1,721,250 research outputs found
Modelling Crossover-Induced Linkage in Genetic Algorithms
No full textThe dynamics of a genetic algorithm undergoing ranking selection, mutation, and two-point crossover for the ones-counting problem is studied using a statistical mechanics approach. This approach has been used previously to study this problem, but with uniform crossover. Two-point crossover induces additional linkage between nearby loci which changes the dynamics significantly. To account for this linkage, the evolution of the auto-correlation function is incorporated into a model of the dynamics. This complicates the analysis and requires several additional approximations to be made. Nevertheless, the model we derive is shown to capture the main features of the dynamics and is in good agreement with simulations
Finite Population Effects for Ranking and Tournament Selection
No full textThe effect of ranking and binary tournament selection on the distribution of fitnesses and genetic correlation is calculated for a finite population. The results are different for continuous and discrete fitness functions. Results for both situations are obtained. These exact results are compared to a previously obtained approximation up to the third cumulant and shown to be in good agreement. Tournament and ranking selection are compared with Boltzmann selection for which exact finite population effects are already known
Fulfilling the needs of gray-sheep users in recommender systems, a clustering solution
No full textRecommender systems apply data mining techniques for filtering unseen information and can predict whether a user would like a given item. This paper focuses on graysheep users problem responsible for the increased error rate in collaborative filtering based recommender systems algorithms. The main contribution of this paper lies in showing that (1) the presence of gray-sheep users can affect the performance— accuracy and coverage—of collaborative filtering based algorithms, depending on the data sparsity and distribution; (2) graysheep users can be identified using clustering algorithms in offline fashion, where the similarity threshold to isolate these users from the rest of clusters can be found empirically; (3) contentbased profile of gray-sheep users can be used for making accurate recommendations. The effectiveness of the proposed algorithm is tested on the MovieLens dataset and community of movie fans in the FilmTrust Website, using mean absolute error, receiver operating characteristic sensitivity, and coverage
Learning the large-scale structure of the max-sat landscape using populations
Full text linkA new algorithm for solving MAX-SAT problems is introduced which clusters good solutions, and restarts the search from the closest feasible solution to the centroid of each cluster. This is shown to be highly efficient for finding good solutions of large MAX-SAT problems. We argue that this success is due to the population learning the large-scale structure of the fitness landscape. Systematic studies of the landscape are presented to support this hypothesis. In addition, a number of other strategies are tested to rule out other possible explanations of the success. Preliminary results are shown indicating that extensions of the proposed algorithm can give similar improvements on other hard optimisation problems
Object detection for crabs in top-view seabed imagery
No full textThis report presents the application of object detection on a database of underwater images of different species of crabs, as well as aerial images of sea lions and finally the Pascal VOC dataset. The model is an end-to-end object detection neural network based on a convolutional network base and a Long Short-Term Memory detector
An improved switching hybrid recommender system using naive Bayes classifier and collaborative filtering
No full textRecommender Systems apply machine learning and data mining techniques for filtering unseen information and can predict whether a user would like a given resource. To date a number of recommendation algorithms have been proposed, where collaborative filtering and content-based filtering are the two most famous and adopted recommendation techniques. Collaborative filtering recommender systems recommend items by identifying other users with similar taste and use their opinions for recommendation; whereas content-based recommender systems recommend items based on the content information of the items. These systems suffer from scalability, data sparsity, over specialization, and cold-start problems resulting in poor quality recommendations and reduced coverage. Hybrid recommender systems combine individual systems to avoid certain aforementioned limitations of these systems. In this paper, we proposed a unique switching hybrid recommendation approach by combining a Naive Bayes classification approach with the collaborative filtering. Experimental results on two different data sets, show that the proposed algorithm is scalable and provide better performance – in terms of accuracy and coverage – than other algorithms while at the same time eliminates some recorded problems with the recommender systems
Novel Significance Weighting Schemes for Collaborative Filtering: Generating Improved Recommendations in Sparse Environments
No full textRecommender systems apply machine learning and data mining techniques for filtering unseen information and can predict whether a user would like a given resource. To date, a number of recommender system algorithms have been proposed, where collaborative filtering is the most famous and adopted recommendation algorithm. Collaborative filtering recommender systems recommend items by identifying other similar users, in case of user-based collaborative filtering, or similar items, in case of item-based collaborative filtering. Significance weighting schemes assign different weights to neighbouring users/items found against an active user/item. Several significance weighting schemes have been proposed [1], [2], [3], [4]. In this paper, we claim that these proposed schemes are flawed by the fact that they can not be applied to general recommender system datasets. We provide the correct generalized significance weighting schemes using different novel heuristics, and by extensive experimental results on three different data sets, show how significance weighting schemes affect the performance of a recommender system. Furthermore, we claim that the conventional weighted sum prediction formula used in item-based [5] collaborative filtering is not correct for very sparse datasets. We provide the correct prediction formula and empirically evaluate it
Building Switching Hybrid Recommender System Using Machine Learning Classifiers and Collaborative Filtering
No full textRecommender systems apply machine learning and data mining techniques for filtering unseen information and can predict whether a user would like a given resource. To date a number of recommendation algorithms have been proposed, where collaborative filtering and content-based filtering are the two most famous and adopted recommendation techniques. Collaborative filtering recommender systems recommend items by identifying other users with similar taste and use their opinions for recommendation; whereas content-based recommender systems recommend items based on the content information of the items. Moreover, machine learning classifiers can be used for recommendation by training them on content information. These systems suffer from scalability, data sparsity, over specialization, and cold-start problems resulting in poor quality recommendations and reduced coverage. Hybrid recommender systems combine individual systems to avoid certain aforementioned limitations of these systems. In this paper, we proposed unique generalized switching hybrid recommendation algorithms that combine machine learning classifiers with the collaborative filtering recommender systems. Experimental results on two different data sets, show that the proposed algorithms are scalable and provide better performance—in terms of accuracy and coverage—than other algorithms while at the same time eliminate some recorded problems with the recommender systems
RotLSTM: rotating memories in recurrent neural networks
No full textLong Short-Term Memory (LSTM) units have the ability to memorize and use long-term dependencies between inputs to generate predictions on time series data. We introduce the concept of modifying the cell state (memory) of LSTMs using rotation matrices parametrized by a new set of trainable weights. This addition shows significant increases of performance on some of the tasks from the bAbI dataset
Novel Heuristics for Coalition Structure Generation in Multi-Agent Systems
No full textA coalition is a set of self-interested agents that agree to cooperate for achieving a set of goals. Coalition formation is an active area of research in multi-agent systems nowadays. Central to this endeavour is the problem of determining which of the many possible coalitions to form in order to achieve some goal, which is called coalition structure generation. Coalition structure generation problem is extremely challenging due to the number of possible solutions that need to be examined, which grows exponentially with the number of agents involved. Generally, agents would enumerate all possible coalitions, store them in memory, and then try to construct the coalition structure that maximizes the sum of the values of the coalitions. However, this is not feasible when we have a large number of agents, and other constraints on execution time, and memory. Hence, there is a need to develop an algorithm that can generate solutions rapidly for large number of agents while providing bounds on the value of solution as well. With this in mind, we propose two new heuristics, namely LocalSearch and GreedySearch, for generating the coalition structure, which satisfy these properties. We empirically show that these heuristics are able to return ‘good-enough’ solutions in very short time. Furthermore, they enhance the performance of state of the art algorithm, IP (proposed by [12]) in terms of increased lower bound, anytime property, and solution quality. Furthermore, we implemented different heuristics for selecting a sub-space in the IP algorithm and show how the time required to find a good-enough solution depends on the selection of a sub-space in the IP algorithm
- …
