Abstract: As biometric systems become widely deployed, the security of identification systems can be easily attacked by various spoof materials. This paper contributes to finding a reliable and practical anti-spoofing method using Convolutional Neural Networks (CNNs) based on the types of loss functions and optimizers. The types of CNNs used in this paper include AlexNet, VGGNet, and ResNet. By using various loss functions including Cross-Entropy, Center Loss, Cosine Proximity, and Hinge Loss, and various loss optimizers which include Adam, SGD, RMSProp, Adadelta, Adagrad, and Nadam, we obtained significant performance changes. We realize that choosing the correct loss function for each model is crucial since different loss functions lead to different errors on the same evaluation. By using a subset of the Livdet 2017 database, we validate our approach to compare the generalization power. It is important to note that we use a subset of LiveDet and the database is the same across all training and testing for each model. This way, we can compare the performance, in terms of generalization, for the unseen data across all different models. The best CNN (AlexNet) with the appropriate loss function and optimizers result in more than 3% of performance gain over the other CNN models with the default loss function and optimizer. In addition to the highest generalization performance, this paper also contains the models with high accuracy associated with parameters and mean average error rates to find the model that consumes the least memory and computation time for training and testing. Although AlexNet has less complexity over other CNN models, it is proven to be very efficient. For practical anti-spoofing systems, the deployed version should use a small amount of memory and should run very fast with high anti-spoofing performance. For our deployed version on smartphones, additional processing steps, such as quantization and pruning algorithms, have been applied in our final model.
Abstract: Clustering is an intensive research for some years
because of its multifaceted applications, such as biology, information
retrieval, medicine, business and so on. The expectation maximization
(EM) is a kind of algorithm framework in clustering methods, one
of the ten algorithms of machine learning. Traditionally, optimization
of objective function has been the standard approach in EM. Hence,
research has investigated the utility of evolutionary computing and
related techniques in the regard. Chemical Reaction Optimization
(CRO) is a recently established method. So the property embedded
in CRO is used to solve optimization problems. This paper presents
an algorithm framework (EM-CRO) with modified CRO operators
based on EM cluster problems. The hybrid algorithm is mainly
to solve the problem of initial value sensitivity of the objective
function optimization clustering algorithm. Our experiments mainly
take the EM classic algorithm:k-means and fuzzy k-means as an
example, through the CRO algorithm to optimize its initial value, get
K-means-CRO and FKM-CRO algorithm. The experimental results
of them show that there is improved efficiency for solving objective
function optimization clustering problems.
Abstract: The crossover probability and mutation probability are the two important factors in genetic algorithm. The adaptive genetic algorithm can improve the convergence performance of genetic algorithm, in which the crossover probability and mutation probability are adaptively designed with the changes of fitness value. We apply adaptive genetic algorithm into a function optimization problem. The numerical experiment represents that adaptive genetic algorithm improves the convergence speed and avoids local convergence.
Abstract: Genetic algorithm is widely used in optimization
problems for its excellent global search capabilities and highly parallel
processing capabilities; but, it converges prematurely and has a poor
local optimization capability in actual operation. Simulated annealing
algorithm can avoid the search process falling into local optimum. A
hybrid genetic algorithm based on simulated annealing is designed by
combining the advantages of genetic algorithm and simulated
annealing algorithm. The numerical experiment represents the hybrid
genetic algorithm can be applied to solve the function optimization
problems efficiently.
Abstract: In this paper, we propose a selective mutation method
for improving the performances of genetic algorithms. In selective
mutation, individuals are first ranked and then additionally mutated
one bit in a part of their strings which is selected corresponding to
their ranks. This selective mutation helps genetic algorithms to fast
approach the global optimum and to quickly escape local optima.
This results in increasing the performances of genetic algorithms.
We measured the effects of selective mutation with four function
optimization problems. It was found from extensive experiments that
the selective mutation can significantly enhance the performances of
genetic algorithms.
Abstract: Ant colony optimization (ACO) and its variants are
applied extensively to resolve various continuous optimization
problems. As per the various diversification and intensification
schemes of ACO for continuous function optimization, researchers
generally consider components of multidimensional state space to
generate the new search point(s). However, diversifying to a new
search space by updating only components of the multidimensional
vector may not ensure that the new point is at a significant distance
from the current solution. If a minimum distance is not ensured
during diversification, then there is always a possibility that the
search will end up with reaching only local optimum. Therefore, to
overcome such situations, a Mahalanobis distance-based
diversification with Nelder-Mead simplex-based search scheme for
each ant is proposed for the ACO strategy. A comparative
computational run results, based on nine nonlinear standard test
problems, confirms that the performance of ACO is improved
significantly with the integration of the proposed schemes in the
ACO.
Abstract: In this paper a data miner based on the learning
automata is proposed and is called LA-miner. The LA-miner extracts
classification rules from data sets automatically. The proposed
algorithm is established based on the function optimization using
learning automata. The experimental results on three benchmarks
indicate that the performance of the proposed LA-miner is
comparable with (sometimes better than) the Ant-miner (a data miner
algorithm based on the Ant Colony optimization algorithm) and CNZ
(a well-known data mining algorithm for classification).
Abstract: In most of the popular implementation of Parallel GAs
the whole population is divided into a set of subpopulations, each
subpopulation executes GA independently and some individuals are
migrated at fixed intervals on a ring topology. In these studies,
the migrations usually occur 'synchronously' among subpopulations.
Therefore, CPUs are not used efficiently and the communication
do not occur efficiently either. A few studies tried asynchronous
migration but it is hard to implement and setting proper parameter
values is difficult.
The aim of our research is to develop a migration method which is
easy to implement, which is easy to set parameter values, and which
reduces communication traffic. In this paper, we propose a traffic
reduction method for the Asynchronous Parallel Distributed GA by
migration of elites only. This is a Server-Client model. Every client
executes GA on a subpopulation and sends an elite information to the
server. The server manages the elite information of each client and
the migrations occur according to the evolution of sub-population in
a client. This facilitates the reduction in communication traffic.
To evaluate our proposed model, we apply it to many function optimization
problems. We confirm that our proposed method performs
as well as current methods, the communication traffic is less, and
setting of the parameters are much easier.
Abstract: The necessity of solving multi dimensional
complicated scientific problems beside the necessity of several
objective functions optimization are the most motive reason of born
of artificial intelligence and heuristic methods.
In this paper, we introduce a new method for multiobjective
optimization based on learning automata. In the proposed method,
search space divides into separate hyper-cubes and each cube is
considered as an action. After gathering of all objective functions
with separate weights, the cumulative function is considered as the
fitness function. By the application of all the cubes to the cumulative
function, we calculate the amount of amplification of each action and
the algorithm continues its way to find the best solutions. In this
Method, a lateral memory is used to gather the significant points of
each iteration of the algorithm. Finally, by considering the
domination factor, pareto front is estimated. Results of several
experiments show the effectiveness of this method in comparison
with genetic algorithm based method.