1,720,987 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
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
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
Learning short synfire chains by self-organization
Full text linkWe develop a model of cortical coding of stimuli by the sequences of activation patterns that they ignite in an initially random network. Hebbian learning then stabilizes these sequences, making them attractors of the dynamics. There is a competition between the capacity of the network and the stability of the sequences; for small stability parameter epsilon (the strength of the mean stabilizing PSP in the neurons in a learned sequence) the capacity is proportional to 1/epsilon2. For epsilon of the order of or less than the PSPs of the untrained network, the capacity exceeds that for sequences learned from tabula rasa
Large Barrier Trees for Studying Search
No full textBarrier trees are a method for representing the landscape structure of high-dimensional discrete spaces such as those that occur in the cost function of combinatorial optimization problems. The leaves of the tree represent local optima and a vertex where subtrees join represents the lowest cost saddle-point between the local optima in the subtrees. This paper introduces an extension to existing Barrier tree methods that make them more useful for studying heuristic optimization algorithms. It is shown that every configuration in the search space can be mapped onto a vertex in the Barrier tree. This provides additional information about the landscape, such as the number of configurations in a local optimum. It also allows the computation of additional statistics such as the correlation between configurations in different parts of the Barrier tree. Furthermore, the mappings allow the dynamic behavior of a heuristic search algorithms to be visualized. This extension is illustrated using an instance of the MAX-3-SAT problem
Learning to count objects in natural images for visual question answering
No full textVisual Question Answering (VQA) models have struggled with counting objects in natural images so far. We identify a fundamental problem due to soft attention in these models as a cause. To circumvent this problem, we propose a neural network component that allows robust counting from object proposals. Experiments on a toy task show the effectiveness of this component and we obtain state-of-the-art accuracy on the number category of the VQA v2 dataset without negatively affecting other categories, even outperforming ensemble models with our single model. On a difficult balanced pair metric, the component gives a substantial improvement in counting over a strong baseline by 6.6%
A Polynomially Searchable Exponential Neighbourhood for Graph Colouring
No full textIn this paper we develop a new graph colouring strategy. Our heuristic is an example of a so called "polynomially searchable exponential neighbourhood" approach. The neighbourhood is that of permutations of the colours of vertices of a subgraph. Our approach provides a solution method for colouring problems with edge weights. Results for initial tests on unweighted K-colouring benchmark problems are presented. Our colour permutation move was found in practice to be too slow to justify its use on these problems. By contrast, our implementation of iterative descent, which incorporates a permutation kickback move, performed extremely well. Moreover, our approach may yet prove valuable for weighted K-colouring. In addition, our approach offers an improved measure of the distance between colourings of a graph. Key words: Graph colouring, Optimisation, Heuristics, Local search, Exponential neighbourhood
Genetic Algorithm for Graph Colouring: Exploration of Galinier and Hao's algorithm
No full textThis paper examines the best current algorithm for solving the Chromatic Number Problem, due to Galinier and Hao (Journal of Combinatorial Optimization,1999, 3(4), pp 379-397). The algorithm combines a Genetic Algorithm with Tabu Search. We show that the algorithm remains powerful even if the Tabu Search component is eliminated, and explore the reasons for its success where other Genetic Algorithms have failed. In addition we propose a generalized algorithm for the Frequency Assignment Problem
Experiments in Bayesian Recommendation
Full text linkThe performance of collaborative filtering recommender systems can suffer when data is sparse, for example in distributed situations. In addition popular algorithms such as memory-based collaborative filtering are rather ad-hoc, making principled improvements difficult. In this paper we focus on a simple recommender based on naïve Bayesian techniques, and explore two different methods of modelling probabilities. We find that a Gaussian model for rating behaviour works well, and with the addition of a Gaussian-Gamma prior it maintains good performance even when data is sparse
- …
