype generation and two novel methods for improving the classification performance are presented: 1) a fitness function named error rank and 2) the multiobjective (MO) optimization strategy. Error rank is proposed to enhance the generation ability of the NN classifier, which takes the ranks of misclassified instances into consideration when designing the fitness function. The MO optimization strategy pursues the performance on multiple subsets of data simultaneously, in order to keep the classifier from overfitting the training set. Experimental results over 31 UCI data sets and 59 additional data sets show that the proposed algorithm outperforms nearly 30 existing prototype generation algorithms.

Inspired by the collective behavior of natural swarm, swarm intelligence algorithms (SIAs) have been developed and widely used for solving optimization problems. When applied to complex problems, a large number of fitness function evaluations are needed to obtain an acceptable solution. To tackle this vital issue, graphical processing units (GPUs) have been used to accelerate the optimization procedure of SIAs. Thanks to their inherent parallelism, SIAs are very suitable for parallel implementation under the GPU platform which have achieved a great success in recent years. This paper presents a comprehensive review of GPU-based parallel SIAs in accordance with a newly proposed taxonomy. Critical concerns for the efficient parallel implementation of SIAs are also described in detail. Moreover, novel criteria are also proposed to evaluate and compare the parallel implementation and algorithm performance universally. The rationality and practicability of the proposed optimization methodology and criteria are verified by careful case study. Finally, our opinions and perspectives on the trends and prospects on the relatively new research domain are also presented for future development.

Inversion is a critical and challenging task in geophysical research. Geophysical inversion can be formulated as an optimization problem to find the best parameters whose forward synthesis data most fit the observed data. The inverse problems are usually highly non-linear, multi-modal as well as ill-posed, so conventional optimization algorithms cannot handle it very efficiently. In the past decades, genetic algorithm (GA) and its many variants are widely applied to inverse problems and achieve great success. Swarm intelligence algorithms are a family of global optimizers inspired by swarm phenomena in nature, and have shown better performance than GA for diverse optimization problems. However, swarm intelligence algorithms are not utilized for geophysical inversion problems until recently and only limited number of works are reported. In this paper, we try to apply two swarm intelligence algorithms, Particle Swarm Optimization (PSO) and Fireworks Algorithm (FWA), to the regional seismic waveform inversion. To explore the advantages and disadvantages of swarm intelligence algorithms over GA, synthetic experiments are conducted by using these two swarm intelligence algorithm and several GA variants as well as Differential Evolution (DE). The experimental results show that, both swarm intelligence algorithms outperform the widely used GA, DE, and the models estimated by swarm intelligence algorithms are closer to the true solution. The promising results imply that swarm intelligence algorithms are a potentially more powerful tool for inversion problems.

In this paper, a novel orienting mutation operator is designed to improve the performance of the fireworks algorithm, which is a recently proposed swarm intelligence algorithm for optimization. For each firework, the orienting mutation operator creates a new promising solution by adding to the firework a proper step size towards the local minimal point. By making use of the ready-made information of the optimization function, the orienting mutation operator enhances the local search ability of the algorithm. Its principles are analyzed and its effect is tested experimentally to show that it is a significant improvement.

Fireworks Algorithm(FWA) is a recently developed swarm intelligence algorithm for single objective optimization problems which gains very promising performances in many areas. In this paper, we extend the original FWA to solve multi-objective optimization problems with the help of S-metric. The S-metric is a frequently used quality measure for solution sets comparison in evolutionary multi-objective optimization algorithms (EMOAs). Besides, S-metric can also be used to evaluate the contribution of a single solution among the solution set. Traditional multi-objective optimization algorithms usually perform a (μ + 1) strategy and update the external archive one by one, while the proposed S-metric based multi-objective fireworks algorithm(S-MOFWA) performs a (μ + μ) strategy, thus converging faster to a set of pareto solutions by three steps: 1)Exploring the solution space by mimicking the explosion of fireworks; 2)Performing a simple selection strategy for choosing the next generation of fireworks according to their S-metric; 3)Utilizing an external archive to maintain the best solution set ever found, with a new archive definition and a novel updating strategy, which can update the archive with μ solutions in a single process. The experimental results on benchmark functions suggest that the proposed S-MOFWA outperforms three other well-known algorithms, i.e. NSGA-II, SPEA2 and PESA2 in terms of the convergence measure and covered space measure.

As a revolutionary swarm intelligence algorithm, fireworks algorithm (FWA) is designed to solve optimization problems. In this paper, the dynamic fireworks algorithm with covariance mutation (dynFWACM) is proposed. After applying the explosion operator, the mutation operator is introduced, which calculates the mean value and covariance matrix of the better sparks and produces sparks according with Gaussian distribution. DynFWACM is compared with the most advanced fireworks algorithms to proof its effectiveness. In addition, 15 functions of CEC 2015 competition on learning based real-parameter single objective optimization are used to test the performance of our new proposed algorithm. The experimental results show that dynFWACM outperforms both AFWA and dynFWA, as well as the experimental results of the 15 functions given.

Fireworks algorithm is a novel swarm intelligence algorithm for solving optimization problems - the latest versions include the adaptive fireworks algorithm and the dynamic fireworks algorithm. However, the mutation operator in the former algorithm was ineffective, whereas there was no mutation operator available in the latter algorithm. In this paper, a mutation operator is proposed, dubbed as the covariance mutation (CM) operator. The CM operator utilizes the information of the sparks with better fitness values to generate potential sparks for finding the optima of functions with higher possibility. Therefore, we proposed the fireworks algorithm with covariance mutation (FWACM) and compared it with the most advanced fireworks algorithms. The experimental results show that FWACM is a significant improvement for fireworks algorithms.

Fireworks algorithm (FWA) is one swarm intelligence algorithm proposed in 2010, which takes the inspiration from the firework explosion process. Compared with other meta-heuristic algorithms, FWA presents a cooperative explosive search manner. In the explosive search manner, the explosion amplitudes, explosion sparks' numbers and explosion dimension selection methods play the key roles for its successful implementation. In this paper, the performance analyses of the different explosion dimension number strategies in FWA and its variants are presented at first, then the exponentially decreased explosion dimension number strategy is introduced for the most recent dynamic search fireworks algorithm (dynFWA), called ed-dynFWA, to enhance its local search ability. To validate the performance of ed-dynFWA, it is used to participate in the CEC 2015 competition for solving learning based optimization problems.

In the field of spam detection, concentration methods have been proposed for feature construction in recent years, which convert emails into fixed length feature vectors. This paper presents a novel method aiming to break through the limit of feature vector's length. Specifically, the method uses a fixed-length sliding window to divide each email into several sections. The number of sections depends on the length of each email. Consequently, length of feature vectors varies from each other and this paper names them variable length concentrations (VLC). This method can acquire adaptive feature vectors according to different lengths of emails. However, general classifiers are not suitable for this kind of feature vectors, because they are not able to handle fixed-length inputs. As a result, this paper applies recurrent neural networks (RNNs), whose inputs are not restricted by the length, to achieve spam detection. Recall, precision, accuracy and F1 measure are taken to evaluate the method's performance. Experimental results on the classic corpora, PU1, PU2, PU3 and PUA, show that VLC performs significantly better than previously proposed methods, which provides support to the effectiveness of our method.

Intelligent optimization algorithms are very effective to tackle complex problems that would be difficult or impossible to solve exactly. A key component within these algorithms is the random number generators (RNGs) which provide random numbers to drive the stochastic search process. Much effort is devoted to develop efficient RNGs with good statistical properties, and many highly optimized libraries are ready to use for generating random numbers fast on both CPUs and other hardware platforms such as GPUs. However, few study is focused on how different RNGs can effect the performance of specific intelligent optimization algorithms. In this paper, we empirically compared 13 widely used RNGs with uniform distribution based on both CPUs and GPUs, with respect to algorithm efficiency as well as their impact on Particle Swarm Optimization (PSO). Two strategies were adopted to conduct comparison among multiple RNGs for multiple objectives. The experiments were conducted on well-known benchmark functions of diverse landscapes, and were run on the GPU for the purpose of accelerating. The results show that RNGs have very different efficiencies in terms of speed, and GPU-based RNGs can be much faster than their CPU-based counterparts if properly utilized. However, no statistically significant disparity in solution quality was observed. Thus it is reasonable to use more efficient RNGs such as Mersenne Twister. The framework proposed in this work can be easily extended to compare the impact of non-uniformly distributed RNGs on more other intelligent optimization algorithms.

In this paper, firstly, the amplitude used in the Enhanced Fireworks Algorithm (EFWA) is analyzed and its lack of adaptability is revealed, and then the adaptive amplitude method is proposed where amplitude is calculated according to the already evaluated fitness of the individuals adaptively. Finally, the Adaptive Fireworks Algorithm (AFWA) is proposed, replacing the amplitude operator in EFWA with the new adaptive amplitude. Some theoretical analyses are made to prove the adaptive explosion amplitude a promising method. Experiments on CEC13's 28 benchmark functions are also conducted in order to illustrate the performance and it turns out that the AFWA where adaptive amplitude is adopted outperforms significantly the EFWA and meanwhile the time consumed is not longer. Moreover, according to experimental results, AFWA performs better than the Standard Particle Swarm Optimization (SPSO).

Fireworks Algorithm (FWA) is a new proposed optimization technique based on swarm intelligence. In FWA, the algorithm generates the explosion sparks and Gaussian mutation sparks by the explosion operator and Gaussian mutation operator to search the global optimum in the problem space. FWA has been applied in various fields of practical optimization problems and gains great success. However, its convergence property has not been analyzed since it has been provided. Same as other swarm intelligence (SI) algorithms, the optimization process of FWA is able to be considered as a Markov process. In this paper, a Markov stochastic process on FWA has been defined, and is used to prove the global convergence of FWA while analyzing its time complexity. In addition, the computation of the approximation region of expected convergence time of FWA has also been given.

The idea of fireworks algorithm (FWA) is inspired by the fireworks explosion in the sky at night. When a firework explodes, a shower of sparks appear around it. In this way, the adjacent area of the firework is searched. By controlling the amplitude of the explosion, the ability of local search for FWA is guaranteed. The way of fireworks algorithm searching the surrounding area can be further improved by differential mutation operator, forming an algorithm called FWA-DM. In this paper, the benchmark suite in the competition of congress of evolutionary computation (CEC) 2014 is used to test the performance of FWA-DM.

We propose an improved version of the recently developed Enhanced Fireworks Algorithm (EFWA) based on an adaptive dynamic local search mechanism. In EFWA, the explosion amplitude (i.e., search area around the current location) of each firework is computed based on the quality of the firework's current location. This explosion amplitude is limited by a lower bound which decreases with the number of iterations in order to avoid the explosion amplitude to be [close to] zero, and in order to enhance global search abilities at the beginning and local search abilities towards the later phase of the algorithm. As the explosion amplitude in EFWA depends solely on the fireworks' fitness and the current number of iterations, this procedure does not allow for an adaptive optimization process. To deal with these limitations, we propose the Dynamic Search Fireworks Algorithm (dynFWA) which uses a dynamic explosion amplitude for the firework at the currently best position. If the fitness of the best firework could be improved, the explosion amplitude will increase in order to speed up convergence. On the contrary, if the current position of the best firework could not be improved, the explosion amplitude will decrease in order to narrow the search area. In addition, we show that one of the EFWA operators can be removed in dynFWA without a loss in accuracy - this makes dynFWA computationally more efficient than EFWA. Experiments on 28 benchmark functions indicate that dynFWA is able to significantly outperform EFWA, and achieves better performance than the latest SPSO version SPSO2011.

In this paper, we consider the target searching problems with a new type of the object: decoys which can be sensed exactly as targets but cannot be collected by the robots. In real-life applications, decoys are very common especially for swarm robots whose hardware should be designed as simple and cheap as possible. This inevitably brings errors and mistakes in the sensing results and the swarm may mistakenly sense certain kinds of environment objects as the target they are looking for. We proposed a simple cooperative strategy to solve this problem, comparing with a non-cooperative strategy as the baseline. The strategies work with other searching algorithms and provide schemes for avoiding decoys. Simulation results demonstrate that the cooperative strategy shares almost the same computation overload yet has better performance in iterations and especially visited times of decoys. The strategy shows great adaptiveness to large scale problems and performs better when more decoys or robots exist in the simulation.

Fireworks algorithm (FWA) is a newly proposed swarm intelligence algorithm, which is used to solve optimization problems. However, the interaction of fireworks in FWA is not sufficient. In this paper, the differential mutation operator is introduced to improve the interaction mechanism of enhanced FWA (EFWA), which is the latest version of FWA. Extensive experiments on 30 benchmark functions were conducted to test the performance of the new algorithm named enhanced fireworks algorithm with differential mutation (FWA-DM). Experimental results have shown that differential mutation operator is able to improve EFWA.

In this paper, an improved group explosion strategy (IGES) is proposed for searching multiple targets using a swarm of simple robots. The strategy is based on our previous work which has several shortcoming especially when number of targets and robots are large. IGES is simple and fast with great adaptability and only one parameter. The simulation results demonstrate that IGES has great efficiency in all aspects including searching time, energy consumption and computation overload. IGES also shows great stability and adaptiveness in both small and large scale problems.

In this paper, we present an improved version of the recently developed Fireworks Algorithm (FWA) based on several modifications. A comprehensive study on the operators of conventional FWA revealed that the algorithm works surprisingly well on benchmark functions which have their optimum at the origin of the search space. However, when being applied on shifted functions, the quality of the results of conventional FWA deteriorates severely and worsens with increasing shift values, i.e., with increasing distance between function optimum and origin of the search space. Moreover, compared to other metaheuristic optimization algorithms, FWA has high computational cost per iteration. In order to tackle these limitations, we present five major improvements of FWA: (i) a new minimal explosion amplitude check, (ii) a new operator for generating explosion sparks, (iii) a new mapping strategy for sparks which are out of the search space, (iv) a new operator for generating Gaussian sparks, and (v) a new operator for selecting the population for the next iteration. The resulting algorithm is called Enhanced Fireworks Algorithm (EFWA). Experimental evaluation on twelve benchmark functions with different shift values shows that EFWA outperforms conventional FWA in terms of convergence capabilities, while reducing the runtime significantly.

In this paper, a group explosion strategy (GES) is proposed for searching multiple targets in obstructive environments using a swarm of simple robots. Considering the limited abilities of on-board sensors, fitness values detected by the robots are discrete in the problem. The strategy introduces schemes from the explosion phenomenon in nature and the whole swarm is self-adaptively divided into small groups which search for targets independently. GES takes the advantages of quick convergence from intra-group cooperation as well as searching multiple targets in parallel from inter-group cooperation. The simulation results demonstrate that GES has great efficiency in energy consumption and targets collecting benefitted from cooperation among robots. GES also shows great stability in obstructive environments.

Swarm intelligence algorithms have been widely used to solve difficult real world problems in both academic and engineering domains. Thanks to the inherent parallelism, various parallelized swarm intelligence algorithms have been proposed to speed up the optimization process, especially on the massively parallel processing architecture GPUs. However, conventional swarm intelligence algorithms are usually not designed specifically for the GPU architecture.They neither can fully exploit the tremendous computational power of GPUs nor can extend effectively as the problem scales go large. To address this shortcoming, a novel GPU-based Fireworks Algorithm (GPU-FWA) is proposed in this paper. In order to fully leverage GPUs' high performance, GPU-FWA modified the original FWA so that it is more suitable for the GPU architecture. An implementation of GPU-FWA on the CUDA platform is presented and then tested on a suite of well-known benchmark optimization problems. We extensively evaluated GPU-FWA and compared it with FWA and PSO, with respect to both running time and solution quality, on a state-of-the-art commodity Fermi GPU.Experimental results demonstrate that GPU-FWA generally outperforms both FWA and PSO, and enjoys a significant speedup as high as 200x, compared to the sequential version of FWA and PSO running on an up-to-date CPU. GPU-FWA also enjoys the advantages of being easy to implement and scalable.

The emergence of Web services has changed the Internet a lot, and greatly facilitated the development of service based software systems. How to select appropriate services and compose them according to given context to satisfy a user's requirement is a big challenge. This paper proposes a novel Genetic Algorithm (GA) method to synthesis web services in a context-aware environment. We first present a context space model to illustrate both contexts and services in a formal way, we utilize GA to compose context-aware services according to users' preference. We transform the problem of service composition to a multi-objective optimization problem. To resolve the conflict and dependencies among services in GA process, we propose a service similarity tree (SST) model to measure the similarity between services. Finally, we design a simulation experiment to evaluate our method. The experiment result shows that our method is a promising one to solve service composition problem in a context-aware environment.

This paper proposes a novel feature construction approach based on term space partition (TSP) aiming to establish a mechanism to make terms play more sufficient and rational roles in email categorization. Dominant terms and general terms are separated by performing a vertical partition of the original term space with respect to feature selection metrics, while spam terms and ham terms are separated by a transverse partition with respect to class tendency. Strategies for constructing discriminative features, named term ratio and term density, are designed on corresponding subspaces. Motivation and principle of the TSP approach is presented in detail, as well as the implementation. Experiments are conducted on five benchmark corpora using cross-validation to evaluate the proposed TSP approach. Comprehensive experimental results suggest that the TSP approach far outperforms the traditional and most widely used feature construction approach in spam filtering, which is named bag-of-words, in both performance and efficiency. In comparison with the heuristic and state-of-the-art approaches, namely CFC and LC, the proposed TSP approach shows obvious advantage in terms of accuracy and μ1measure, as well as high precision, which is warmly welcomed in real spam filtering. Furthermore, the TSP approach performs quite similar with CFC in efficiency of processing incoming emails, while much faster than LC. In addition, it is shown that the TSP approach cooperates well with both unsupervised and supervised feature selection metrics, which endows it with flexible capability in the real world.

This paper proposes a multi-resolution-concentration (MRC) based feature construction approach for spam filtering by progressively partitioning an email into local areas on smaller and smaller resolutions. The MRC approach depicts a dynamic process of gradual refinement in locating the pathogens by calculating concentrations of detectors on local areas, and is considered to be able to extract the position-correlated and process-correlated information from emails. Furthermore, A weighted MRC (WMRC) approach is presented by considering the different activity levels of detectors in calculation of concentrations. A generic structure of the MRC model, which mainly contains detector sets construction and multi-resolution concentrations calculation, is designed. The implementations of MRC and WMRC approaches are described in detail. Experiments are conducted on five benchmark corpora using cross-validation to evaluate the proposed MRC model. Comprehensive experimental results suggest that the MRC and WMRC approaches perform far better than the prevalent bag-of-words approach in both performance and efficiency. Compared with the concentration based feature construction approach and local-concentration based feature extraction approach, MRC and WMRC achieve higher accuracy and μ1measure, which demonstrates the effectiveness of the MRC model. In addition, it is shown that both the MRC and WMRC approaches cooperate well with variety of classification methods, which endows the MRC model with flexible capability in the real world.

To solve the spam problem, many statistical learning methods and AIS methods have been proposed and applied. In essence, statistical learning methods and AIS methods have quite different origins, and they try to find the solutions from distinct aspects. In recent works, we proposed several hybrid methods, which combined immune theory with statistical methods in spam filtering. In this paper, we briefly review and analyze these works and possible extensions, and demonstrate the rationality of building hybrid immune models for spam filtering. In addition, a generic framework of an immune based model is presented, and online implementation strategies are given. It is well demonstrated that how to apply the immune based model to building an intelligent email server.

In a context-aware pervasive environment, it is crucial to provide an efficient adaption mechanism to rapidly discover appropriate services and compose them to achieve user desired situation when context change. In this paper, we propose a GA(Genetic Algorithm) based novel service composition method in a context-aware environment. We first use Context Space model to represent context, situation, user preference and services in a unified formal way. Then we transform the service composition problem into a multi-objective optimization problem and utilize GA to find the optimal service composition based on user's preference. To affiliate the connection between context and services, we use a vector to represent each service to reflect which context types a service can change so as to utilize GA to find the best service composition. Confliction of services may happen during the GA process. In order to resolve the conflicts, we propose a SST(Service Similarity Tree) method to measure the similarity among services to find out the best alternative. Finally, we design and implement a simulation experiment to verify our method. The results shows that our method can leverage GA to synthesis appropriate services to achieve user desired goal in an efficient way.

Local learning employs locality adjusting mechanisms to give local function estimation for each query, while global learning tries to capture the global distribution characteristics of the entire training set. When fitting well with local characteristics of each individual region, the locality parameter may help local learning to improve performance. However, the real data distribution is impossible to get for a real-world problem, and thus an optimal locality is hard to get for each query. In addition, it is quite time-consuming to build an independent local model for each query. To solve these problems, we present strategies for estimating and tuning locality according to local distribution. Based on local distribution estimation, global learning and local learning are combined to achieve a good compromise between capacity and locality. In addition, multi-objective learning principles for the combination are also given. In implementation, a unique global model is first built on the entire training set based on empirical minimization principle. For each query, it is measured that whether the global model can well fit the vicinity space of the query. When an uneven local distribution is found, the locality of the model is tuned, and a specific local model will be built on the local region. To investigate the performance of hybrid models, we apply them to a typical learning problem-spam filtering, in which data are always found to be unevenly distributed. Experiments were conducted on five real-world corpora, namely PU1, PU2, PU3, PUA, and TREC07. It is shown that the hybrid models can achieve a better compromise between capacity and locality, and hybrid models outperform both global learning and local learning.

This paper presents an empirical study on the influence of approximation approaches on accelerating the fireworks algorithm search by elite strategy. In this study, we use three sampling data methods to approximate fitness landscape, i.e. the best fitness sampling method, the sampling distance near the best fitness individual sampling method and the random sampling method. For each approximation methods, we conduct a series of combinative evaluations with the different sampling method and sampling number for accelerating fireworks algorithm. The experimental evaluations on benchmark functions show that this elite strategy can enhance the fireworks algorithm search capability effectively. We also analyze and discuss the related issues on the influence of approximation model, sampling method, and sampling number on the fireworks algorithm acceleration performance.

In this paper, we propose an efficient algorithm for computing the Euclidean distance transform of two-dimensional binary image, called PBEDT (Perpendicular Bisector Euclidean Distance Transform). PBEDT is a two-stage independent scan algorithm. In the first stage, PBEDT computes the distance from each point to its closest feature point in the same column using one time column-wise scan. In the second stage, PBEDT computes the distance transform for each point by row with intermediate results of the previous stage. By using the geometric properties of the perpendicular bisector, PBEDT directly computes the segmentation by feature points for each row and each segment corresponding to one feature point. Furthermore, by using integer arithmetic to avoid time consuming float operations, PBEDT still achieves exact results. All these methods reduce the computational complexity significantly. Consequently, an efficient and exact linear time Euclidean distance transform algorithm is implemented. Detailed comparison with state-of-the-art linear time Euclidean distance transform algorithms shows that PBEDT is the fastest on most cases, and also the most stable one with respect to image contents.

In this paper, we propose a genetic programming algorithm to design the morphological image enhancement procedure. Given a group of morphological operations and logical operations as function set, this algorithm evolves to produce a rational procedure which can enhance the input images. A novel mechanism which combines the ground truth method and feature significance is brought forward to evaluate the performance of images enhanced by generated procedures. In each generation, the best fitted individuals are selected on the basis of fitness values, and some individuals participate in crossover or mutation with a probability. After each generation, this algorithm outputs the best individual. Seven morphological operations and five logical operations are used in this algorithm. Furthermore, the structuring elements of morphological operations are randomly generated and varied in the whole pattern space. These methods promote the expressive ability of generated procedures. Examined by the binary image feature extraction, the procedure generated by this algorithm is more accurate and intelligible than previous work. In the task of gray scale image enhancement, the generated procedure is applied to infrared finger vein images to enhance the region of interest. More accurate features are extracted and the accuracy of authentication is promoted.

Along with the evolution of computer viruses, the number of file samples that need to be analyzed has constantly increased. An automatic and robust tool is needed to classify the file samples quickly and efficiently. Inspired by the human immune system, we developed a local concentration based virus detection method, which connects a certain number of two-element local concentration vectors as a feature vector. In contrast to the existing data mining techniques, the new method does not remember exact file content for virus detection, but uses a non-signature paradigm, such that it can detect some previously unknown viruses and overcome the techniques like obfuscation to bypass signatures. This model first extracts the viral tendency of each fragment and identifies a set of statical structural detectors, and then uses an information-theoretic preprocessing to remove redundancy in the detectors’ set to generate ‘self’ and ‘nonself’ detector libraries. Finally, ‘self’ and ‘nonself’ local concentrations are constructed by using the libraries, to form a vector with an array of two elements of local concentrations for detecting viruses efficiently. Several standard data mining classifiers, including K -nearest neighbor (KNN), radial basis function (RBF) neural networks, and support vector machine (SVM), are leveraged to classify the local concentration vector as the feature of a benign or malicious program and to verify the effectiveness and robustness of this approach. Experimental results show that the proposed approach not only has a much faster speed, but also gives around 98% of accuracy.

Inspired from the biological immune system, we propose a local concentration (LC)-based feature extraction approach for anti-spam. The LC approach is considered to be able to effectively extract position-correlated information from messages by transforming each area of a message to a corresponding LC feature. Two implementation strategies of the LC approach are designed using a fixed-length sliding window and a variable-length sliding window. To incorporate the LC approach into the whole process of spam filtering, a generic LC model is designed. In the LC model, two types of detector sets are at first generated by using term selection methods and a well-defined tendency threshold. Then a sliding window is adopted to divide the message into individual areas. After segmentation of the message, the concentration of detectors is calculated and taken as the feature for each local area. Finally, all the features of local areas are combined as a feature vector of the message. To evaluate the proposed LC model, several experiments are conducted on five benchmark corpora using the cross-validation method. It is shown that the LC approach cooperates well with three term selection methods, which endows it with flexible applicability in the real world. Compared to the global-concentration-based approach and the prevalent bag-of-words approach, the LC approach has better performance in terms of both accuracy andF1measure. It is also demonstrated that the LC approach is robust against messages with variable message length.

As a new kind of vehicle with low fuel cost and low emissions, the hybrid electric vehicle (HEV) has been paid much attention in recent years. The key technique in the HEV is adopting the optimal control strategy for the best performance. As the premise, correct driving condition discrimination has an extremely important significance. This paper proposes an intelligent multifeature statistical approach to automatically discriminate the driving condition of the HEV. First, this approach periodically samples the driving cycle. Then, it extracts multiple statistical features and tests their significance by statistical analysis to select effective features. Afterward, it applies a support vector machine (SVM) and other machine-learning methods to intelligently and automatically discriminate the driving conditions. Compared with others, the proposed approach can compute fast and discriminate in real time during the whole HEV running mode. In our experiments, it reaches an accuracy value of 95%. As a result, our approach can completely mine the valid information from the data and extract multiple features that have clear meanings and significance. Finally, according to the prediction experiment by a neural network, the fitting experiment by the autoregressive moving average model, and the simulation results of the control strategy, it turns out that our proposed approach raises the efficiency of considerably controlling the HEV.

Inspired from Biological Immune System, we propose a local concentration based feature extraction (LC) approach for anti-spam. A general anti-spam model is built to incorporate the LC approach with term selection methods and classifiers. In the LC model, each message is divided into areas by a sliding window. At each area, a two-dimensional feature is constructed by calculating the concentrations of spam and legitimate email. Then all the features of each area are combined together as a whole feature vector. Several experiments are conducted on four benchmark corpora, by using 10-fold cross-validation. It is shown that the LC approach can extract the effective position correlated information from messages. Compared to the prevalent Bag-of-Words approach, the LC has better performance in terms of both accuracy and F1measure. Most significantly, the LC approach can reduce feature dimensionality greatly and has much faster speed.

This paper gives an applicable genetic programming(GP) approach to solve the binary image analysis and gray scale image enhancement problems. By showing a section of binary image and the corresponding goal image, this algorithm automatically produces a mathematic morphological operation sequence to transform the target into the goal. While the operation sequence is applied to the whole image, the objective of image analysis is achieved. With well-defined chromosome structure and evolution strategy, the effectiveness of evolution is promoted and more complex morphological operations can be composed in a short sequence. In addition, this algorithm is also applied to infrared finger vein gray scale images to enhance the region of interest. Whose effect is examined by an application of identity authentication, and the accuracy of authentication is promoted.

A novel particle swarm optimization with triggered mutation (PSO-TM) is presented in this paper for better performance. First, a technique is designed to evaluate the "health" of swarm. When the swarm is successively "unhealthy" for a certain number of iterations, uniform mutation is applied to the position of each particle in a probabilistic way. If the mutations produce worse particles, the memorized previous positions are retrieved as current positions of these particles, hence the normal evolution process of the swarm will not be fiercely interrupted by such bad mutations. Experiments are conducted on 29 benchmark test functions to show the promising performance of our proposed PSOTM. The results show that the PSO-TM performs much better than the standard PSO on almost all of the 29 test functions, especially those multimodal, complex ones of hybrid composition. Besides, PSO-TM adds little computation complexity to the standard PSO, and runs almost equally fast. Furthermore, we have implemented PSO-TM based on Graphic Processing Unit(GPU) in parallel. Compared with the CPU-based standard PSO, the proposed PSO-TM can reach a speedup of 25×, as well as an improved optimizing performance.

As a new kind of vehicles with low fuel cost and low emission, hybrid electric vehicle (HEV) has been given more and more attention in recent years. The key technique in the HEV is the optimal control strategy for the best performance. This paper proposed a new torque control strategy with charge buffer (TCSCB) to control the two power sources of the HEV. The TCSCB is based on the control of engine torque which make the control strategy easily distribute the output power to the engine and motor. In this control strategy, the real time optimization based on the engine efficiency map increases engine efficiency observably. The charge buffer reduces the dramatic fluctuation of the engine torque to improve the fuel economy. The prediction engine torque based on the neural network improves the control performance by the future information greatly. The simulation results showed the TCSCB could reach a higher fuel economy and lower emission compared to the current control strategies. In order to optimize the control performances, the parameters in the TCSCB were also discussed in details.

Neural network techniques have proven to be flexible in pattern recognition and information processing in complex environments.They typically include BP networks, RBF networks, support vector machine (SVM) and other similar biologically motivated models.The neural network techniques are able to enhance recognition accuracy, and have found applications in real-world environments.This special issue addresses neural network techniques in pattern recognition and information processing problems.The first paper ''Kernel based improved discriminant analysis and its application to face recognition,'' coauthored by Dake Zhou and Zhenmin Tang, presents a variant of KDA called kernel-based improved discriminant analysis (KIDA).In the proposed framework, original samples are projected firstly into a feature space by an implicit nonlinear mapping.After reconstructing betweenclass scatter matrix in the feature space by weighted

A novel parallel approach to run standard particle swarm optimization (SPSO) on Graphic Processing Unit (GPU) is presented in this paper. By using the general-purpose computing ability of GPU and based on the software platform of Compute Unified Device Architecture (CUDA) from NVIDIA, SPSO can be executed in parallel on GPU. Experiments are conducted by running SPSO both on GPU and CPU, respectively, to optimize four benchmark test functions. The running time of the SPSO based on GPU (GPU-SPSO) is greatly shortened compared to that of the SPSO on CPU (CPU-SPSO). Running speed of GPU-SPSO can be more than 11 times as fast as that of CPU-SPSO, with the same performance, compared to CPU-SPSO, GPU-SPSO shows special speed advantages on large swarm population applications and high dimensional problems, which can be widely used in real optimizing problems.

As a new kind of vehicles with low fuel cost and low emission, hybrid electric vehicle (HEV) has been given more and more attentions in recent years. The key technique in the HEV is adopting the optimal control strategy for the best performance. As the premise, a correct driving condition discrimination has an extremely important significance. This paper proposes an intelligent multi-feature statistical approach to discriminate the driving conditions of the HEV automatically. First of all, this approach samples the driving cycle periodically. Then it extracts multiple statistical features and tests their significance by statistical analysis. After that, it applies SVM and other machine learning methods to discriminate the driving conditions intelligently and automatically. Compared to the others, the proposed approach can compute fast and discriminate in real time during the whole HEV running. In our experiments, it reaches an accuracy of 97%. As a result, our approach can mine the valid information in the data completely and extract multiple features which have clear meanings and significance. Finally, according to the prediction experiment by a neural network and the fitting experiment by the ARMA model, it turns out that our proposed approach raises the efficiency of controlling the HEV considerably.

Inspired by human immune system, a concentration based feature construction (CFC) approach which utilizes a two-element concentration vector as the feature vector is proposed for spam detection in this paper. In the CFC approach, dasiaselfpsila and dasianon-selfpsila concentrations are constructed by using dasiaselfpsila and dasianon-selfpsila gene libraries, respectively, and subsequently are used to form a vector with two elements of concentrations for characterizing the e-mail efficiently. As a result, the design of classifier actually amounts to establishing a mapping between two real-value inputs and one binary output. The classification of the e-mail is considered as an optimization problem aiming at minimizing a formulated cost function. A clonal particle swarm optimization (CPSO) algorithm proposed by the leading author is also employed for this purpose. Several classifiers including linear discriminant, multi-layer neural networks and support vector machine are used to verify the effectiveness and robustness of the CFC approach. Experimental results demonstrate that the proposed CFC approach not only has a very much fast speed but also gives 97% and 99% of accuracy just using a two-element concentration feature vector on corpus PU1 and Ling, respectively.

In this paper, a concentric spatial extension based particle swarm optimization (CSE-PSO) is proposed by combining the spatial extension with the brood sorting in ant colonies, which leads to a concentric spatial extension scheme for the PSO. The brood sorting in ant colonies endows the particles in PSO with different radii adaptively according their distances to the best position of the swarm. In such a way, the search space in the CSE-PSO is not only enlarged greatly but also the diversity of the swarm in the CSE-PSO is increased accordingly. Meanwhile, a better trade-off between exploration and exploitation in the PSO is achieved by the concentric spatial extension. Simulation results on the fifteen benchmark test functions announced in IEEE CEC'2005 show that the proposed CSE-PSO is not only capable of speeding up the convergence but also improving the performance of global optimizer greatly on all the fifteen benchmark test functions.

A novel particle swarm optimization algorithm based on advance and retreat strategy and clone mechanism (ARC-PSO) is proposed in this paper. It is well known that the advance-and-retreat strategy is a simple and effective method of one-dimensional search. We use the advance-and-retreat strategy to endow the clones with faster speed to find nearby local basins before next clonal operation. Furthermore, in the next clonal operation, the search space is enlarged greatly and the diversity of clones is increased. When the fitness value turns better after last ldquoflyingrdquo, the cloned particle advances. On the contrary, the cloned particle retreats then searches in the reverse direction of the last ldquoflyingrdquo with a small step-size of the previous velocity. Thus, the swarm has strong optimization ability. Comparisons among the proposed ARC-PSO, the conventional standard particle swarm optimization (SPSO) and the pure clone particle swarm optimization (CPSO) on thirteen benchmark test functions are presented in this paper. Experimental results show that the proposed ARC-PSO is capable of speeding up the evolution process significantly and improving the performance of global optimizer greatly.

As advances in green automotives, hybrid electric vehicle (BEV) has being given more and more attention in recent years. The power management control strategy of BEV is the key problem that determines the efficiency and pollution emission level of the BEV, which requires the forecast of driving load situation of BEV in advance. This paper proposes an efficient approach for forecasting the driving load of the BEV by using Discrete Cosine Transform (DCT) and Support Vector Machine (SVM). The DCT is used to extract features from raw data, and reduce the dimensionality of feature which will result in an efficient SVM classification. The SVM is used to classify the current driving load into one of five presetting levels of the driving load of the BEV In such way, we can predict the driving load efficiently and accurately, which leads to a reasonable control to the BEV and gives as a high efficiency and low emission level as possible. Finally, a number of experiments are conducted to verify the validity of our proposed approach. Compared to current methods, our proposed approach gives a considerably promising performance through extensive experiments and comparison tests.

Particle swarm optimization (PSO) is a stochastic global optimization algorithm inspired by social behavior of bird flocking in search for food, which is a simple but powerful, and widely used as a problem-solving technique to a variety of complex problems in science and engineering. A novel particle swarm optimization algorithm based on immunity-clonal strategies, called as clonal particle swarm optimization (CPSO), is proposed at first in this paper. By cloning the best individual of ten succeeding generations, CPSO has better optimization solving capability and faster convergence performance than the conventional standard particle swarm optimization (SPSO) based on a number of simulations. A detailed description and explanation of the CPSO algorithm are given in the paper. Several experiments on six benchmark test functions are conducted to demonstrate that the proposed CPSO algorithm is able to speedup the evolution process and improve the performance of global optimizer greatly, while avoiding the premature convergence on the multidimensional variable space.

We present a mechanism to train support vector machines (SVMs) with a hybrid kernel and minimal Vapnik-Chervonenkis (VC) dimension. After describing the VC dimension of sets of separating hyperplanes in a high-dimensional feature space produced by a mapping related to kernels from the input space, we proposed an optimization criterion to design SVMs by minimizing the upper bound of the VC dimension. This method realizes a structural risk minimization and utilizes a flexible kernel function such that a superior generalization over test data can be obtained. In order to obtain a flexible kernel function, we develop a hybrid kernel function and a sufficient condition to be an admissible Mercer kernel based on common Mercer kernels (polynomial, radial basis function, two-layer neural network, etc.). The nonnegative combination coefficients and parameters of the hybrid kernel are determined subject to the minimal upper bound of the VC dimension of the learning machine. The use of the hybrid kernel results in a better performance than those with a single common kernel. Experimental results are discussed to illustrate the proposed method and show that the SVM with the hybrid kernel outperforms that with a single common kernel in terms of generalization power.

Demixing independent source signals from their nonlinear mixtures is a very important issue in many scenarios. This paper presents a novel method for blindly separating unobservable independent source signals from their nonlinear mixtures. The demixing system is modeled using a parameterized neural network whose parameters can be determined under the criterion of independence of its outputs. Compared to conventional gradient-based approaches, the GA-based approach for blind source separation is characterized by high accuracy, high robustness, and high convergence rate. Simulation results are discussed to demonstrate that the proposed GA-based approach is capable of separating independent sources from their nonlinear mixtures generated by a parametric separation model.

This paper presents a novel method for blindly separating unobservable independent source signals from their nonlinear mixtures. The demixing system is modeled using a parameterized neural network whose parameters can be determined under the criterion of independence of its outputs. Two cost functions based on higher order statistics are established to measure the statistical dependence of the outputs of the demixing system. The proposed method utilizes a genetic algorithm (GA) to minimize the highly nonlinear and nonconvex cost functions. The GA-based global optimization technique is able to obtain superior separation solutions to the nonlinear blind separation problem from any random initial values. Compared to conventional gradient-based approaches, the GA-based approach for blind source separation is characterized by high accuracy, robustness, and convergence rate. In particular, it is very suitable for the case of limited available data. Simulation results are discussed to demonstrate that the proposed GA-based approach is capable of separating independent sources from their nonlinear mixtures generated by a parametric separation model.

This paper proposes a novel neural-network approach to blind source separation in nonlinear mixture. The approach utilizes a radial basis function (RBF) neural-network to approximate the inverse of the nonlinear mixing mapping which is assumed to exist and able to be approximated using an RBF network. A contrast function which consists of the mutual information and partial moments of the outputs of the separation system, is defined to separate the nonlinear mixture. The minimization of the contrast function results in the independence of the outputs with desirable moments such that the original sources are separated properly. Two learning algorithms for the parametric RBF network are developed by using the stochastic gradient descent method and an unsupervised clustering method. By virtue of the RBF neural network, this proposed approach takes advantage of high learning convergence rate of weights in the hidden layer and output layer, natural unsupervised learning characteristics, modular structure, and universal approximation capability. Simulation results are presented to demonstrate the feasibility, robustness, and computability of the proposed method.

We apply a recurrent neural network to support vector machine (SVM) training for pattern recognition. Specifically, a primal-dual neural network is exploited to solve the quadratic programming problem encountered in training SVMs. The properties of the network allow one to design SVMs without adjustable network parameters and give a better solution for ill-posed problems.

A novel neural network approach is developed for nonlinear blind separation using a radial b axis function (RBF) network and an information theoretic criterion. By utilizing the universal approximation ability and local response property of an RBF network the proposed separation method is characterized by fast convergence and strong demixing ability. After its learning process, the RBF network is able to separate independent signals effectively from their nonlinear mixtures by the nonlinear channel model without the prior knowledge of the source signals and mixing channels. Experimental results illustrate the validity and effectiveness of the proposed method.