35 research outputs found
An analysis of cooperative coevolutionary differential evolution as neural networks optimizer
No full textDifferential Evolution for Neural Networks (DENN) is an optimizer for neural network weights based on Differential Evolution. Although DENN has shown good performance with middle-size networks, the number of weights is an evident limitation of the approach. The aim of this work is to figure out if coevolutionary strategies implemented on top of DENN could be of help during the optimization phase. Moreover, we studied two of the classical problems connected to the application of evolutionary computation, i.e. the stagnation and the lack of population diversity, and the use of a crowding strategy to address them. The system has been tested on classical benchmark classification problems and experimental results are presented and discussed
Neural Random Access Machines Optimized by Differential Evolution
No full textRecently a research trend of learning algorithms by means of deep learning techniques has started. Most of these are different implementations of the controller-interface abstraction: they use a neural controller as a “processor" and provide different interfaces for input, output and memory management. In this trend, we consider of particular interest the Neural Random-Access Machines, called NRAM, because this model is also able to solve problems which require indirect memory references. In this paper we propose a version of the Neural Random-Access Machines, where the core neural controller is trained with Differential Evolution meta-heuristic instead of the usual backpropagation algorithm. Some experimental results showing that this approach is effective and competitive are also presented
Planning with Graded Fluents and Actions
No full textThis work can be seen as a rst approach to a new
planning model that takes into account the possibility
to express actions and uents with non-boolean
values. According to this model, a planning problem
is dened using both graded (multi-valued) and
classical (boolean) uents. Moreover, actions that
can have different application degrees can be defined. In this work a PDDL extension allowing to
describe such new problems is proposed and a planning
algorithm for such problems is presented
Memes evolution in a memetic variant of particle swarm optimization
No full textIn this work, a coevolving memetic particle swarm optimization (CoMPSO) algorithm is presented. CoMPSO introduces the memetic evolution of local search operators in particle swarm optimization (PSO) continuous/discrete hybrid search spaces. The proposed solution allows one to overcome the rigidity of uniform local search strategies when applied to PSO. The key contribution is that memes provides each particle of a PSO scheme with the ability to adapt its exploration dynamics to the local characteristics of the search space landscape. The objective is obtained by an original hybrid continuous/discrete meme representation and a probabilistic co-evolving PSO scheme for discrete, continuous, or hybrid spaces. The coevolving memetic PSO evolves both the solutions and their associated memes, i.e. the local search operators. The proposed CoMPSO approach has been experimented on a standard suite of numerical optimization benchmark problems. Preliminary experimental results show that CoMPSO is competitive with respect to standard PSO and other memetic PSO schemes in literature, and its a promising starting point for further research in adaptive PSO local search operators
Autoencoders for unsupervised real-time bridge health assessment
No full textOver the last decades, the rising number of aging infrastructures has progressively fueled much interest toward the field of structural health monitoring. Following the increasing popularity of artificial intelligence algorithms, an autoencoder-based damage detection technique within the context of unsupervised learning is proposed in this paper to provide support for practical engineering applications. The developed methodology uses the autoencoder to reconstruct raw acceleration sequences of user-defined length collected from a healthy structure. To quantify the errors between the original input and the reconstructed output, which may be representative of damage occurrence, two indexes of reconstruction loss are selected as damage-sensitive features. To support damage detection, a selected number of short-time sequences are finally grouped into a unique macrosequence. The novel procedure can effectively both work at the single sensor level, as well as combine the predictive models using an ensemble learning strategy. Avoiding system identification, results obtained in the Z24 bridge demonstrate that the proposed method is quite effective for local damage detection with limited computational effort and using a limited number of sensors, thereby suitable to be easily applicable in the context of real-time bridge assessment
A Proposal for Planning with graded fluents and actions (Pianificare con azioni e fluenti graduati)
No full textLinear temporal logic as an executable semantics for planning languages
No full textThis paper presents an approach to artificial intelligence planning based on linear temporal logic (LTL). A simple and easy-to-use planning language is described, PDDL-K (Planning Domain Description Language with control Knowledge), which allows one to specify a planning problem together with heuristic information that can be of help for both pruning the search space and finding better quality plans. The semantics of the language is given in terms of a translation into a set of LTL formulae. Planning is then reduced to “executing” the LTL encoding, i.e. to model search in LTL. The feasibility of the approach has been successfully tested by means of the system Pdk, an implementation of the proposed method
Towards a Parallel Search Engine for Planning Systems Based on Linear Time Logic
No full textA planning problem can be entirely encoded as a set of linear temporal logic (LTL) formulae, in such a way that planning is reduced to model search. In order for this approach to be effective, it is important to enhance the performances of LTL provers. In this work, we study a parallel algorithm for LTL model search, based on the tableaux calculus. In paritcular, the approach presented here is based on the “divide et impera” approach: a task in tableaux construction is identified that can be split into smaller homogeneous processes. The parallelization acts during the construction of each time state: the set of formulas to be expanded is split into k disjoint subsets (where k is the number of processes), the k tableaux expansions are carried out in parallel, and the k results are suitably combined. First promising experimental results are also presented: they are based on the algorithm implementation on a cluster of non homogeneous machines
An Intelligent Cache Management for Data Analysis at CMS
No full textIn this work, we explore a score-based approach to manage a cache system. With the proposed method, the cache can better discriminate the input requests and improve the overall performances. We created a score based discriminator using the file statistics. The score represents the weight of a file. We tested several functions to compute the file weight used to determine whether a file has to be stored in the cache or not. We developed a solution experimenting on a real cache manager named XCache, that is used within the Compact Muon Solenoid (CMS) data analysis workflow. The aim of this work is optimizing to reduce maintaining costs of the cache system without compromising the user experience
