Reaction Systems (RSs) are a successful natural computing framework inspired by chemical reaction networks. A RS consists of a set of entities and a set of reactions. Entities can enable or inhibit each reaction, and are produced by reactions or provided by the environment. In a previous paper, we defined an original labelled transition system (LTS) semantics for RSs in the structural operational semantics (SOS) style. This approach has several advantages: (i) it provides a formal specification of the RS dynamics that enables the reuse of many formal analysis techniques and favors the implementation of tools, and (ii) it facilitates the definition of extensions of the RS framework by simply modifying some of the SOS rules in a modular way. In this paper, we demonstrate the extensibility of the framework by defining two quantitative variants of RSs: with reaction delays/durations, and with concentration levels. We provide a prototype logic programming implementation and apply our tool to a RS model of Th cells differentiation in the immune system.
Freeform optical systems include lens or mirror surfaces with shapes that have no axis of revolution in order to reach better performance in off-axis designs such as Head-Mounted Displays (HMD). They have an increased number of design parameters compared to rotationally symmetrical optical systems which makes their optimization more difficult. In this paper we presented results of several studies where various optimization techniques were used to optimize a freeform HMD prism with two mirrors. It was demonstrated that algorithms which performed well on traditional systems were not capable of obtaining high-quality solutions for the system under investigation. This paper presents a new parallel asynchronous memetic algorithm designed for optimizing freeform optical systems. The algorithm incorporates local search techniques into an asynchronous parallel computing procedure thus helping to speed-up the convergence to a high-quality solution. The results of a comparative study, presented in this paper along with the algorithm description, demonstrate that the proposed method is not only capable of obtaining a good solution but is also comparable with high-end commercial optical software packages.
We propose an analogy-based data augmentation approach for sentiment and style transfer named Analogy-Based Controllable Data Augmentation (ABCD). The object of data augmentation is to expand the number of sentences based on a limited amount of available data. We are given two unpaired corpora with different styles. In data augmentation, we retain the original text style while changing words to generate new sentences. We first train a self-attention-based convolutional neural network to compute the distribution of the contribution of each word to style in a given sentence. We call the words with high style contribution style-characteristic words. By substituting content words and style-characteristic words separately, we generate two new sentences. We use an analogy between the original sentence and these two additional sentences to generate another sentence. The results show that our proposed approach decrease perplexity by about 4 points and outperforms baselines on three transfer datasets.
This paper describes in details an innovative technique of injection of a global (or generally large-scale) quality measure into a deep neural network (DNN) in order to compensate for the tendency of DNNs to found the resulting classification virtually from a superposition of local neighbourhood transformations and projections. We used a state probability-like feature as the global quality measure and injected it into a DNN-based classifier deployed in a specific task of determining which parts of a web page are of certain interest for further processing by NLP techniques. Our goal was to decompose web sites of various internet discussion forums to useful content, i.e. the posts of users, and useless content, i.e. forum graphics, menus, banners, advertisements, etc.
Selection hyper-heuristics have proven to be effective in solving various real-world problems. Hyper-heuristics differ from traditional heuristic approaches in that they explore a heuristic space rather than a solution space. These techniques select constructive or perturbative heuristics to construct a solution or improve an existing solution respectively. Previous work has shown that the set of problem-specific heuristics made available to the hyper-heuristic for selection has an impact on the performance of the hyper-heuristic. Hence, there have been initiatives to determine the appropriate set of heuristics that the hyper-heuristic can select from. However, there has not been much research done in this area. Furthermore, previous work has focused on determining a set of heuristics that is used throughout the lifespan of the hyper-heuristic with no change to this set during the application of the hyper-heuristic. This paper investigates dynamic heuristic set selection (DHSS) which applies dominance to select the set of heuristics at different points during the lifespan of a selection hyper-heuristic. The DHSS approach was evaluated on the benchmark set for the CHeSC cross-domain hyper-heuristic challenge. DHSS was found to improve the performance of the best performing hyper-heuristic for this challenge.
Data classification is a real-world problem that is encountered daily in various problem domains. Genetic programming (GP) has proved to be one of the most versatile algorithms leading to its popularity as a classification algorithm. However, due to its large number of parameters, the manual design process of GP is considered to be a time consuming tedious task. As a result, there have been initiatives by the machine learning community to automate the design of GP classification algorithms. In this paper, we propose the transfer of the design knowledge gained from the automated design of GP classification algorithms from a specific source domain and apply it to design GP classification algorithms for a target domain. The results of the experiments demonstrate that the proposed approach is capable of evolving classifiers that achieve results that are competitive when compared to automated designed classifiers and better than manually tuned parameter classifiers. To the best of our knowledge, this is the first study that examines transfer learning in automated design. The proposed approach is shown to achieve positive transfer.
The task of generation constructive hyper-heuristics concerns itself with generating new heuristics for problem domains via some kind of mechanism that combines low-level heuristic components into new heuristics. The movie scene scheduling problem is a recently developed combinatorial problem for which there are relatively few low-level heuristics. This paper focused on the application of a novel ant-based generation constructive hyper-heuristic to develop new constructive heuristics for the problem. The ant-based generation constructive hyper-heuristic was applied to create components that were themselves produced from existing heuristics and domain knowledge regarding the movie scene scheduling problem. The results of the research demonstrated that the ant-based hyper-heuristic was successful in the domain. It outperformed the existing set of human-derived constructive heuristics across a wide variety of problem classes and over several instances within the movie scene scheduling problem. The success of this research suggests that other hyper-heuristic methods, such as a generation perturbative one, could be applied to the movie scene scheduling problem in the future.
The Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) is a popular algorithm for solving Multi-Objective Problems (MOPs). The main component of MOEA/D is to decompose a MOP into easier sub-problems using a set of weight vectors. The choice of the number of weight vectors significantly impacts the performance of MOEA/D. However, the right choice for this number varies, given different MOPs and search stages. We adaptively change the number of vectors by removing unnecessary vectors and adding new ones in empty areas of the objective space. Our MOEA/D variant uses the Consolidation Ratio to decide when to change the number of vectors and to decide where to add or remove these weighted vectors. We investigate the effects of this adaptive MOEA/D against MOEA/D with a poorly chosen set of vectors, a MOEA/D with fine-tuned vectors and MOEA/D with Adaptive Weight Adjustment on two commonly used benchmark functions. We analyse the algorithms in terms of hypervolume, IGD and entropy performance. Our results show that the proposed method is equivalent to MOEA/D with fine-tuned vectors and superior to MOEA/D with poorly defined vectors. Thus, our adaptive mechanism mitigates problems related to the choice of the number of weight vectors in MOEA/D, increasing the final performance of MOEA/D by filling empty areas of the objective space and avoiding premature stagnation of the search progress.
A collection of k mobile agents is arbitrarily deployed in the edges of a directed torus network where agents perpetually move to the successor edge. Each node has a switch that allows one agent of the two incoming edges to pass to its successor edge in every round. The goal is to obtain a switch scheduling to reach and maintain a configuration where the longest queue length is minimum. We consider a synchronous system. We use the concept of conflict graphs to model the local conflicts that occur with incident links. We show that there does not exist an algorithm that can reduce the number of agents in any conflict cycle of the conflict graph providing that all the links have at least 2 agents at every round. Hence, the lower bound is at least the average queue length of the conflict cycle with the maximum average queue length. Next, we present a centralized algorithm that computes a strategy in O ( n log n ) time for each round that attains the optimal queue length in O ( σ n ) rounds where n is the number of nodes in the network and σ is the standard deviations of the queue lengths in the initial setting. Our technique is based on network flooding on conflict graphs. Next, we consider a distributed system where nodes have access to the length of their queues and use communication to self-coordinate with nearby nodes. We present a local algorithm using only the information of the queue lengths at distance two. We show that the algorithm attains the optimal queue length in O ( σ C max 2 ) rounds where C max is the length of the longest conflict cycle with the maximum average queue length.
In the geometric Amoebot model, programmable matter is viewed as a very large number of identical micro/nano-sized entities, called particles , operating on a hexagonal tessellation of the plane, with limited computational capabilities, interacting only with neighboring particles, and moving from a grid node to an empty neighboring node. An important requirement, common to most research in this model, is that the particles must be connected at all times. Within this model, a central concern has been the formation of geometric shapes; in particular, the line is the elementary shape used as the basis to form more complex shapes, and as a step to solve complex tasks. If some of the particles on the line are faulty it might be necessary for the non-faulty particles to reconstruct a line that does not contain faulty particles. In this paper we study the Connected Line Recovery problem of reconstructing the line without violating the connectivity requirement. We provide a complete feasibility characterization of the problem, identifying the conditions necessary for its solvability, and constructively proving the sufficiency of those conditions. Our algorithm allows the non-faulty particles to solve the problem, regardless of the initial distribution of the faults and of their number.
For many Automatic Speech Recognition (ASR) tasks audio features as spectrograms show better results than Mel-frequency Cepstral Coefficients (MFCC), but in practice they are hard to use due to a complex dimensionality of a feature space. The following paper presents an alternative approach towards generating compressed spectrogram representation, based on Convolutional Variational Autoencoders (VAE). A Convolutional VAE model was trained on a subsample of the LibriSpeech dataset to reconstruct short fragments of audio spectrograms (25 ms) from a 13-dimensional embedding. The trained model for a 40-dimensional (300 ms) embedding was used to generate features for corpus of spoken commands on the GoogleSpeechCommands dataset. Using the generated features an ASR system was built and compared to the model with MFCC features.
Inspired by the NP-hardness of string folding problems modeling the natural process of protein folding, we discuss the idea of solving instances of NP-hard problems (e.g., string folding problems) of moderate size by letting artificially assembled proteins to fold. The accuracy with which one can combinatorially model the protein folding process, e.g., by string folding, as well as the precision with which one could experimentally estimate the energy of folded artificial proteins are crucial issues.
This paper proposes a three-player envy-free discrete assignment protocol of a divisible good, in which the utility of some portion of the good can be positive for some players and negative for the others. Such a good is called mixed manna. For mixed manna, current discrete envy-free cake-cutting or chore-division protocols cannot be applied. A naive protocol to achieve an envy-free division of mixed manna for three players needs an initial division of given mixed manna into eight pieces. This paper shows a new three-player envy-free discrete division protocol that needs an initial division into two pieces. After the initial division, it is shown that each of the pieces can be divided by modifying current envy-free cake-cutting and chore-division protocols.
Card-based protocols allow to perform secure multiparty computations using a deck of physical cards, and rely on shuffle actions such as the (normal) shuffle, the random cut, and the random bisection cut. A shuffle action is mathematically defined by a pair of a permutation set (which is a subset of the symmetric group) and a probability distribution on it; while one can theoretically consider any shuffle action in mind, he or she may be unable to determine whether it can be easily implemented by human hands. As one of the most general results, Koch and Walzer showed that any uniform closed shuffle (meaning that its permutation set is a subgroup and its distribution is uniform) can be implemented by human hands with the help of additional cards. However, there are several existing protocols which use non-uniform and/or non-closed shuffles. To implement these specific shuffles, Nishimura et al. proposed an idea of using (special) physical cases that can store piles of cards as well as Koch and Walzer proposed an implementation of a specific non-closed shuffle with additional cards. Because their implementations handle a limited class of non-uniform and/or non-closed shuffles, it is still open to find a general method for implementing any shuffle. In this paper, we solve the above problem; we implement “any” shuffle with only additional cards, provided that every probability of its distribution is a rational number. Therefore, our implementation works for any non-closed or non-uniform shuffle (if the distribution is rational as above).
This paper extends a PSO-based nonlinear regression technique to dynamic environments whereby the induced model dynamically adjusts when an environmental change is detected. As such, this work hybridizes a PSO designed for dynamic environments with a least-squares approximation technique to induce structurally optimal nonlinear regression models. The proposed model was evaluated experimentally and compared with the dynamic PSOs, namely multi-swarm, reinitialized, and charged PSOs, to optimize the model structure and the regression parameters in the dynamic environment. The obtained results show that the proposed model was adaptive to the changing environment to yield structurally optimal models which consequently, outperformed the dynamic PSOs for the given datasets.
This study looks at digital learning material as its qualified virtual practice/virtual experiment learning content and designs the best learning content to satisfy the existing definition of functions and services of digital learning material. The virtual practice/virtual experiment learning content service should meet the requirements. It is created considering a link to/synchronization with future learning content services built and operated by distance universities. This research defined learning model scenarios based on failure-experience learning model in oxygen titration for chemical experiments. The research developed the failure-experience learning scenario for chemical experiments' oxygen titration and developed a prototype for digital learning material service which will be provided to learners. It also conducted a survey asking learners on the developed virtual experiment learning material service. The survey results found that the service could be effective to improve the learning efficiency.
This paper attempts a new scheme of a semi-asynchronous parallel evolutionary algorithm (PEA), named time-limitation asynchronous PEA (TLAPEA). TLAPEA takes a balance between the search capability and the computational efficiency of PEA by synchronizing the solution evaluations within a particular waiting time before generating solutions. To reduce the idling time to wait for the slower evaluation of solutions, TLAPEA waits for a while for other solutions after the evaluation of a solution completes. The waiting time is decided from the average evaluation time of solutions and a new asynchrony parameter. This paper conducts an experiment to compare the proposed method with the full synchronous and asynchronous parallel evolutionary algorithm on multi-objective optimization problems. The experiment uses a state-of-the-art indicator-based multi-objective evolutionary algorithm, \(I_{\mathrm {SDE}}+\). Our experiment examines several variances of evaluation time on a parallel computing simulation. The experimental result reveals that TLAPEA with shorter time limitation obtains a high quality of solutions quicker than the synchronous and asynchronous ones regardless of the variance of the evaluation time.
Quantum-inspired algorithms are efficient for solving global search optimization problems. Nevertheless, their application is limited by two main requirements: a knowledge of a cost function and a big computational effort. To address both limitations, this paper presents a global optimization algorithm mixing a Quantum Diffusion Monte Carlo (DMC) method and an Evolution Strategy (ES). The proposed approach takes advantage of quantum models for efficiently finding solutions close the global optimum by applying the DMC. However, the DMC needs to be adapted to an optimization process. In order to improve relatively slow convergence rate of the DMC, we integrate a step size control for a diffusion displacement process of the DMC. Then, an (1+1)-ES is applied to find the minimum value of the cost function. Experimentation shows that the proposed approach can find global minima with fewer objective function evaluations than Quantum Particle Swarm Optimization and Quantum Annealing. Thus, it seems to be promising for solving black-box problems, where a reliable trade-off between exploration and exploitation is required.
Engineering approaches to machine learning (including robot learning) typically seek for the best learning algorithm for a particular problem, or a set problems. In contrast, the mammalian brain appears as a toolbox of different learning strategies, so that any newly encountered situation can be autonomously learned by an animal with a combination of existing learning strategies. For example, when facing a new navigation problem, a rat can either learn a map of the environment and then plan to find a path to its goal within this map. Alternatively, it can learn sequences of egocentric movements in response to identifiable features of the environment. For about 15 years, computational neuroscientists have searched for the mammalian brain’s coordination mechanisms which enable it to find efficient, if not necessarily optimal, combinations of existing learning strategies to solve new problems. Understanding such coordination principles of multiple learning strategies could have great implications in robotics, to enable robots to autonomously determine which learning strategies are appropriate in different contexts. Here, we review some of the main neuroscience models for the coordination of learning strategies and present some of the early results obtained when applying these models to robot learning. We moreover highlight important energy costs which can be reduced with such bio-inspired solutions compared to current deep reinforcement learning approaches. We conclude by sketching a roadmap for further developing such bio-inspired hybrid learning approaches to robotics.
The field of quantum information is becoming more known to the general public. However, effectively demonstrating the concepts underneath quantum science and technology to the general public can be a challenging job. We investigate, extend, and much expand here “quantum candies” (invented by Jacobs), a pedagogical model for intuitively describing some basic concepts in quantum information, including quantum bits, complementarity, the no-cloning principle, and entanglement. Following Jacob’s quantum candies description of the well known quantum key distribution protocol BB84, we explicitly demonstrate various additional quantum cryptography protocols using quantum candies in an approachable manner. The model we investigate can be a valuable tool for science and engineering educators who would like to help the general public to gain more insights about quantum science and technology: most parts of this paper, including many protocols for quantum cryptography, are expected to be easily understandable by a layperson without any previous knowledge of mathematics, physics, or cryptography.
