Ultra-lightweight CNN design based on neural architecture search and knowledge distillation: A novel method to build the automatic recognition model of space target ISAR images

In this paper,a novel method of ultra-lightweight convolution neural network(CNN)design based on neural architecture search(NAS)and knowledge distillation(KD)is proposed.It can realize the automatic construction of the space target inverse synthetic aperture radar(ISAR)image recognition model with ultra-lightweight and high accuracy.This method introduces the NAS method into the radar image recognition for the first time,which solves the time-consuming and labor-consuming problems in the artificial design of the space target ISAR image automatic recognition model(STIIARM).On this basis,the NAS model’s knowledge is transferred to the student model with lower computational complexity by the flow of the solution procedure(FSP)distillation method.Thus,the decline of recognition accuracy caused by the direct compression of model structural parameters can be effectively avoided,and the ultralightweight STIIARM can be obtained.In the method,the Inverted Linear Bottleneck(ILB)and Inverted Residual Block(IRB)are firstly taken as each block’s basic structure in CNN.And the expansion ratio,output filter size,number of IRBs,and convolution kernel size are set as the search parameters to construct a hierarchical decomposition search space.Then,the recognition accuracy and computational complexity are taken as the objective function and constraint conditions,respectively,and the global optimization model of the CNN architecture search is established.Next,the simulated annealing(SA)algorithm is used as the search strategy to search out the lightweight and high accuracy STIIARM directly.After that,based on the three principles of similar block structure,the same corresponding channel number,and the minimum computational complexity,the more lightweight student model is designed,and the FSP matrix pairing between the NAS model and student model is completed.Finally,by minimizing the loss between the FSP matrix pairs of the NAS model and student model,the student model’s weight adjustment is completed.Thus the ultra-lightweight and high accuracy STIIARM is obtained.The proposed method’s effectiveness is verified by the simulation experiments on the ISAR image dataset of five types of space targets.

Using Genetic Algorithms to Improve the Search of the Weight Space in Cascade－Correlation Neural Network

In this paper, we use the global search characteristics of genetic algorithms to help search the weight space of the neurons in the cascade-correlation architecture. The cascade-correlation learning architecture is a technique of training and building neural networks that starts with a simple network of neurons and adds additional neurons as they are needed to suit a particular problem. In our approach, instead ofmodifying the genetic algorithm to account for convergence problems, we search the weight-space using the genetic algorithm and then apply the gradient technique of Quickprop to optimize the weights. This hybrid algorithm which is a combination of genetic algorithms and cascade-correlation is applied to the two spirals problem. We also use our algorithm in the prediction of the cyclic oxidation resistance of Ni- and Co-base superalloys.

Levy Constrained Search in Fock Space：An Alternative Approach to Noninteger Electron Number

By extending the Levy wavefunction constrained search to Fock Space,one can define a wavefunction constrained search for electron densities in systems having noninteger number of electrons.For pure-state v-representable densities,the results are equivalent to what one would obtain with the zero-temperature grand canonical ensemble.In other cases,the wavefunction constrained search in Fock space presents an upper bound to the grand canonical ensemble functional.One advantage of the Fock-space wavefunction constrained search functional over the zero-temperature grand-canonical ensemble constrained search functional is that certain specific excited states(i.e.,those that are not ground-statev-representable) are the stationary points of the Fock-space functional.However,a potential disadvantage of the Fock-space constrained search functional is that it is not convex.

Indexing the bit-code and distance for fast KNN search in high-dimensional spaces

Various index structures have recently been proposed to facilitate high-dimensional KNN queries, among which the techniques of approximate vector presentation and one-dimensional (1D) transformation can break the curse of dimensionality. Based on the two techniques above, a novel high-dimensional index is proposed, called Bit-code and Distance based index (BD). BD is based on a special partitioning strategy which is optimized for high-dimensional data. By the definitions of bit code and transformation function, a high-dimensional vector can be first approximately represented and then transformed into a 1D vector, the key managed by a B+-tree. A new KNN search algorithm is also proposed that exploits the bit code and distance to prune the search space more effectively. Results of extensive experiments using both synthetic and real data demonstrated that BD out- performs the existing index structures for KNN search in high-dimensional spaces.

A local space transfer learning-based parallel Bayesian optimization with its application

The optimization of process parameters in polyolefin production can bring significant economic benefits to the factory.However,due to small data sets,high costs associated with parameter verification cycles,and difficulty in establishing an optimization model,the optimization process is often restricted.To address this issue,we propose using a transfer learning Bayesian optimization strategy to improve the efficiency of parameter optimization while minimizing resource consumption.Specifically,we leverage Gaussian process(GP)regression models to establish an integrated model that incorporates both source and target grade production task data.We then measure the similarity weights of each model by comparing their predicted trends,and utilize these weights to accelerate the solution of optimal process parameters for producing target polyolefin grades.In order to enhance the accuracy of our approach,we acknowledge that measuring similarity in a global search space may not effectively capture local similarity characteristics.Therefore,we propose a novel method for transfer learning optimization that operates within a local space(LSTL-PBO).This method employs partial data acquired through random sampling from the target task data and utilizes Bayesian optimization techniques for model establishment.By focusing on a local search space,we aim to better discern and leverage the inherent similarities between source tasks and the target task.Additionally,we incorporate a parallel concept into our method to address multiple local search spaces simultaneously.By doing so,we can explore different regions of the parameter space in parallel,thereby increasing the chances of finding optimal process parameters.This localized approach allows us to improve the precision and effectiveness of our optimization process.The performance of our method is validated through experiments on benchmark problems,and we discuss the sensitivity of its hyperparameters.The results show that our proposed method can significantly improve the efficiency of process parameter optimization,reduce the dependence on source tasks,and enhance the method's robustness.This has great potential for optimizing processes in industrial environments.

Enhanced Differentiable Architecture Search Based on Asymptotic Regularization

In differentiable search architecture search methods,a more efficient search space design can significantly improve the performance of the searched architecture,thus requiring people to carefully define the search space with different complexity according to various operations.Meanwhile rationalizing the search strategies to explore the well-defined search space will further improve the speed and efficiency of architecture search.With this in mind,we propose a faster and more efficient differentiable architecture search method,AllegroNAS.Firstly,we introduce a more efficient search space enriched by the introduction of two redefined convolution modules.Secondly,we utilize a more efficient architectural parameter regularization method,mitigating the overfitting problem during the search process and reducing the error brought about by gradient approximation.Meanwhile,we introduce a natural exponential cosine annealing method to make the learning rate of the neural network training process more suitable for the search procedure.Moreover,group convolution and data augmentation are employed to reduce the computational cost.Finally,through extensive experiments on several public datasets,we demonstrate that our method can more swiftly search for better-performing neural network architectures in a more efficient search space,thus validating the effectiveness of our approach.

Enhanced self-adaptive evolutionary algorithm for numerical optimization

There are many population-based stochastic search algorithms for solving optimization problems. However, the universality and robustness of these algorithms are still unsatisfactory. This paper proposes an enhanced self-adaptiveevolutionary algorithm （ESEA） to overcome the demerits above. In the ESEA, four evolutionary operators are designed to enhance the evolutionary structure. Besides, the ESEA employs four effective search strategies under the framework of the self-adaptive learning. Four groups of the experiments are done to find out the most suitable parameter values for the ESEA. In order to verify the performance of the proposed algorithm, 26 state-of-the-art test functions are solved by the ESEA and its competitors. The experimental results demonstrate that the universality and robustness of the ESEA out-perform its competitors.

Harmony search algorithm with differential evolution based control parameter co-evolution and its application in chemical process dynamic optimization

A modified harmony search algorithm with co-evolutional control parameters(DEHS), applied through differential evolution optimization, is proposed. In DEHS, two control parameters, i.e., harmony memory considering rate and pitch adjusting rate, are encoded as a symbiotic individual of an original individual(i.e., harmony vector). Harmony search operators are applied to evolving the original population. DE is applied to co-evolving the symbiotic population based on feedback information from the original population. Thus, with the evolution of the original population in DEHS, the symbiotic population is dynamically and self-adaptively adjusted, and real-time optimum control parameters are obtained. The proposed DEHS algorithm has been applied to various benchmark functions and two typical dynamic optimization problems. The experimental results show that the performance of the proposed algorithm is better than that of other HS variants. Satisfactory results are obtained in the application.

An iterated local coordinate-exchange algorithm for constructing experimental designs for multi-dimensional constrained spaces

Iterated local search(ILS)is used to construct the optimal experimental designs for multi-dimensional constrained spaces,in which the inner loop is based on the stochastic coordinate-exchange(SCE)algorithm.Every time a local optimal solution is found by the SCE algorithm,the perturbation operator is applied to it,and then a new solution is explored in the areas where the exchange of coordinates may produce improvement,so as to retain the features and attributes of the current optimal solution and avoid the defects of random restart.We implement the iterated local coordinate-exchange algorithm for experimental designs in the multi-dimensional constrained spaces.In addition,sensitivity analysis was conducted to analyze the impacts of the parameters on the performance of the proposed algorithm.Also we compared the performance of the proposed algorithm to the SCE algorithm using the random restart strategy.The analysis shows that the proposed algorithm is better than the SCE algorithm in terms of efficiency and quality,especially in the experimental designs for high-dimensional constrained space.

Motion Planning Algorithm and Simulation for Space Manipulators

The specificities of collision-free path planning of space manipulators are analyzed. Path planning strategies are presented in consideration of these specificities, and an implementation procedure is also described in detail according to these strategies.

Self-Adaptive Algorithms for the Split Common Fixed Point Problem of the Demimetric Mappings

The split common fixed point problem is an inverse problem that consists in finding an element in a fixed point set such that its image under a bounded linear operator belongs to another fixed-point set. In this paper, we present new iterative algorithms for solving the split common fixed point problem of demimetric mappings in Hilbert spaces. Moreover, our algorithm does not need any prior information of the operator norm. Weak and strong convergence theorems are given under some mild assumptions. The results in this paper are the extension and improvement of the recent results in the literature.

Scheduling Optimization of Space Object Observations for Radar

An optimizing method of observation scheduling based on time-division multiplexing is proposed in this paper,and its efficiency is verified by outdoor experiments. The initial observation scheduling is first obtained by using a semi-random search algorithm,and secondly the connection time pair（ CTP） between adjacent objects is optimized by using a genetic algorithm. After obtaining these two parameters,the final observation scheduling can be obtained. According to pre-designed tracks between each adjacent objects in observation order,the seamless observation of neighboring targets is derived by automatically steering the antenna beam,so the observation efficiency is improved.

Architecture of a Commercialized Search Engine Using Mobile Agents

Shopping Search Engine(SSE)implies a unique challenge for validating distinct items available online in market place.For sellers,having a user finding relevant search results on top is very difficult.Buyers tend to click on and buy from the listings which appear first.Search engine optimization devotes that goal to influence such challenges.In current shopping search platforms,lots of irrelevant items retrieved from their indices;e.g.retrieving accessories of exact items rather than retrieving the items itself,regardless the price of item were considered or not.Also,users tend to move from shoppers to another searching for appropriate items where the time is crucial for consumers.In our proposal,we exploit the drawbacks of current shopping search engines,and the main goal of this research is to combine and merge multiple search results retrieved from some highly professional shopping sellers in the commercial market.Experimental results showed that our approach is more efficient and robust for retrieving a complete list of desired and relevant items with respect to all query space.

神经架构搜索综述

近几年,深度学习因具有强大的表征能力,已经在许多领域中取得了突破性的进展,而神经网络的架构对它的性能至关重要。然而,高性能的神经网络架构设计严重依赖研究人员的先验知识和经验,神经网络参数量庞大,难以设计最优的神经网络架构,因此自动神经架构搜索(NAS)获得了极大的关注。NAS是一种使用机器学习的方法,可以在不需要大量人力的情况下,自动搜索最优网络架构的技术,是未来神经网络设计的重要手段之一。NAS本质上是一个搜索优化问题,通过对搜索空间、搜索策略和性能评估策略的设计,自动搜索最优的网络结构。从搜索空间、搜索策略和性能评估策略这3个方面详细且全面地分析、比较和总结目前NAS的研究进展,方便读者快速了解神经架构搜索的发展过程和各项技术的优缺点,并提出NAS未来可能的研究发展方向。

空间机器人预定义任意时间最优轨迹规划

针对双臂自由漂浮空间机器人(FFSR)在太空抓取任务中的轨迹规划问题,设计了一种FFSR系统的可实现误差预定义时间收敛的新型轨迹规划算法,用布谷鸟搜索算法进行参数优化,以实现末端执行器误差的快速收敛并获得平滑轨迹。首先,导出FFSR的基于姿态误差的运动学模型;然后,将累积危险场避碰算法应用于预定义任意时间轨迹规划,实现高跟踪精度的快速避碰轨迹;最后,利用布谷鸟搜索算法对预定义任意时间参数进行优化,得到机械臂低关节角速度的运动轨迹。仿真结果表明所提算法能够在预定义的时间内实现末端执行器误差快速收敛,而且能够得到更平滑的运动轨迹。

约束空间工业机器人姿态搜索及避障研究

为解决约束空间下机器人无碰撞目标姿态求解问题,提出一种机器人姿态快速搜索算法,并结合4种不同的路径规划算法实现工业机器人的多目标路径规划。首先,对机器人运动框架初步配置完成后,通过计算位姿矩阵得到了场景内物体坐标的动态相对关系,并在ROS中搭建规划场景;其次,提出一种基于模型和机器人逆解的机器人姿态快速搜索算法RFP,并在RoboDK中对其进行验证,验证结果表明该算法有效;最后,根据任务需求设计了路径规划流程并引入二次规划减少算法随机性的影响,提高规划成功率,应用了RRT等算法进行对比分析,仿真结果表明RRT-connect算法的规划速度与成功率高于其它算法,为样机的实现奠定了基础。

潜在空间中的策略搜索强化学习方法

策略搜索是深度强化学习领域中一种能够解决大规模连续状态空间和动作空间问题的高效学习方法,被广泛应用在现实问题中。然而,此类方法通常需要花费大量的学习样本和训练时间,且泛化能力较差,学到的策略模型难以泛化至环境中看似微小的变化。为了解决上述问题,提出了一种基于潜在空间的策略搜索强化学习方法。将学习状态表示的思想拓展到动作表示上,即在动作表示的潜在空间中学习策略,再将动作表示映射到真实动作空间中。通过表示学习模型的引入,摒弃端到端的训练方式,将整个强化学习任务划分成大规模的表示模型部分和小规模的策略模型部分,使用无监督的学习方法来学习表示模型,使用策略搜索强化学习方法学习小规模的策略模型。大规模的表示模型能保留应有的泛化性和表达能力,小规模的策略模型有助于减轻策略学习的负担,从而在一定程度上缓解深度强化学习领域中样本利用率低、学习效率低和动作选择泛化性弱的问题。最后,在智能控制任务CarRacing和Cheetah中验证了引入潜在空间中的状态表示和动作表示的有效性。

复合联运物流运输网络建模与路径优化

在全球经济一体化背景下,物流行业成为不可或缺的经济支柱之一,在经济中的作用不断显现,复合联运逐渐成为物流行业的重要运输方式。以空铁海复合联运作为研究对象,构建不同的搜索空间,确定边界条件并分析货运影响参数,包括时间、成本、距离。以遗传算法为原则,求解建立空铁海复合联运模型并进行路径优化。确定3组节点组,每组都包含普通货物、特殊货物、航线拥堵3种情况,分别优化成本、时间、距离,为复合联运业务提供路径选择依据。通过分析,在规定的搜索空间内分别选出了最优时间路线、最优成本路线、最优距离路线。期望可以在物流运输过程中实现降本增效、减少风险、提高行业竞争力的目的。

考虑综合性能最优的非短视快速天基雷达多目标跟踪资源调度算法

合理有效的资源调度是天基雷达效能得以充分发挥的关键。针对天基雷达多目标跟踪资源调度问题,建立了综合考虑目标威胁度、跟踪精度与低截获概率(LPI)的代价函数;考虑目标的不确定、天基平台约束以及长远期期望代价,建立了多约束下的基于部分可观测的马尔可夫决策过程(POMDP)的资源调度模型;采用拉格朗日松弛法将多约束下的多目标跟踪资源调度问题转换分解为多个无约束的子问题;针对连续状态空间、连续动作空间及连续观测空间引起的维数灾难问题,采用基于蒙特卡罗树搜索(MCTS)的在线POMDP算法—POMCPOW算法进行求解,最终提出了一种综合多指标性能的非短视快速天基雷达多目标跟踪资源调度算法。仿真表明,与已有调度算法相比,所提算法资源分配更合理,系统性能更优。

基于多标签集成学习的螺旋CT机故障诊断研究

医学应用领域计算机X线断层摄影螺旋机由于复杂的结构和较高的集成度在实际故障定位和检测中具有极高的难度;为解决这个问题,研究对螺旋CT机故障定位与检测问题进行了分析,提出一种多标签集成学习方法;该方法采用了折半查找算法获取螺旋CT机的故障数据,同时有效结合现有的卷积神经网络和循环神经网络的文本表征网络,通过自适应标签关系增强方法找出标签间的依赖关系,并利用加权约简标签集的不平衡学习能有效杜绝模型可扩展性低和模型泛化性弱等问题;经损失值、准确度、运行时间、精准率、灵敏度5个指标的实例测试结果表明,研究所给出的方法均相对于其他3种较为创新的多标签集成学习方法更具优势,且提升数值均超过2%,训练集的各个指标数据均比测试集相应数值更高;训练集和测试集中空时网络聚类约简的多标签集成学习方法的精准率分别为93.12%和87.26%,召回率分别为86.35%和84.25%;该方法能精准快速查找螺旋CT机的故障类型和故障部位,极大程度降低维修成本和延长设备的使用年限。