| Cosson R, Santana R, Derbel B and Liefooghe A (2022), "Multi-objective NK landscapes with heterogeneous objectives", In Proceedings of the Genetic and Evolutionary Computation Conference. , pp. 502-510.
[Abstract] [BibTeX] [URL]
|
| Abstract: So far, multi-objective NK landscapes have been investigated under the assumption of a homogeneous nature of the involved objectives in terms of difficulty. However, we argue that problems with heterogeneous objectives, e.g., in terms of multi-modality, can be challenging for multi-objective evolutionary algorithms, and deserve further considerations. In this paper, we propose a model of multi-objective NK landscapes, where each objective has a different degree of variable interactions (𝐾), as a benchmark to investigate heterogeneous multi-objective optimization problems. We show that the use of a rank-annotated neighborhood network with labeled local optimal solutions, together with landscape metrics extracted from the heterogeneous objectives, thoroughly characterize bi-objective NK landscapes with a different level of heterogeneity among the objectives |
BibTeX:
@inproceedings{Cosson_et_al:2022,
author = {Cosson, Raphaël and Santana, Roberto and Derbel, Bilel and Liefooghe, Arnaud},
title = {Multi-objective NK landscapes with heterogeneous objectives},
booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference},
year = {2022},
pages = {502--510},
url = {https://dl.acm.org/doi/abs/10.1145/3512290.3528858}
}
|
| Lima RH, Magalhães D, Pozo A, Mendiburu A and Santana R (2022), "A grammar-based GP approach applied to the design of deep neural networks", Genetic Programming and Evolvable Machines. , pp. 1-26. Springer.
[Abstract] [BibTeX] [URL]
|
| Abstract: Deep Learning has been very successful in automating the feature engineering process, widely applied for various tasks, such as speech recognition, classification, segmentation of images, time-series forecasting, among others. Deep neural networks (DNNs) incorporate the power to learn patterns through data, following an end-to-end fashion and expand the applicability in real world problems, since less pre-processing is necessary. With the fast growth in both scale and complexity, a new challenge has emerged regarding the design and configuration of DNNs. In this work, we present a study on applying an evolutionary grammar-based genetic programming algorithm (GP) as a unified approach to the design of DNNs. Evolutionary approaches have been growing in popularity for this subject as Neuroevolution is studied more. We validate our approach in three different applications: the design of Convolutional Neural Networks for image classification, Graph Neural Networks for text classification, and U-Nets for image segmentation. The results show that evolutionary grammar-based GP can efficiently generate different DNN architectures, adapted to each problem, employing choices that differ from what is usually seen in networks designed by hand. This approach has shown a lot of promise regarding the design of architectures, reaching competitive results with their counterparts. |
BibTeX:
@article{Lima_et_al:2022,
author = {Lima, Ricardo HR and Magalhães, Dimmy and Pozo, Aurora and Mendiburu, Alexander and Santana, Roberto},
title = {A grammar-based GP approach applied to the design of deep neural networks},
journal = {Genetic Programming and Evolvable Machines},
publisher = {Springer},
year = {2022},
pages = {1--26},
url = {https://link.springer.com/article/10.1007/s10710-022-09432-0}
}
|
| Mei N, Santana R and Soto D (2022), "Informative neural representations of unseen contents during higher-order processing in human brains and deep artificial networks", Nature Human Behaviour. Vol. 6(5), pp. 720-731. Nature Publishing Group.
[Abstract] [BibTeX] [URL]
|
| Abstract: A framework to pinpoint the scope of unconscious processing is critical to improve models of visual consciousness. Previous research observed brain signatures of unconscious processing in visual cortex, but these were not reliably identified. Further, whether unconscious contents are represented in high-level stages of the ventral visual stream and linked parieto-frontal areas remains unknown. Using a within-subject, high-precision functional magnetic resonance imaging approach, we show that unconscious contents can be decoded from multi-voxel patterns that are highly distributed alongside the ventral visual pathway and also involving parieto-frontal substrates. Classifiers trained with multi-voxel patterns of conscious items generalized to predict the unconscious counterparts, indicating that their neural representations overlap. These findings suggest revisions to models of consciousness such as the neuronal global workspace. We then provide a computational simulation of visual processing/representation without perceptual sensitivity by using deep neural networks performing a similar visual task. The work provides a framework for pinpointing the representation of unconscious knowledge across different task domains. |
BibTeX:
@article{Mei_et_al:2022,
author = {Mei, Ning and Santana, Roberto and Soto, David},
title = {Informative neural representations of unseen contents during higher-order processing in human brains and deep artificial networks},
journal = {Nature Human Behaviour},
publisher = {Nature Publishing Group},
year = {2022},
volume = {6},
number = {5},
pages = {720--731},
url = {https://www.nature.com/articles/s41562-021-01274-7}
}
|
| Murua M, Galar D and Santana R (2022), "Solving the multi-objective Hamiltonian cycle problem using a Branch-and-Fix based algorithm", Journal of Computational Science. Vol. 60, pp. 101578. Elsevier.
[Abstract] [BibTeX] [URL]
|
| Abstract: The Hamiltonian cycle problem consists of finding a cycle in a given graph that passes through every single vertex exactly once, or determining that this cannot be achieved. In this investigation, a graph is considered with an associated set of matrices. The entries of each of the matrix correspond to a different weight of an arc. A multi-objective Hamiltonian cycle problem is addressed here by computing a Pareto set of solutions that minimize the sum of the weights of the arcs for each objective. Our heuristic approach extends the Branch-and-Fix algorithm, an exact method that embeds the problem in a stochastic process. To measure the efficiency of the proposed algorithm, we compare it with a multi-objective genetic algorithm in graphs of a different number of vertices and density. The results show that the density of the graphs is critical when solving the problem. The multi-objective genetic algorithm performs better (quality of the Pareto sets) than the proposed approach in random graphs with high density; however, in these graphs it is easier to find Hamiltonian cycles, and they are closer to the multi-objective traveling salesman problem. The results reveal that, in a challenging benchmark of Hamiltonian graphs with low density, the proposed approach significantly outperforms the multi-objective genetic algorithm. |
BibTeX:
@article{Murua_et_al:2022,
author = {Murua, M and Galar, D and Santana, R},
title = {Solving the multi-objective Hamiltonian cycle problem using a Branch-and-Fix based algorithm},
journal = {Journal of Computational Science},
publisher = {Elsevier},
year = {2022},
volume = {60},
pages = {101578},
url = {https://www.sciencedirect.com/science/article/pii/S1877750322000151}
}
|
| Santana R, Liefooghe A and Derbel B (2022), "Boomerang-shaped neural embeddings for NK landscapes", In Proceedings of the Genetic and Evolutionary Computation Conference. , pp. 858-866.
[Abstract] [BibTeX] [URL]
|
| Abstract: Understanding the landscape underlying NK models is of fundamental interest. Different representations have been proposed to better understand how the ruggedness of the landscape is influenced by the model parameters, such as the problem dimension, the degree of non-linearity and the structure of variable interactions. In this paper, we propose to use neural embedding, that is a continuous vectorial representation obtained as a result of applying a neural network to a prediction task, in order to investigate the characteristics of NK landscapes. The main assumption is that neural embeddings are able to capture important features that reflect the difficulty of the landscape. We propose a method for constructing NK embeddings, together with metrics for evaluating to what extent this embedding space encodes valuable information from the original NK landscape. Furthermore, we study how the embedding dimensionality and the parameters of the NK model influence the characteristics of the NK embedding space. Finally, we evaluate the performance of optimizers that solve the continuous representations of NK models by searching for solutions in the embedding space. |
BibTeX:
@inproceedings{Santana_et_al:2022,
author = {Santana, Roberto and Liefooghe, Arnaud and Derbel, Bilel},
title = {Boomerang-shaped neural embeddings for NK landscapes},
booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference},
year = {2022},
pages = {858--866},
url = {https://dl.acm.org/doi/abs/10.1145/3512290.3528856}
}
|
| Santana R (2022), "An embedding space for SARS-CoV-2 epitope-based vaccines", In European Journal of Clinical Investigation. Vol. 52
[Abstract] [BibTeX]
|
| Abstract: Epitopes are relatively small peptide chains which play an important role in the immune response. They are identified by T cells and B cells which participate in the immune response in humans. Epitope-based vaccines synthesize epitopes that when inoculated stimulate the natural response to a pathogen. This type of vaccines have been proposed for pathogens such as influenza, tuberculosis, dengue, and more recently SARS-CoV-2. In this talk we address the problem of learning an embedding representation of epitopes useful for the design of epitope vaccines. Different criteria should be taken into account in the design of an epitope vaccine, including the biding affinity of the epitopes to one or more major histocompatibility complex (MHC) alleles, the extent to which they cover haplotype distribution of the target population, etc. The goal of the epitope embedding design is capturing in the representation specific immunogenic characteristics of the epitopes in relation to the different MHC alleles. Starting from an original set of peptides (peptide vocabulary), e.g., those extracted from the genome of a pathogen, we propose methods to generate artificial sequences of such peptides. The sequence generation method is used to create large datasets of sequences (vaccine corpora) which are assumed to exhibit some latent semantics related to the way epitopes interact among them and with alleles to provide an immunogenic effect. We use the corpora to create neural epitope embeddings learned in an unsupervised way. We them explore the space of embeddings and discuss how to use them to define intrinsic and extrinsic tasks related to vaccine design. Using a large set of SARS-CoV-2 T cell epitope candidates, we show how to address the vaccine design problem in the epitope embedding space, and provide evidence that such embeddings can be used for solving downstream tasks related to epitope-based vaccine design. |
BibTeX:
@inproceedings{Santana:2022a,
author = {Santana, R},
title = {An embedding space for SARS-CoV-2 epitope-based vaccines},
booktitle = {European Journal of Clinical Investigation},
year = {2022},
volume = {52}
}
|
| Santana R and Shakya S (2022), "Evolutionary approaches with adaptive operators for the bi-objective TTP problem",
In 2022 IEEE Symposium Series on Computational Intelligence (IEEE SSCI). Singapore , pp. 1-8.
[Abstract] [BibTeX] [URL]
|
| Abstract: One characteristic feature of the traveling thief problem (TTP) is the existence of different facets of instance difficulty that make no single optimization algorithm to excel over the others. Multi-objective variants of the TTP add to the challenging features of the single-objective TTP the difficulties associated to keep a set of diverse non-dominated solutions. In this paper we propose an adaptive hybrid TTP optimization algorithm based on the probabilistic application of variation operators. The algorithm combines variation operators conceived for different facets of instance difficulty and adapts their frequency of application to the characteristics of the instances. The introduced algorithm shows a robust behavior across different bi-objective TTP instances producing results competitive with those achieved by state of the art algorithms |
BibTeX:
@inproceedings{Santana_and_Shakya:2022,
author={R. Santana and S. Shakya},
title={Evolutionary approaches with adaptive operators for the bi-objective TTP problem},
booktitle={2022 IEEE Symposium Series on Computational Intelligence (IEEE SSCI)},
year={2022},
pages={1-8},
url={https://ieeessci2022.org/index.html}
}
|
| Vadillo J and Santana R (2022), "On the human evaluation of universal audio adversarial perturbations", Computers & Security. Vol. 112, pp. 102495.
[Abstract] [BibTeX] [DOI] [URL]
|
| Abstract: Human-machine interaction is increasingly dependent on speech communication, mainly due to the remarkable performance of Machine Learning models in speech recognition tasks. However, these models can be fooled by adversarial examples, which are inputs intentionally perturbed to produce a wrong prediction without the changes being noticeable to humans. While much research has focused on developing new techniques to generate adversarial perturbations, less attention has been given to aspects that determine whether and how the perturbations are noticed by humans. This question is relevant since high fooling rates of proposed adversarial perturbation strategies are only valuable if the perturbations are not detectable. In this paper we investigate to which extent the distortion metrics proposed in the literature for audio adversarial examples, and which are commonly applied to evaluate the effectiveness of methods for generating these attacks, are a reliable measure of the human perception of the perturbations. Using an analytical framework, and an experiment in which 36 subjects evaluate audio adversarial examples according to different factors, we demonstrate that the metrics employed by convention are not a reliable measure of the perceptual similarity of adversarial examples in the audio domain. |
BibTeX:
@article{Vadillo_and_Santana:2022,
author = {Jon Vadillo and Roberto Santana},
title = {On the human evaluation of universal audio adversarial perturbations},
journal = {Computers & Security},
year = {2022},
volume = {112},
pages = {102495},
url = {https://www.sciencedirect.com/science/article/pii/S0167404821003199},
doi = {10.1016/j.cose.2021.102495}
}
|
| Vadillo J, Santana R and Lozano JA (2022), "Analysis of dominant classes in universal adversarial perturbations", Knowledge-Based Systems. Vol. 236, pp. 107719.
[Abstract] [BibTeX] [DOI] [URL]
|
| Abstract: The reasons why Deep Neural Networks are susceptible to being fooled by adversarial examples remains an open discussion. Indeed, many different strategies can be employed to efficiently generate adversarial attacks, some of them relying on different theoretical justifications. Among these strategies, universal (input-agnostic) perturbations are of particular interest, due to their capability to fool a network independently of the input in which the perturbation is applied. In this work, we investigate an intriguing phenomenon of universal perturbations, which has been reported previously in the literature, yet without a proven justification: universal perturbations change the predicted classes for most inputs into one particular (dominant) class, even if this behavior is not specified during the creation of the perturbation. In order to justify the cause of this phenomenon, we propose a number of hypotheses and experimentally test them using a speech command classification problem in the audio domain as a testbed. Our analyses reveal interesting properties of universal perturbations, suggest new methods to generate such attacks and provide an explanation of dominant classes, under both a geometric and a data-feature perspective. |
BibTeX:
@article{Vadillo_et_al:2022,
author = {Jon Vadillo and Roberto Santana and Jose A. Lozano},
title = {Analysis of dominant classes in universal adversarial perturbations},
journal = {Knowledge-Based Systems},
year = {2022},
volume = {236},
pages = {107719},
url = {https://www.sciencedirect.com/science/article/pii/S0950705121009643},
doi = {10.1016/j.knosys.2021.107719}
}
|