Advancements in Natural Language Processing (NLP), have led to the emergence of Large Language Models (LLMs) such as GPT, Llama, Claude, and Gemini, which excel across a range of tasks but require extensive fine-tuning to align their outputs with human expectations. A widely used method for achieving this alignment is Reinforcement Learning from Human Feedback (RLHF), which, despite its success, faces challenges in accurately modelling human preferences. In this paper, we introduce GazeReward, a novel framework that integrates implicit feedback -- and specifically eye-tracking (ET) data -- into the Reward Model (RM). In addition, we explore how ET-based features can provide insights into user preferences. Through ablation studies we test our framework with different integration methods, LLMs, and ET generator models, demonstrating that our approach significantly improves the accuracy of the RM on established human preference datasets. This work advances the ongoing discussion on optimizing AI alignment with human values, exploring the potential of cognitive data for shaping future NLP research.

Evolutionary multi-objective optimization (EMO) is a highly active research field that has attracted many researchers and practitioners over the past three decades. Surprisingly, until now the goal of almost all EMO algorithms is to compute a suitable finite size representation of the Pareto set/front of a given MOP or at least a part of it. In other words, the quest is restricted to optimal solutions. In this work, we argue that the entire set of nearly optimal solutions – which includes all optimal ones – is of potential interest for the decision maker as they contain in addition to the optimal solutions alternative realizations or backup solutions. We further make a first effort to reliably compute the set of nearly optimal solutions via EMO algorithms. To this end, we first propose a new set of interest, NQ, and analyze its topology. In a next step, we propose an unbounded archiver that aims to capture NQ,and analyze it with respect to monotonicity and limit behavior. After this, we discuss the related subset selection problem which comes with unbounded archivers leading to four different algorithms. Finally, we numerically investigate the behavior of the archiver and the selection strategies, and present some results when using the archiver as external archiver to three widely used MOEAs indicating the benefit of the new approach.

Data imputation and data generation have important applications across many domains where incomplete or missing data can hinder accurate analysis and decision-making. Diffusion models have emerged as powerful generative models capable of capturing complex data distributions across various data modalities such as image, audio, and time series. Recently, they have been also adapted to generate tabular data. In this article, we propose a diffusion model for tabular data that introduces three key enhancements: (1) a conditioning attention mechanism, (2) an encoder–decoder transformer as the denoising network, and (3) dynamic masking. The conditioning attention mechanism is designed to improve the model’s ability to capture the relationship between the condition and synthetic data. The transformer layers help model interactions within the condition (encoder) or synthetic data (decoder), while dynamic masking enables our model to efficiently handle both missing data imputation and synthetic data generation tasks within a unified framework. We conduct a comprehensive evaluation by comparing the performance of diffusion models with transformer conditioning against state-of-the-art techniques such as Variational Autoencoders, Generative Adversarial Networks, and Diffusion Models, on benchmark datasets. Our evaluation focuses on the assessment of the generated samples with respect to three important criteria, namely: (1) machine learning efficiency, (2) statistical similarity, and (3) privacy risk mitigation. For the task of data imputation, we consider the efficiency of the generated samples across different levels of missing features. The results demonstrate average superior machine learning efficiency and statistical accuracy compared to the baselines, while maintaining privacy risks at a comparable level, particularly showing increased performance in datasets with a large number of features. By conditioning the data generation on a desired target variable, the model can mitigate systemic biases, generate augmented datasets to address data imbalance issues, and improve data quality for subsequent analysis. This has significant implications for domains such as healthcare and finance, where accurate, unbiased, and privacy-preserving data are critical for informed decision-making and fair model outcomes.

This is a corrigendum for the article “Diffusion Models for Tabular Data Imputation and Synthetic Data Generation” published in ACM Trans. Knowl. Discov. Data 19(6): 125:1-125:32 (2025).

Algorithms relating the amount of population's diversity to the elapsed period of execution have yielded important improvements. Particularly, schemes with a gradual shift from exploration to exploitation have excelled in several areas of Evolutionary Algorithms. A fairly recent method that applies this design principle is the Genetic Programming variant with Dynamic Management of Diversity (GP-DMD). GP-DMD applies a diversity-based replacement strategy that takes into account a user-defined function or policy that sets the amount of diversity desired in the population. Despite the improvements attained by GP-DMD, it is unable to precisely follow the user-defined policy in some cases. This calls into question its ability to perform a gradual shift from exploration to exploitation and hinders its extension to develop more complex dynamic and adaptive algorithms. This paper proposes the Genetic Programming variant with Controlled Dynamic Management of Diversity (GP-CDMD) which incorporates a novel replacement strategy that aims to improve its tracking capabilities. This is done through a probabilistic selection that takes into account the desired amount of diversity to restrict the diversity of the population. Results in the Symbolic Regression benchmark problem show a significant improvement in the tracking error, which results in features of the dynamics of the population that are more similar to the expected ones. This achievement facilitates the design of more complex diversity-based dynamic and adaptive optimizers and allows for better analyses on the implications of diversity in the GP area.

Most state-of-the-art Multiobjective Evolutionary Algorithms (moeas) promote the preservation of diversity of objective function space but neglect the diversity of decision variable space. The aim of this article is to show that explicitly managing the amount of diversity maintained in the decision variable space is useful to increase the quality of moeas when taking into account metrics of the objective space. Our novel Variable Space Diversity-based MOEA (vsd-moea) explicitly considers the diversity of both decision variable and objective function space. This information is used with the aim of properly adapting the balance between exploration and intensification during the optimization process. Particularly, at the initial stages, decisions made by the approach are more biased by the information on the diversity of the variable space, whereas it gradually grants more importance to the diversity of objective function space as the evolution progresses. The latter is achieved through a novel density estimator. The new method is compared with state-of-art moeas using several benchmarks with two and three objectives. This novel proposal yields much better results than state-of-the-art schemes when considering metrics applied on objective function space, exhibiting a more stable and robust behavior.

Current end-to-end approaches to Spoken Language Translation (SLT) rely on limited training resources, especially for multilingual settings. On the other hand, Multilingual Neural Machine Translation (MultiNMT) approaches rely on higher-quality and more massive data sets. Our proposed method extends a MultiNMT architecture based on language-specific encoders-decoders to the task of Multilingual SLT (Multi-SLT). Our method entirely eliminates the dependency from MultiSLT data and it is able to translate while training only on ASR and MultiNMT data. Our experiments on four different languages show that coupling the speech encoder to the MultiNMT architecture produces similar quality translations compared to a bilingual baseline (±0.2 BLEU) while effectively allowing for zero-shot MultiSLT. Additionally, we propose using an Adapter module for coupling the speech inputs. This Adapter module produces consistent improvements up to +6 BLEU points on the proposed architecture and +1 BLEU point on the end-to-end baseline.

Models based on attention mechanisms have shown unprecedented speech recognition performance. However, they are computationally expensive and unnecessarily complex for keyword spotting, a task targeted to small-footprint devices. This work explores the application of Lambda networks, an alternative framework for capturing long-range interactions without attention, for the keyword spotting task. We propose a novel ResNet-based model by swapping the residual blocks by temporal Lambda layers. Furthermore, the proposed architecture is built upon uni-dimensional temporal convolutions that further reduce its complexity. The presented model does not only reach state-of-the-art accuracies on the Google Speech Commands dataset, but it is 85% and 65% lighter than its Transformer-based (KWT) and convolutional (ResNet15) counterparts while being up to 100× faster. To the best of our knowledge, this is the first attempt to explore the Lambda framework within the speech domain and therefore, we unravel further research of new interfaces based on this architecture.

Menu planning is a complex task that involves finding a combination of menu items by taking into account several kinds of features, such as nutritional and economical, among others. In order to deal with the menu planning as an optimization problem, these features are transformed into constraints and objectives. Several variants of this problem have been defined and metaheuristics have been significantly successful solving them. In the last years, Memetic Algorithms (MAs) with explicit control of diversity have lead the attainment of high-quality solutions in several combinatorial problems. The main aim of this paper is to show that these types of methods are also viable for the menu planning problem. Specifically, a simple problem formulation based on transforming the menu planning into a single-objective constrained optimization problem is used. An MA that incorporates the use of iterated local search and a novel crossover operator is designed. The importance of incorporating an explicit control of diversity is studied. This is performed by using several well-known strategies to control the diversity, as well as a recently devised proposal. Results show that, for solving this problem in a robust way, the incorporation of explicit control of diversity and ad-hoc operators is mandatory.

End-to-end models for raw audio generation are a challenge, specially if they have to work with non-parallel data, which is a desirable setup in many situations. Voice conversion, in which a model has to impersonate a speaker in a recording, is one of those situations. In this paper, we propose Blow, a single-scale normalizing flow using hypernetwork conditioning to perform many-to-many voice conversion between raw audio. Blow is trained end-to-end, with non-parallel data, on a frame-by-frame basis using a single speaker identifier. We show that Blow compares favorably to existing flow-based architectures and other competitive baselines, obtaining equal or better performance in both objective and subjective evaluations. We further assess the impact of its main components with an ablation study, and quantify a number of properties such as the necessary amount of training data or the preference for source or target speakers.

One of the issues that might affect the performance of Differential Evolution (DE) is premature convergence. In such cases, and especially in long-term executions, due to the nature of the reproduction phase of DE, computational resources might not be employed efficiently. In DE this issue is usually tackled by altering the reproduction phase or by attaching an external archive. However, in other fields such as in combinatorial optimization, methods that amend the replacement phase have been very successful. In this paper, two variants of DE are extended by including a replacement phase that explicitly relates the amount of diversity maintained in the population with the number of generations evolved and with the stopping criterion. One of the variants is the classic DE/rand/1/bin whereas the other one is Success-History based Adaptive DE (SHADE), a state-of-the-art approach. In the case of the classic variant, the principles employed in combinatorial optimization could be used straightforwardly. However, when integrating it with state-of-the-art techniques that already incorporate mechanisms to preserve diversity, additional modifications were required. Experimental validation has been performed with the benchmarks provided for the 2013 IEEE Congress on Evolutionary Computation competition on real parameter optimization. The results of the best-ranked DE in such a competition could be improved further. Additionally, some problems that had not been solved to optimality ever by any proposal, could be successfully solved.

Most recombination operators are designed with the aim of altering its exploration capabilities depending on the distance between the parents involved in the process. However, relating the exploration capability of crossover operators only to the distance between parents might be a drawback in long-term executions because diversity could not be large enough after some generations, so the search process might stagnate. This paper proposes some extensions of the Simulated Binary Crossover (SBX) to generate dynamic variants of SBX (DSBX). The dynamic variants consider the stopping criterion to alter their internal operation. The main objective of the extensions is to induce a gradual change from exploration to intensification in the search process, which is performed by dynamically altering three different features of the original SBX. In order to validate the effectiveness of our proposal DSBX is integrated in three different state-of-the-art Multi-objective Evolutionary Algorithms (MOEAs). The experimental validation performed with the popular DTLZ, WFG and UF benchmarks shows a significant improvement of all the MOEAs when applying the novel crossover operator. Additionally, our proposal is also tested against schemes that incorporate differential evolution operators, showing quite competitive results.

Real-world problems complexity is often a consequence of the interdependence of the sub-problems that compose them. The Travelling Thief Problem (TTP) is a novel benchmark problem that aims to be a good model of this interdependence. It combines two classical well known problems: The Travelling Salesman Problem (TSP) and the Knapsack Problem (KP). Some state-of-the-art techniques, after a complex initialization, alternate between two different optimization stages, one focused on the tour and the other one on the selection of items. The optimization of the tour usually involves a simple trajectory-based search, meaning that important drawbacks might appear when the initial optimized TSP tours are not suitable for the TTP problem. In this paper a Guided Local Search (GLS) approach is proposed to improve further the tour optimization and compared against state-of-the-art techniques. Experimental validation shows the important benefits provided by our proposal, meaning that in fact many state-of-the-art techniques are too focused in a subset of the search space. Although at the moment the computational cost of our method is large, meaning that this approach does not scale well, new best-known solutions were generated in three well-known TTP instances. The main benefits were obtained in small and medium-sized instances.

Vehicle Routing Problem (VRP) is a widely known NP-Hard combinatorial optimization problem. This paper presents a proposal of a memetic algorithm (MA) with simulated annealing (SA) as trajectory-based method for solving the Capacitated Vehicle Routing Problem with Time Windows (CVRPTW). A novel crossover operator, the Single Breaking-point Sequence Based Crossover (SBSBX), is introduced and compared with a widely used operator, the Sequence-based Crossover (SBX). One of the principles behind the design of SBSBX is to reduce the disruptive behavior of SBX, with the aim of providing additional intensification. Initial studies show that the different crossover operators heavily impact the preservation of diversity in the population. Thus, two different parent-selection operators that induce different selection pressure are applied: random selection and binary tournament. The proposal is validated using the well-known Solomon's benchmark. The experimental validation shows that in some of the tested methods premature convergence is an important issue, whereas in other cases convergence is not attained. Overall, the combination of SBSBX and random selection attains the most promising results. In fact, a new best-known solution could be generated for one commonly used instance.

In recent years, several memetic algorithms with explicit mechanisms to delay convergence have shown great promise when solving 9×9 Sudoku puzzles. This paper analyzes and extends state-of-the-art schemes for dealing with Sudoku puzzles of larger dimensionality. Two interesting aspects are analyzed: the importance of the encoding and its relation with the way of managing the diversity. Specifically, three different ways of encoding the individuals and six different methods, including four that control the diversity in a special way, are studied. Computational results are shown with twenty 16×16 Sudoku puzzles. Contrary to the low-dimensional case, important differences appear among the several ways of controlling diversity. Specifically, a method that incorporates multi-objective concepts in the replacement phase to deal with the diversity, resulted in the most promising method. Results show that both the encoding and the way of managing diversity are crucial to attain high success probabilities in large Sudoku puzzles. They also show that, while the analyzed encodings induce different search space sizes, this feature is not enough to justify the differences in the performance attained by them.

Metaheuristics have yielded very promising results for the frequency assignment problem (FAP). However, the results obtainable using currently published methods are far from ideal in complex, large-scale instances. This paper applies and extends some of the most recent advances in evolutionary algorithms to two common variants of the FAP, and shows how, in traditional techniques, two common issues affect their performance: 1) premature convergence and 2) the way in which neutral networks are handled. A recent replacement-based diversity management strategy is successfully applied to alleviate the premature convergence drawback. Additionally, by properly defining a distance metric, the performance in the presence of neutrality can also be greatly improved. The replacement strategy combines the principle of transforming a single-objective problem into a multiobjective one by considering diversity as an additional objective, with the idea of adapting the balance induced between exploration and exploitation to the requirements of the different optimization stages. Tests with 44 publicly available instances yield very competitive results. New best-known frequency plans were generated for 11 instances, whereas in the remaining ones the best-known solutions were replicated. Comparisons with a large number of strategies designed to delay convergence of the population clearly show the advantages of our novel proposals.

The past few years have seen several variants of Evolutionary Algorithms (EAs) applied to solving Sudoku puzzles. Given that EAs with simple components do not work properly, considerable efforts have gone into designing ad-hoc evolutionary operators that profit from a knowledge of the problem. In this paper, we show that one of the main reasons for the improper behavior of EAs when dealing with difficult Sudoku puzzles is the appearance of premature convergence. Memetic algorithms with readily available genetic operators can be used to solve the hardest known Sudoku puzzles when general and well-known methods for avoiding premature convergence are incorporated. Among the approaches tested, a recently proposed method that is based on adopting multi-objective concepts for the solution of single-objective problems has shown remarkable performance. To our knowledge, the methods presented in this paper are the first EAs capable of solving the three Sudoku puzzles that are regarded as the most difficult ones known to date.

Premature convergence is one of the best-known drawbacks that affects the performance of evolutionary algorithms. An alternative for dealing with this problem is to explicitly try to maintain proper diversity. In this paper, a new replacement strategy that preserves useful diversity is presented. The novelty of our method is that it combines the idea of transforming a single-objective problem into a multiobjective one, by considering diversity as an explicit objective, with the idea of adapting the balance induced between exploration and exploitation to the various optimization stages. Specifically, in the initial phases, larger amounts of diversity are accepted. The diversity measure considered in this paper is based on calculating distances to the closest surviving individual. Analyses with a multimodal function better justify the design decisions and provide greater insight into the working operation of the proposal. Computational results with a packing problem that was proposed in a popular contest illustrate the usefulness of the proposal. The new method significantly improves on the best results known to date for this problem and compares favorably against a large number of state-of-the-art schemes.

People and objects will soon share the same digital network for information exchange in a world named as the age of the cyber-physical systems. The general expectation is that people and systems will interact in real-time. This poses pressure onto systems design to support increasing demands on computational power, while keeping a low power envelop. Additionally, modular scaling and easy programmability are also important to ensure these systems to become widespread. The whole set of expectations impose scientific and technological challenges that need to be properly addressed. The AXIOM project (Agile, eXtensible, fast I/O Module) will research new hardware/software architectures for cyber-physical systems to meet such expectations. The technical approach aims at solving fundamental problems to enable easy programmability of heterogeneous multi-core multi-board systems. AXIOM proposes the use of the task-based OmpSs programming model, leveraging low-level communication interfaces provided by the hardware. Modular scalability will be possible thanks to a fast interconnect embedded into each module. To this aim, an innovative ARM and FPGA-based board will be designed, with enhanced capabilities for interfacing with the physical world. Its effectiveness will be demonstrated with key scenarios such as Smart Video-Surveillance and Smart Living/Home (domotics).

The Traveling Salesman Problem (TSP) is one of the most well-known NP-hard combinatorial optimization problems. In order to deal with large TSP instances, several heuristics and metaheuristics have been devised. In this paper, a novel memetic scheme that incorporates a new diversity-based replacement strategy is proposed and applied to the largest instances of the TSPLIB benchmark. The novelty of our method is that it combines the idea of transforming a single-objective problem into a multi-objective one, by considering diversity as an explicit objective, with the idea of adapting the balance induced between exploration and exploitation to the various optimization stages. In addition, the intensification capabilities of the individual learning method incorporated in the memetic scheme are also adapted by taking into account the stopping criterion. Computational results show the clear superiority of our scheme when compared against state-of-the-art schemes. To our knowledge, our proposal is the first evolutionary scheme that readily solves an instance with more than 30,000 cities to optimality.

One of the main disadvantages of Evolutionary Algorithms (EAs) is that they converge towards local optima for some problems. In recent years, diversity-based multi-objective EAs have emerged as a promising technique to prevent from local optima stagnation when optimising single-objective problems. An additional drawback of EAs is the large dependency between the quality of the results provided and the setting of their parameters. By the use of parameter control methods, parameter values can be adapted during the run of an EA. The aim of control approaches is not only to improve the robustness of the controlled algorithm, but also to boost its efficiency. In this paper we apply a novel hybrid parameter control scheme based on Fuzzy Logic and Hyper-heuristics to simultaneously adapt several numeric and symbolic parameters of a diversity-based multi-objective EA. An extensive experimental evaluation is carried out, which includes a comparison between the hybrid control proposal and a wide range of configurations of the diversity-based multi-objective EA with fixed parameters. Results demonstrate that our control proposal is able to find similar or even better solutions than those obtained by the best configuration of the diversity-based scheme with fixed parameters in a significant number of benchmark problems, demonstrating the advantages of parameter control over parameter tuning for these test cases.

Differential Evolution (DE) is a very efficient meta-heuristic for optimization over continuous spaces which has gained much popularity in recent years. Several parameter control strategies have been proposed to automatically adapt its internal parameters. The most advanced DE variants take into account the feedback obtained in the optimization process to guide the dynamic setting of the DE parameters. Indeed, the automatic adaptation of the crossover rate (CR) has attracted a lot of research in the last decades. In most of such strategies, the quality of using a given CR value is measured by considering the probability of performing a replacement in the DE selection stage when such a value is applied. One of the main contributions of this paper is to experimentally show that the probability of replacement induced by the application of a given CR value and the quality of the obtained results are not as correlated as expected. This might cause a performance deterioration that avoids the achievement of good quality solutions even in the long-term. In addition, the experimental evaluation developed with a set of optimization problems of varying complexities clarifies some of the advantages and drawbacks of the different tested strategies. The only component varied among the different tested schemes has been the CR control strategy. The study presented in this paper provides advances in the understanding of the inner working of several state-of-the-art adaptive DE variants.

The maintenance of a proper diversity is an important issue for the correct behavior of Evolutionary Algorithms (EAs). The loss of diversity might lead to stagnation in suboptimal regions, producing the effect known as “premature convergence”. Several methods to avoid premature convergence have been previously proposed. Among them, the use of Multi-objective Evolutionary Algorithms (MOEAs) is a promising approach. Several ways of using MOEAs for single-objective optimization problems have been devised. The use of an additional objective based on calculating the diversity that each individual introduces in the population has been successfully applied by several researchers. Several ways of measuring the diversity have also been tested. In this work, the main weaknesses of some of the previously presented approaches are analyzed. Considering such drawbacks, a new scheme whose aim is to maintain a better diversity than previous approaches is proposed. The proposed approach is empirically validated using a set of well-known single-objective benchmark problems. Our preliminary results indicate that the proposed approach provides several advantages in terms of premature convergence avoidance. An analysis of the convergence in the average-case is also carried out. Such an analysis reveals that the better ability of our proposed approach to deal with premature convergence produces a reduction in the convergence speed in the average-case for several of the benchmark problems adopted.

Evolutionary Algorithms (EAs) are one of the most popular strategies for solving optimisation problems. To define a configuration of an EA several components and parameters must be specified. Therefore, one of the main drawbacks of EAs is the complexity of their parameter setting. Another problem is that EAs might have a tendency to converge towards local optima for many problems. For this reason, several methods to deal with local optima stagnation have been designed. Multiobjectivisation, which consists in the reformulation of mono-objective problems as multi-objective ones, is one of such methods. Some multiobjectivisation methods require the specification of parameters by the user. In some cases, the quality of the obtained solutions has been improved by these methods. However, they usually introduce more components and parameters into the optimisation scheme. The main contribution of this work is to deeply analyse the robustness of multiobjectivisation approaches with parameters. Several large scale continuous optimisation problems have been multiobjectivised in order to perform such a study. Extracted conclusions might allow designing methods which profit from multiobjectivisation with parameters, without incorporating additional parameters to the whole optimisation scheme. By this way, the parameter setting could be performed in an easier way. The experimental evaluation has provided promising results.

The goal of face detection is to determine the presence of faces in arbitrary images, along with their locations and dimensions. As it happens with any graphics workloads, these algorithms benefit from data-level parallelism. Existing parallelization efforts strictly focus on mapping different divide and conquer strategies into multicore CPUs and GPUs. However, even the most advanced single-chip many-core processors to date are still struggling to effectively handle real-time face detection under high-definition video workloads. To address this challenge, face detection algorithms typically avoid computations by dynamically evaluating a boosted cascade of classifiers. Unfortunately, this technique yields a low ALU occupancy in architectures such as GPUs, which heavily rely on large SIMD widths for maximizing data-level parallelism. In this paper we present several techniques to increase the performance of the cascade evaluation kernel, which is the most resource-intensive part of the face detection pipeline. Particularly, the usage of concurrent kernel execution in combination with cascades generated with the Gentle Boost algorithm solves the problem of GPU underutilization, and achieves a 5X speedup in 1080p videos on average over the fastest known implementations, while slightly improving the accuracy. Finally, we also studied the parallelization of the cascade training process and its scalability under SMP platforms. The proposed parallelization strategy exploits both task and data-level parallelism and achieves a 3.5X speedup over single-threaded implementations.

This work presents a novel two-step algorithm to estimate the\norientation of speakers in a smart-room environment equipped\nwith microphone arrays. First the position of the speaker is\nestimated by the SRP-PHAT algorithm, and the time delay of\narrival for each microphone pair with respect to the detected\nposition is computed. In the second step, the value of the cross-\ncorrelation at the estimated time delay is used as the fundamen-\ntal characteristic from where to derive the speaker orientation. The proposed method performs consistently better than other state-of-the-art acoustic techniques with a purposely recorded database and the CLEAR head pose database.

Evolutionary Algorithms (EAs) are one of the most popular strategies for solving optimisation problems. Several variants of EAs are seen to exist. They usually have several components and parameters which must be fixed. Therefore, one of the main drawbacks of EAs is the complexity of their parameter setting. Multiobjectivisation consists in the reformulation of mono-objective problems as multi-objective ones. Multiobjectivisation has been used in several fields as a mechanism to avoid premature convergence in local optima. However, since they usually introduce more components and parameters into the optimisation scheme, they hinder even more the parameter setting of an EA. A hyper heuristic can be viewed as a heuristic that iteratively chooses between a set of given low-level (meta)-heuristics in order to solve an optimisation problem. Hence, hyper heuristics have been used as an approach to facilitate the application of EAs. In this work, a parallel hyper heuristic is applied to a set of well-known optimisation benchmark problems. The contribution of the work is twofold. First, the adaptation level - amount of considered historical knowledge - of the hyper heuristic is analysed. Moreover, the contribution of considering multiobjectivisation inside the model is studied. Computational results show the benefits of parallel hyper heuristics and multiobjectivisation.

Simultaneous speech poses a challenging problem for conventional speaker diarization systems. In meeting data, a substantial amount of missed speech error is due to speaker overlaps, since usually only one speaker label per segment is assigned. Furthermore, simultaneous speech included in training data can lead to corrupt speaker models and thus worse segmentation performance. In this paper, we propose the use of three spatial cross-correlation-based features together with spectral information for speaker overlap detection on distant microphones. Different microphone-pair data are fused by means of principal component analysis. We have obtained an improvement of the speaker diarization system over the baseline by discarding overlap segments from model training and assigning two speaker labels to them according to likelihoods in Viterbi decoding. In experiments conducted on the AMI Meeting corpus, we achieve a relative DER reduction of 11.2% and 17.0% for single- and multi-site data, respectively. The improvement of clustering with techniques such as beamforming and TDOA-feature stream also leads to a higher effectiveness of the overlap labeling algorithm. Preliminary experiments with NIST RT data show DER improvement on the RT'09 meeting recordings as well.

This work presents a set of approaches used to deal with the Frequency Assignment Problem (FAP), which is one of the key issues in the design of Global System for Mobile Communications (GSM) networks. The used formulation of the fap is focused on aspects which are relevant for real-world GSM networks. The best up to date frequency plans for the considered version of the fap had been obtained by using parallel memetic algorithms. However, such approaches suffer from premature convergence with some real world instances. Multiobjectivisation is a technique which transforms a mono-objective optimisation problem into a multi-objective one with the aim of avoiding stagnation. A Multiobjectivised Memetic Algorithm, based on the well-known Non-Dominated Sorting Genetic Algorithm II (NSGA-II) together with its required operators, is presented in this paper. Several multiobjectivised schemes, based on the addition of an artificial objective, are analysed. They have been combined with a novel crossover operator. Computational results obtained for two different real-world instances of the fap demonstrate the validity of the proposed model. The new model provides benefits in terms of solution quality, and in terms of time saving. The previously known best frequency plans for both tested real-world networks have been improved.

Bin Packing problems are NP-hard problems with many practical applications. A variant of a Bin Packing Problem was proposed in the GECCO 2008 competition session. The best results were achieved by a mono-objective Memetic Algorithm (MA). In order to reduce the execution time, it was parallelised using an island-based model. High quality results were obtained for the proposed instance. However, subsequent studies concluded that stagnation may occur for other instances. The term multiobjectivisation refers to the transformation of originally mono-objective problems as multi-objective ones. Its main aim is to avoid local optima. In this work, a multiobjectivised MA has been applied to the gecco 2008 Bin Packing Problem. Several multiobjectivisation schemes, which use problem-dependent and problem-independent information have been tested. Also, a parallelisation of the multiobjectivised MA has been developed. Results have been compared with the best up to date mono-objective approaches. Computational results have demonstrated the validity of the proposals. They have provided benefits in terms of solution quality, and in terms of time saving.

Most research on Strip Packing Problems is focused on the single-objective formulation of the problem. However, in this work we deal with a more general and practical variant of the problem, which not only seeks to optimize the usage of the raw material, but also the production process. For the problem solution, we have applied some of the most-known multi-objective evolutionary algorithms, since they have shown a promising behavior when affording multi-objective real-world problems. For an initial implementation, we proposed a solution codification which is based on a complete representation of the pattern layouts. Such an approach was promising but wasn't suitable to afford large instances. For this reason, we have focused on the design of a codification which can be much more competitive when compared to some tailor-made methods. In this sense, we present a hyperheuristic-based codification as an alternative to combine heuristics in such a way that a heuristic's strengths make up for the drawbacks of another. Results demonstrate the advantage of using multi-objective approaches, hyperheuristic-based representations, and of course, the importance on the selection of appropriate solution codifications.

A substantial portion of errors of the conventional speaker\ndiarization systems on meeting data can be accounted to overlapped\nspeech. This paper proposes the use of several spatial\nfeatures to improve speech overlap detection on distant channel\nmicrophones. These spatial features are integrated into a\nspectral-based system by using principal component analysis\nand neural networks. Different overlap detection hypotheses\nare used to improve diarization performance with both overlap\nexclusion and overlap labeling. In experiments conducted\non AMI Meeting Corpus we demonstrate a relative DER improvement\nof 11.6% and 14.6% for single- and multi-site data,\nrespectively.

This work presents several approaches used to deal with the 2D packing problem proposed in the GECCO 2008 contest session. A memetic algorithm, together with the specifically designed local search and variation operators, are presented. A novel parallel model was used to parallelize the approach. The model is a hybrid algorithm which combines a parallel island-based scheme with a hyperheuristic approach. An adaptive behavior is added to the island-based model by applying the hyperheuristic procedure. The main operation of the island-based model is kept, but the configurations of the memetic algorithms executed on each island are dynamically mapped. The model grants more computational resources to those configurations that show a more promising behavior. For this purpose a specific criterion was designed in order to select the configurations with better success expectations. Computational results obtained for the contest problem demonstrate the validity of the proposed model. The best reported solutions for the problem contest instance have been achieved by using the here presented approaches.

The detection of the acoustic events (AEs) that are naturally \nproduced in a meeting room may help to describe the human \nand social activity that takes place in it. When applied to \nspontaneous recordings, the detection of AEs from only audio \ninformation shows a large amount of errors, which are mostly due to temporal overlapping of sounds. In this paper, a system to detect and recognize AEs using both audio and video information is presented. A feature-level fusion strategy is used, and the structure of the HMM-GMM based system considers each class separately and uses a one-against-all strategy for training. Experime ntal AED results with a new and rather spontaneous dataset are presented which show the advantage of the proposed approach.

ULL∷A-Team tool is a library that provides a skeleton to solve multi-objective optimization problems by applying evolutionary algorithms. In addition to providing sequential implementations of some of the best-known evolutionary algorithms, the skeleton provides great flexibility in obtaining parallel schemes. This flexibility is achieved by specifying configurations that allow the execution of different parallel evolutionary models: homogeneous island-based model, heterogeneous island-based model and self-adaptive island-based model. To solve a particular problem, the user must specify all its properties by defining a set of C++classes. Additionally, the user can also incorporate new evolutionary algorithms to the tool. This work explains how to carry out this task using IBEA algorithm as a case study. In order to check the contribution of the new algorithm, the computational results obtained for the multi-objective knapsack problem are presented.

The radio network design (RND) is an NP-hard optimization problem which consists of the maximization of the coverage of a given area while minimizing the base station deployment. Solving RND problems efficiently is relevant to many fields of application and has a direct impact in the engineering, telecommunication, scientific, and industrial areas. Numerous works can be found in the literature dealing with the RND problem, although they all suffer from the same shortfall: a noncomparable efficiency. Therefore, the aim of this paper is twofold: first, to offer a reliable RND comparison base reference in order to cover a wide algorithmic spectrum, and, second, to offer a comprehensible insight into accurate comparisons of efficiency, reliability, and swiftness of the different techniques applied to solve the RND problem. In order to achieve the first aim we propose a canonical RND problem formulation driven by two main directives: technology independence and a normalized comparison criterion. Following this, we have included an exhaustive behavior comparison between 14 different techniques. Finally, this paper indicates algorithmic trends and different patterns that can be observed through this analysis.

This paper presents the graphical user interface (GUI) for a remote service used to solve the 2-dimensional guillotine cutting stock problem as applied to the textile industry. This interface allows for the patterns in the regional dress of the city of La Orotava, in the Canary Islands, to be defined and manipulated. The user chooses the garment, quantity and sizes to be made, as well as the amount of material available. This information is then used by the system to arrange the patterns. Individual patterns may also be selected. Internally, the system implements a variant of the Viswanathan-Bagchi algorithm to solve the 2-dimensional cutting stock problem. The solution is calculated remotely in a dedicated server. The user may, however, keep track of all those possibilities rejected by the algorithm in its search for the final solution. He can also modify the final solution by dragging the patterns with the mouse into the desired position on the surface. The graphical user interface and the remote service were implemented using Java, while the solving algorithm was coded using C/C++.

This work presents a new parallel model for the solution of multi-objective optimization problems. The model combines a parallel island-based scheme with a hyperheuristic approach in order to raise the level of generality at which most current evolutionary algorithms operate. This way, a wider range of problems can be tackled since the strengths of one algorithm can compensate for the weaknesses of another. Computational results demonstrate that the model grants more computational resources to those algorithms that show a more promising behaviour.

The Frequency Assignment Problem (FAP) is one of the key issues in the design of GSM networks (Global System for Mobile communications), and will remain important in the foreseeable future. There are many versions of FAP, most of them benchmarking-like problems. We use a formulation of FAP, developed in published work, that focuses on aspects which are relevant for real-world GSM networks. In this paper, we have designed, adapted, and evaluated several types of metaheuristic for different time ranges. After a detailed statistical study, results indicate that these metaheuristics are very appropriate for this FAP. New interference results have been obtained, that significantly improve those published in previous research.

This paper proposes an initialization for an agglomerative\nsystem applied to speaker diarization in the meeting environment.\nThe initialization is based on a previous clustering of the\ntemporal sequence generated by the estimation of the Time Delay\nof Arrival (TDOA) among pair of sensors. That initial clustering\nhas the purpose of obtaining initial classes with speaker\ninformation from a sole speaker. The aim is to ensure the purity\nof the initial segments based on the position of the speakers in\na meeting along time. The TDOA initialization was tested with\nthe dataset used in the RT07s evaluation where an improvement\nof the diariazation error rate is obtained with respect to the classical\nuniform initialization. The most of the experiments show\nthat the purity of the beginning segments leads to a better clustering\non the posterior hierarchical strategy based on cepstral\nfeatures.

This paper presents a novel approach to speaker orientation\nestimation in a SmartRoom environment equipped with\nmultiple microphones. The ratio between the high and low\nband energies (HLBR) received at each microphone has been\nshown in our previous work to be a potentially approach to estimate\nthe direction of the voice produced by a speaker. In this\nwork, for each microphone pair, a smoothed CPS phase is obtained\nby a proper windowing of the main peak of the crosscorrelation\nsequence estimated with the GCC-PHAT method,\nand a HLBR is computed from the processed CPS. The proposed\nmethod keeps the computational simplicity of the HLBR\nalgorithm while adding the robustness offered by the GCCPHAT\ntechnique. Experimental preliminary results were conducted\nover a database recorded purposely in the UPC Smart\nroom, and over the CLEAR head pose database. The proposed\nmethod performs consistently better than other state-of-the-art\ntechniques with both databases.

This work analyses the difficulties of parallelizing the best known sequential algorithm for the 2D cutting stock problem. All the approaches to parallelize the algorithm strive against its highly irregular computation structure and its sequential nature. A distributed-memory parallel algorithm has been designed through a time-driven task intercommunication service. The service allows to introduce a load balancing scheme that tries to hide the non-homogeneous work load nature of the involved single tasks. Experimental results obtained for MVAPICH infiniband and MPICH gigabit Ethernet implementations prove the efficiency of the communication and balancing schemes and show the almost linear speedup achieved by the parallel algorithm.

Many real-world problems are based on the optimization of more than one objective function. This work presents a tool for the resolution of multi-objective optimization problems based on the cooperation of a set of algorithms. The invested time in the resolution is decreased by means of a parallel implementation of an evolutionary team algorithm. This model keeps the advantages of heterogeneous island models but also allows to assign more computational resources to the algorithms with better expectations. The elitist scheme applied aims to improve the results obtained with single executions of independent evolutionary algorithms. The user solves the problem without the need of knowing the internal operation details of the used evolutionary algorithms. The computational results obtained on a cluster of PCs for some tests available in the literature are presented.

This paper presents a multimodal approach to head pose estimation and 3D gaze orientation of individuals in a SmartRoom environment equipped with multiple cameras and microphones. We first introduce the two monomodal approaches as reference. In video, we estimate head orientation from color information by exploiting spatial redundancy among cameras. Audio information is processed to estimate the direction of the voice produced by a speaker making use of the directivity characteristics of the head radiation pattern. Two multimodal information fusion schemes working at data and decision levels are analyzed in terms of accuracy and robustness of the estimation. Experimental results conducted over the CLEAR evaluation database are reported and the comparison of the proposed multimodal head pose estimation algorithms with the reference monomodal approaches proves the effectiveness of the proposed approach.

The head orientation of human speakers in a smart-room affects the quality of the signals recorded by far-field microphones, and consequently influences the performance of the technologies deployed based on those signals. Additionally, knowing the orientation in these environments can be useful for the development of several multimodal advanced services, for instance, in microphone network management. Consequently, head orientation estimation has recently become a growing interesting research topic. In this paper, we propose two different approaches to head orientation estimation on the basis of multimicrophone recordings: first, an approach based on the generalization of the well-known SRP-PHAT speaker localization algorithm, and second a new approach based on measurements of the ratio between the high and the low band speech energies. Promising results are obtained in both cases, with a generalized better performance of the algorithms based on speaker localization methods. Index Terms: head orientation estimation, microphone arrays, speaker tracking

This work presents a skeleton for the resolution of multi-objective optimization problems using the improved version of the strength Pareto evolutionary algorithm (SPEA2). From the same problem specification, the skeleton derives sequential and distributed parallel solvers. The user interface for the problem definition consists of a set of classes and methods which are described in detail. The internal implementation of both solvers and their configuration parameters are explained. An application example to solve the optimization of a broadcasting strategy in metropolitan MANETs is given. The computational results obtained for this example in a homogeneous cluster of PCs give evidence of the quality of the approach

Reliable measures of speaker positions are needed for computational perception of human activities taking place in a smartroom environment. In this work, it is described the development process and the experiments conducted in the design and implementation of an Audio Person Tracking system for smart-room environments. The proposed system is based on the SRP-PHAT algorithm, as it is known to perform robustly in most environmental conditions. Novelties proposed are aimed to enhance the accuracy of the system independently on the application scenario and to reduce the computational complexity. Index Terms: sound localization, source tracking, microphone arrays.

Reliable measures of speaker positions are needed for computational perception of human activities taking place in a smart-room environment. In this work, we investigate the effect of talkers head orientation on the accuracy of acoustical source localization techniques and its relation with the talker directivity pattern and room reverberation. Two different representative speaker localization techniques are assessed, steered response power and a crossing lines based method, in both cases on the basis of the estimated delays between pairs of microphones with the GCC-PHAT algorithm. A small database has been collected at the UPC’s smart room for evaluation. The results show how the localization error heavily depends on the head orientation, and also the fact that the space exploration based technique is much more robust to head orientation changes than the crossing lines technique, due to the way the contributions from the various microphones are combined.