Search

Deep learning fostered a leap ahead in automated skin lesion analysis in the last two years. Those models, however, are expensive to train and difficult to parameterize. Objective: We investigate methodological issues for designing and evaluating deep learning models for skin lesion analysis. We explore ten choices faced by researchers: use of transfer learning, model architecture, train dataset, image resolution, type of data augmentation, input normalization, use of segmentation, duration of training, additional use of Support Vector Machines, and test data augmentation. Methods: We perform two full factorial experiments, for five different test datasets, resulting in 2560 exhaustive trials in our main experiment, and 1280 trials in our assessment of transfer learning. We analyze both with multi-way analyses of variance (ANOVA). We use the exhaustive trials to simulate sequential decisions and ensembles, with and without the use of privileged information from the test set. Results main experiment: Amount of train data has disproportionate influence, explaining almost half the variation in performance. Of the other factors, test data augmentation and input resolution are the most influential. Deeper models, when combined, with extra data, also help. — transfer experiment: Transfer learning is critical, its absence brings huge performance penalties. — simulations: Ensembles of models are the best option to provide reliable results with limited resources, without using privileged information and sacrificing methodological rigor. Conclusions and Significance: Advancing research on automated skin lesion analysis requires curating larger public datasets. Indirect use of privileged information from the test set to design the models is a subtle, but frequent methodological mistake that leads to overoptimistic results. Ensembles of models are a cost-effective alternative to the expensive full-factorial and to the unstable sequential designs.

International audience; Continual learning aims to learn tasks sequentially, with (often severe) constraints on the storage of old learning samples, without suffering from catastrophic forgetting. In this work, we propose prescient continual learning, a novel experimental setting, to incorporate existing information about the classes, prior to any training data. Usually, each task in a traditional continual learning setting evaluates the model on present and past classes, the latter with a limited number of training samples. Our setting adds future classes, with no training samples at all. We introduce Ghost Model, a representation-learning-based model for continual learning using ideas from zero-shot learning. A generative model of the representation space in concert with a careful adjustment of the losses allows us to exploit insights from future classes to constraint the spatial arrangement of the past and current classes. Quantitative results on the AwA2 and aP\&Y datasets and detailed visualizations showcase the interest of this new setting and the method we propose to address it.

Despite the growing availability of high-quality public datasets, the lack of training samples is still one of the main challenges of deep-learning for skin lesion analysis. Generative Adversarial Networks (GANs) appear as an enticing alternative to alleviate the issue, by synthesizing samples indistinguishable from real images, with a plethora of works employing them for medical applications. Nevertheless, carefully designed experiments for skin-lesion diagnosis with GAN-based data augmentation show favorable results only on out-of-distribution test sets. For GAN-based data anonymization $-$ where the synthetic images replace the real ones $-$ favorable results also only appear for out-of-distribution test sets. Because of the costs and risks associated with GAN usage, those results suggest caution in their adoption for medical applications., Accepted to the ISIC Skin Image Analysis Workshop @ CVPR 2021

The field of neural networks has seen significant advances in recent years with the development of deep and convolutional neural networks. Although many of the current works address real-valued models, recent studies reveal that neural networks with hypercomplex-valued parameters can better capture, generalize, and represent the complexity of multidimensional data. This paper explores the quaternion-valued convolutional neural network application for a pattern recognition task from medicine, namely, the diagnosis of acute lymphoblastic leukemia. Precisely, we compare the performance of real-valued and quaternion-valued convolutional neural networks to classify lymphoblasts from the peripheral blood smear microscopic images. The quaternion-valued convolutional neural network achieved better or similar performance than its corresponding real-valued network but using only 34% of its parameters. This result confirms that quaternion algebra allows capturing and extracting information from a color image with fewer parameters.

Lifelong learning has attracted much attention, but existing works still struggle to fight catastrophic forgetting and accumulate knowledge over long stretches of incremental learning. In this work, we propose PODNet, a model inspired by representation learning. By carefully balancing the compromise between remembering the old classes and learning new ones, PODNet fights catastrophic forgetting, even over very long runs of small incremental tasks --a setting so far unexplored by current works. PODNet innovates on existing art with an efficient spatial-based distillation-loss applied throughout the model and a representation comprising multiple proxy vectors for each class. We validate those innovations thoroughly, comparing PODNet with three state-of-the-art models on three datasets: CIFAR100, ImageNet100, and ImageNet1000. Our results showcase a significant advantage of PODNet over existing art, with accuracy gains of 12.10, 6.51, and 2.85 percentage points, respectively. Code is available at https://github.com/arthurdouillard/incremental_learning.pytorch, Comment: Accepted at ECCV 2020

Data-driven models are now deployed in a plethora of real-world applications - including automated diagnosis - but models learned from data risk learning biases from that same data. When models learn spurious correlations not found in real-world situations, their deployment for critical tasks, such as medical decisions, can be catastrophic. In this work we address this issue for skin-lesion classification models, with two objectives: finding out what are the spurious correlations exploited by biased networks, and debiasing the models by removing such spurious correlations from them. We perform a systematic integrated analysis of 7 visual artifacts (which are possible sources of biases exploitable by networks), employ a state-of-the-art technique to prevent the models from learning spurious correlations, and propose datasets to test models for the presence of bias. We find out that, despite interesting results that point to promising future research, current debiasing methods are not ready to solve the bias issue for skin-lesion models., Comment: Accepted to the ISIC Skin Image Analysis Workshop @ CVPR 2020

Segmenting skin lesions images is relevant both for itself and for assisting in lesion classification, but suffers from the challenge in obtaining annotated data. In this work, we show that segmentation may improve with less data, by selecting the training samples with best inter-annotator agreement, and conditioning the ground-truth masks to remove excessive detail. We perform an exhaustive experimental design considering several sources of variation, including three different test sets, two different deep-learning architectures, and several replications, for a total of 540 experimental runs. We found that sample selection and detail removal may have impacts corresponding, respectively, to 12% and 16% of the one obtained by picking a better deep-learning model., Accepted to the ISIC Skin Image Analysis Workshop @ CVPR 2020

An ensemble method should cleverly combine a group of base classifiers to yield an improved classifier. The majority vote is an example of a methodology used to combine classifiers in an ensemble method. In this paper, we propose to combine classifiers using an associative memory model. Precisely, we introduce ensemble methods based on recurrent correlation associative memories (RCAMs) for binary classification problems. We show that an RCAM-based ensemble classifier can be viewed as a majority vote classifier whose weights depend on the similarity between the base classifiers and the resulting ensemble method. More precisely, the RCAM-based ensemble combines the classifiers using a recurrent consult and vote scheme. Furthermore, computational experiments confirm the potential application of the RCAM-based ensemble method for binary classification problems.

Hypercomplex-valued neural networks, including quaternion-valued neural networks, can treat multi-dimensional data as a single entity. In this paper, we present the quaternion-valued recurrent projection neural networks (QRPNNs). Briefly, QRPNNs are obtained by combining the non-local projection learning with the quaternion-valued recurrent correlation neural network (QRCNNs). We show that QRPNNs overcome the cross-talk problem of QRCNNs. Thus, they are appropriate to implement associative memories. Furthermore, computational experiments reveal that QRPNNs exhibit greater storage capacity and noise tolerance than their corresponding QRCNNs., Comment: arXiv admin note: substantial text overlap with arXiv:1909.09227

Recent literature has explored automated pornographic detection - a bold move to replace humans in the tedious task of moderating online content. Unfortunately, on scenes with high skin exposure, such as people sunbathing and wrestling, the state of the art can have many false alarms. This paper is based on the premise that incorporating motion information in the models can alleviate the problem of mapping skin exposure to pornographic content, and advances the bar on automated pornography detection with the use of motion information and deep learning architectures. Deep Learning, especially in the form of Convolutional Neural Networks, have striking results on computer vision, but their potential for pornography detection is yet to be fully explored through the use of motion information. We propose novel ways for combining static (picture) and dynamic (motion) information using optical flow and MPEG motion vectors. We show that both methods provide equivalent accuracies, but that MPEG motion vectors allow a more efficient implementation. The best proposed method yields a classification accuracy of 97.9% - an error reduction of 64.4% when compared to the state of the art - on a dataset of 800 challenging test cases. Finally, we present and discuss results on a larger, and more challenging, dataset.

Hypercomplex-valued neural networks, including quaternion-valued neural networks, can treat multi-dimensional data as a single entity. In this paper, we introduce the quaternion-valued recurrent projection neural networks (QRPNNs). Briefly, QRPNNs are obtained by combining the non-local projection learning with the quaternion-valued recurrent correlation neural network (QRCNNs). We show that QRPNNs overcome the cross-talk problem of QRCNNs. Thus, they are appropriate to implement associative memories. Furthermore, computational experiments reveal that QRPNNs exhibit greater storage capacity and noise tolerance than their corresponding QRCNNs., Accepted to be Published in: Proceedings of the 8th Brazilian Conference on Intelligent Systems (BRACIS 2019), October 15-18, 2019, Salvador, BA, Brazil

Convolutional neural networks (CNNs) deliver exceptional results for computer vision, including medical image analysis. With the growing number of available architectures, picking one over another is far from obvious. Existing art suggests that, when performing transfer learning, the performance of CNN architectures on ImageNet correlates strongly with their performance on target tasks. We evaluate that claim for melanoma classification, over 9 CNNs architectures, in 5 sets of splits created on the ISIC Challenge 2017 dataset, and 3 repeated measures, resulting in 135 models. The correlations we found were, to begin with, much smaller than those reported by existing art, and disappeared altogether when we considered only the top-performing networks: uncontrolled nuisances (i.e., splits and randomness) overcome any of the analyzed factors. Whenever possible, the best approach for melanoma classification is still to create ensembles of multiple models. We compared two choices for selecting which models to ensemble: picking them at random (among a pool of high-quality ones) vs. using the validation set to determine which ones to pick first. For small ensembles, we found a slight advantage on the second approach but found that random choice was also competitive. Although our aim in this paper was not to maximize performance, we easily reached AUCs comparable to the first place on the ISIC Challenge 2017., Comment: ISIC Skin Image Analysis Workshop @ CVPR 2019

An autoassociative memory is an input-output system designed for the storage and recall of a finite set of items. In this work, we present the class of complete lattice projection autoassociative memories (CLPAMs). A CLPAM is a non-distributive autoassociative memory defined by a neural network with a hidden layer of morphological neurons. More importantly, a CLPAM is formulated using only the partial ordering of a complete lattice. As an example of CLPAM, we introduce the so called distance-based projection autoassociative memories (DBPAMs) which exhibit an excellent tolerance to salt-and-pepper noise.

A pet that goes missing is among many people's worst fears: a moment of distraction is enough for a dog or a cat wandering off from home. Some measures help matching lost animals to their owners; but automated visual recognition is one that - although convenient, highly available, and low-cost - is surprisingly overlooked. In this paper, we inaugurate that promising avenue by pursuing face recognition for dogs. We contrast four ready-to-use human facial recognizers (EigenFaces, FisherFaces, LBPH, and a Sparse method) to two original solutions based upon convolutional neural networks: BARK (inspired in architecture-optimized networks employed for human facial recognition) and WOOF (based upon off-the-shelf OverFeat features). Human facial recognizers perform poorly for dogs (up to 60.5% accuracy), showing that dog facial recognition is not a trivial extension of human facial recognition. The convolutional network solutions work much better, with BARK attaining up to 81.1% accuracy, and WOOF, 89.4%. The tests were conducted in two datasets: Flickr-dog, with 42 dogs of two breeds (pugs and huskies); and Snoopybook, with 18 mongrel dogs., Comment: 17 pages, 8 figures, 1 table, Multimedia Tools and Applications

Mathematical morphology is a theory with applications in image processing and analysis. This paper presents a quantale-based approach to color morphology based on the CIELab color space in spherical coordinates. The novel morphological operations take into account the perceptual difference between color elements by using a distance-based ordering scheme. Furthermore, the novel approach allows for the use of non-flat structuring elements. An illustrative example reveals that non-flat dilations and erosions may preserve more features of a color image than their corresponding flat operations. Furthermore, the novel non-flat morphological operators yielded promising results on experiments concerning the detection of the boundaries of objects on color images.

Dilation and erosion are two elementary operations from mathematical morphology, a non-linear lattice computing methodology widely used for image processing and analysis. The dilation-erosion perceptron (DEP) is a morphological neural network obtained by a convex combination of a dilation and an erosion followed by the application of a hard-limiter function for binary classification tasks. A DEP classifier can be trained using a convex-concave procedure along with the minimization of the hinge loss function. As a lattice computing model, the DEP classifier assumes the feature and class spaces are partially ordered sets. In many practical situations, however, there is no natural ordering for the feature patterns. Using concepts from multi-valued mathematical morphology, this paper introduces the reduced dilation-erosion (r-DEP) classifier. An r-DEP classifier is obtained by endowing the feature space with an appropriate reduced ordering. Such reduced ordering can be determined using two approaches: one based on an ensemble of support vector classifiers (SVCs) with different kernels and the other based on a bagging of similar SVCs trained using different samples of the training set. Using several binary classification datasets from the OpenML repository, the ensemble and bagging r-DEP classifiers yielded mean higher balanced accuracy scores than the linear, polynomial, and radial basis function (RBF) SVCs as well as their ensemble and a bagging of RBF SVCs.

In this paper, we present a dynamic wildfire warning map that combines both spatial and weather information. In particular, our wildfire early warning model is obtained by aggregating two indexes called wildfire risk and wildfire danger. The wildfire risk index, which is based on georeferenced features such as altitude and forest type, measures the fuel necessary for a wildfire to start at a certain location on a map. The wildfire danger uses weather conditions to yield temporal information concerning the possibility of a wildfire to spread. Machine learning techniques and fuzzy logic operations are used to determine the wildfire risk and danger indexes from available data. Although both wildfire risk and wildfire danger indexes can be used separately, using concepts from fuzzy logic, they can be combined to yield a wildfire warning system that takes into account both weather and static information. We illustrate the wildfire early warning model by considering weather and geographical data for the state of Acre.

Prior art on automated screening of diabetic retinopathy and direct referral decision shows promising performance; yet most methods build upon complex hand-crafted features whose performance often fails to generalize. Objective We investigate data-driven approaches that extract powerful abstract representations directly from retinal images to provide a reliable referable diabetic retinopathy detector. Methods We gradually build the solution based on convolutional neural networks, adding data augmentation, multi-resolution training, robust feature-extraction augmentation, and a patient-basis analysis, testing the effectiveness of each improvement. Results The proposed method achieved an area under the ROC curve of 98.2% (95% CI: 97.4–98.9%) under a strict cross-dataset protocol designed to test the ability to generalize — training on the Kaggle competition dataset and testing using the Messidor-2 dataset. With a 5 × 2-fold cross-validation protocol, similar results are achieved for Messidor-2 and DR2 datasets, reducing the classification error by over 44% when compared to most published studies in existing literature. Conclusion Additional boost strategies can improve performance substantially, but it is important to evaluate whether the additional (computation- and implementation-) complexity of each improvement is worth its benefits. We also corroborate that novel families of data-driven methods are the state of the art for diabetic retinopathy screening. Significance: By learning powerful discriminative patterns directly from available training retinal images, it is possible to perform referral diagnostics without detecting individual lesions.

Autoassociative memories (AMs) are models inspired by the human ability to store and retrieve information by association. A fuzzy associative memory is used for the storage and recall of fuzzy sets. Inspired by the fuzzy morphological associative memories, we recently introduced the class of max-C projection fuzzy autoassociative memories (maxC PFAMs). In few words, a max-C PFAM projects the input vector into the set of all max-C combinations of the stored vectors. In this paper, we present its dual version, the class of min-D PFAMs, which projects an input into the set of all min-D combinations. In this paper we also address some theoretical issues of the two PFAM models.

In this paper we introduce the class of fuzzy kernel associative memories (fuzzy KAMs). Fuzzy KAMs are derived from single-step generalized exponential bidirectional fuzzy associative memories by interpreting the exponential of a fuzzy similarity measure as a kernel function. The output of a fuzzy KAM is obtained by summing the desired responses weighted by a normalized evaluation of the kernel function. Furthermore, in this paper we propose to estimate the parameter of a fuzzy KAM by maximizing the entropy of the model. We also present two approaches for pattern classification using fuzzy KAMs. Computational experiments reveal that fuzzy KAM-based classifiers are competitive with well-known classifiers from the literature.

Skin cancer is by far the most common type of cancer. Early detection is the key to increase the chances for successful treatment significantly. Currently, Deep Neural Networks are the state-of-the-art results on automated skin cancer classification. To push the results further, we need to address the lack of annotated data, which is expensive and require much effort from specialists. To bypass this problem, we propose using Generative Adversarial Networks for generating realistic synthetic skin lesion images. To the best of our knowledge, our results are the first to show visually-appealing synthetic images that comprise clinically-meaningful information.

Mathematical morphology is a theory with applications in image processing and analysis. In a supervised approach to mathematical morphology, pixel values are ranked according to sets of foreground and background elements specified a priori by the user. In this paper, we introduce a supervised fuzzy color-based approach to color mathematical morphology that provides an elegant alternative to the support vector machine-based approach developed by Velasco-Forero and Angulo. Briefly, color elements are ranked according to the degree of truth of the proposition “the considered color is a foreground color but it is not a background color” in the new supervised color morphological approach. Furthermore, the vagueness and uncertainty inherent to the description of colors by humans can be naturally incorporated in the new approach using the concept of fuzzy colors.

Deep learning models show remarkable results in automated skin lesion analysis. However, these models demand considerable amounts of data, while the availability of annotated skin lesion images is often limited. Data augmentation can expand the training dataset by transforming input images. In this work, we investigate the impact of 13 data augmentation scenarios for melanoma classification trained on three CNNs (Inception-v4, ResNet, and DenseNet). Scenarios include traditional color and geometric transforms, and more unusual augmentations such as elastic transforms, random erasing and a novel augmentation that mixes different lesions. We also explore the use of data augmentation at test-time and the impact of data augmentation on various dataset sizes. Our results confirm the importance of data augmentation in both training and testing and show that it can lead to more performance gains than obtaining new images. The best scenario results in an AUC of 0.882 for melanoma classification without using external data, outperforming the top-ranked submission (0.874) for the ISIC Challenge 2017, which was trained with additional data.

Max-C projection autoassociative fuzzy memories (max-C PAFMs) are memory models designed for the storage and recall of fuzzy sets. In few words, a max-C PAFM projects the input fuzzy set into the family of all max-C combinations of the stored items. In this paper, we focus on a particular max-C PAFM called Zadeh max-C PAFM. The Zadeh max-C PAFM is the most robust max-C PAFM with respect to dilative noise. Furthermore, by masking the noise contained in a corrupted input, it exhibits excellent tolerance to any kind of noise. Besides introducing the Zadeh max-C PAFM, in this paper we point out a potential application of the Zadeh max-C PAFM for face recognition.

Autoassociative morphological memories (AMMs) are robust and computationally efficient memory models with unlimited storage capacity. In this paper, we present the max-plus and min-plus projection autoassociative morphological memories (PAMMs) as well as their compositions. Briefly, the max-plus PAMM yields the largest max-plus combination of the stored vectors which is less than or equal to the input. Dually, the vector recalled by the min-plus PAMM corresponds to the smallest min-plus combination which is larger than or equal to the input. Apart from unlimited absolute storage capacity and one step retrieval, PAMMs and their compositions exhibit an excellent noise tolerance. Furthermore, the new memories yielded quite promising results in classification problems with a large number of features and classes.

Knowledge transfer impacts the performance of deep learning -- the state of the art for image classification tasks, including automated melanoma screening. Deep learning's greed for large amounts of training data poses a challenge for medical tasks, which we can alleviate by recycling knowledge from models trained on different tasks, in a scheme called transfer learning. Although much of the best art on automated melanoma screening employs some form of transfer learning, a systematic evaluation was missing. Here we investigate the presence of transfer, from which task the transfer is sourced, and the application of fine tuning (i.e., retraining of the deep learning model after transfer). We also test the impact of picking deeper (and more expensive) models. Our results favor deeper models, pre-trained over ImageNet, with fine-tuning, reaching an AUC of 80.7% and 84.5% for the two skin-lesion datasets evaluated., 4 pages

Automatically detecting violence in videos is paramount for enforcing the law and providing the society with better policies for safer public places. In addition, it may be essential for protecting minors from accessing inappropriate contents on-line, and for helping parents choose suitable movie titles for their children. However, this is an open problem as the very definition of violence is subjective and may vary from one society to another. Detecting such nuances from video footages with no human supervision is very challenging. Clearly, when designing a computer-aided solution to this problem, we need to think of efficient (quickly harness large troves of data) and effective detection methods (robustly filter what needs special attention and further analysis). In this vein, we explore a content description method for violence detection founded upon temporal robust features that quickly grasp video sequences, automatically classifying violent videos. The used method also holds promise for fast and effective classification of other recognition tasks (e.g., pornography and other inappropriate material). When compared to more complex counterparts for violence detection, the method shows similar classification quality while being several times more efficient in terms of runtime and memory footprint.

Face recognition systems are gaining momentum with current developments in computer vision. At the same time, tactics to mislead these systems are getting more complex, and counter-measure approaches are necessary. Following the current progress with convolutional neural networks (CNN) in classification tasks, we present an approach based on transfer learning using a pre-trained CNN model using only static features to recognize photo, video or mask attacks. We tested our approach on the REPLAY-ATTACK and 3DMAD public databases. On the REPLAY-ATTACK database our accuracy was 99.04% and the half total error rate (HTER) of 1.20%. For the 3DMAD, our accuracy was of 100.00% and HTER 0.00%. Our results are comparable to the state-of-the-art.

An autoassociative memory (AM) that projects an input pattern onto a linear subspace is referred to as a subspace projection AM (SPAM). The optimal linear AM (OLAM), which can be used for the storage and recall of real-valued patterns, is an example of SPAM. In this brief we introduce a novel SPAM model based on the robust M-estimation method. In contrast to the OLAM and many other associative memory models, the robust SPAM represents a neural network in which the synaptic weights are iteratively adjusted during the retrieval phase. Computational experiments concerning the reconstruction of corrupted gray-scale images reveal that the novel memories exhibit an excellent tolerance with respect to salt and pepper noise as well as some tolerance with respect to Gaussian noise and blurred input images.

In recent years, deep architectures have been used for transfer learning with state-of-the-art performance in many datasets. The properties of their features remain, however, largely unstudied under the transfer perspective. In this work, we present an extensive analysis of the resiliency of feature vectors extracted from deep models, with special focus on the trade-off between performance and compression rate. By introducing perturbations to image descriptions extracted from a deep convolutional neural network, we change their precision and number of dimensions, measuring how it affects the final score. We show that deep features are more robust to these disturbances when compared to classical approaches, achieving a compression rate of 98.4%, while losing only 0.88% of their original score for Pascal VOC 2007., This article corresponds to the accepted version at IEEE ICIP 2016. We will link the DOI as soon as it is available

Adversarial examples have raised questions regarding the robustness and security of deep neural networks. In this work we formalize the problem of adversarial images given a pretrained classifier, showing that even in the linear case the resulting optimization problem is nonconvex. We generate adversarial images using shallow and deep classifiers on the MNIST and ImageNet datasets. We probe the pixel space of adversarial images using noise of varying intensity and distribution. We bring novel visualizations that showcase the phenomenon and its high variability. We show that adversarial images appear in large regions in the pixel space, but that, for the same task, a shallow classifier seems more robust to adversarial images than a deep convolutional network., Comment: Copyright 2016 IEEE. This manuscript was accepted at the IEEE International Joint Conference on Neural Networks (IJCNN) 2016. We will link the published version as soon as the DOI is available

Advantages of semiconducting small molecules-as opposed to semiconducting polymers-include synthetic simplicity, monodispersity, low cost, and ease of purification. One purported disadvantage of small-molecule films is reduced mechanical robustness. This paper measures the tensile modulus and crack-onset strain for pure films of the high-performance solution-processable small-molecule donors 7,7'-[4,4-bis(2-ethylhexyl)-4H-silolo[3,2-b:4,5-b']dithiophene-2,6-diyl]bis[6-fluoro-4-(5'-hexyl-[2,2'-bithiophen]-5-yl)benzo[c][1,2,5]thiadiazole] (DTS(FBTTh2)2), 2,5-di(2-ethylhexyl)-3,6-bis(5″-n-hexyl-[2,2',5',2″]terthiophen-5-yl)-pyrrolo[3,4-c]pyrrole-1,4-dione (SMDPPEH), and 6,13-bis(triisopropylsilylethynyl)pentacene (TIPS-pentacene), the acceptor 5,5'-(2,1,3-benzothiadiazole-4,7-diyldi-2,1-ethenediyl)bis[2-hexyl-1H-isoindole-1,3(2H)-dione] (HPI-BT), blends of DTS(FBTTh2)2 and SMDPPEH with [6,6]-phenyl C71 butyric acid methyl ester (PC71BM) and with HPI-BT, and bulk heterojunction films processed with the additives 1,8-diiodooctane (DIO) and polystyrene (PS). The most deformable films of solution-processed organic semiconductors are found to exhibit tensile moduli and crack-onset strains comparable to those measured for conjugated polymers. For example, the tensile modulus of as-cast DTS(FBTTh2)2 is 0.68 GPa (i.e., comparable to poly(3-hexylthiophene) (P3HT), the common polymer), while it exhibits no cracks when stretched on an elastomeric substrate to strains of 14%. While this high degree of stretchability is lost upon the addition of PC71BM (4.2 GPa, 1.42%), it can be partially recovered using processing additives. Tensile modulus and crack-onset strain are highly correlated, which is typical of van der Waals solids. Increased surface roughness was correlated to increased modulus and brittleness within films of similar composition. Decreased stiffness for soluble molecular semiconductors can be rationalized by the presence of alkyl side chains, which decrease the van der Waals attraction between molecules in the crystalline grains. These measurements and observations could have important consequences for the stability of devices based on molecular semiconductors, especially those destined for stretchable or ultraflexible applications, or those demanding mechanical robustness during roll-to-roll fabrication or use in the outdoor environment.

Lattice computing models such as the morphological neural networks and fuzzy neurocomputing models are becoming increasingly important with the advent of granular computing. In particular, the morphological perceptron with competitive learning (MP/CL), introduced by Sussner and Esmi, exhibited satisfactory classification results in some well known classification problems. On the downside, the MP/CL is subject to overfitting in which the network learns singular characteristics from the training data. In this paper, we propose a learning strategy based on a certain genetic algorithm to circumvent the overfitting problem of MP/CL. Computational experiments revealed that the novel model can achieve similar classification results but using a smaller number of hidden neurons. Keywords-neural networks; morphological perceptron; classification problem; genetic algorithm.

As web technologies and social networks become part of the general public's life, the problem of automatically detecting pornography is into every parent's mind - nobody feels completely safe when their children go online. In this paper, we focus on video-pornography classification, a hard problem in which traditional methods often employ still-image techniques - labeling frames individually prior to a global decision. Frame-based approaches, however, ignore significant cogent information brought by motion. Here, we introduce a space-temporal interest point detector and descriptor called Temporal Robust Features (TRoF). TRoF was custom-tailored for efficient (low processing time and memory footprint) and effective (high classification accuracy and low false negative rate) motion description, particularly suited to the task at hand. We aggregate local information extracted by TRoF into a mid-level representation using Fisher Vectors, the state-of-the-art model of Bags of Visual Words (BoVW). We evaluate our original strategy, contrasting it both to commercial pornography detection solutions, and to BoVW solutions based upon other space-temporal features from the scientific literature. The performance is assessed using the Pornography-2k dataset, a new challenging pornographic benchmark, comprising 2000 web videos and 140h of video footage. The dataset is also a contribution of this work and is very assorted, including both professional and amateur content, and it depicts several genres of pornography, from cartoon to live action, with diverse behavior and ethnicity. The best approach, based on a dense application of TRoF, yields a classification error reduction of almost 79% when compared to the best commercial classifier. A sparse description relying on TRoF detector is also noteworthy, for yielding a classification error reduction of over 69%, with 19× less memory footprint than the dense solution, and yet can also be implemented to meet real-time requirements.

We recently employed concepts of mathematical morphology to introduce fuzzy morphological associative memories (FMAMs), a broad class of fuzzy associative memories (FAMs). We observed that many well-known FAM models can be classified as belonging to the class of FMAMs. Moreover, we developed a general learning strategy for FMAMs using the concept of adjunction of mathematical morphology. In this paper, we describe the properties of FMAMs with adjunction-based learning. In particular, we characterize the recall phase of these models. Furthermore, we prove several theorems concerning the storage capacity, noise tolerance, fixed points, and convergence of auto-associative FMAMs. These theorems are corroborated by experimental results concerning the reconstruction of noisy images. Finally, we successfully employ FMAMs with adjunction-based learning in order to implement fuzzy rule-based systems in an application to a time-series prediction problem in industry.

Autoassociative fuzzy implicative memories (AFIMs) are models that exhibit optimal absolute storage capacity and an excellent tolerance with respect to incomplete or eroded patterns. As a consequence, they can be effectively used for the reconstruction of gray-scale images. In practice, however, applications of AFIMs are confined to images of small size due to computational limitations. In order to circumvent this computational overhead and, motivated by the sparsity of biological neural networks, this paper introduces the class of sparsely connected AFIMs (SCAFIMs). Such as the original AFIMs, SCAFIMs exhibit optimal absolute storage capacity and tolerance with respect to incomplete or eroded patterns. By means of computational experiments, we investigate the performance of SCAFIMs with different network topologies and compare the novel models with other techniques for the reconstruction of gray-scale images.

Fuzzy associative memories (FAMs) can be used as a powerful tool for implementing fuzzy rule-based systems. The insight that FAMs are closely related to mathematical morphology (MM) has recently led to the development of new fuzzy morphological associative memories (FMAMs), in particular implicative fuzzy associative memories (IFAMs). As the name FMAM indicates, these models belong to the class of fuzzy morphological neural networks (FMNNs). Thus, each node of an FMAM performs an elementary operation of fuzzy MM. Clarifying several misconceptions about FMAMs that have recently appeared in the literature, we provide a general framework for FMAMs within the class of FMNN. We show that many well-known FAM models fit within this framework and can therefore be classified as FMAMs. Moreover, we employ certain concepts of duality that are defined in the general theory of MM in order to derive a large class of strategies for learning and recall in FMAMs.

Diabetic retinopathy (DR) is the leading cause of blindness in adults, but can be managed if detected early. Automated DR screening helps by indicating which patients should be referred to the doctor. However, current techniques of automated screening still depend too much on the detection of individual lesions. In this study, we bypass lesion detection, and directly train a classifier for DR referral. Additional novelties are the use of state-of-the-art mid-level features for the retinal images: BossaNova and Fisher Vector. Those features extend the classical Bags of Visual Words and greatly improve the accuracy of complex classification tasks. The proposed technique for direct referral is promising, achieving an area under the curve of 96.4%, thus, reducing the classification error by almost 40% over the current state of the art, held by lesion-based techniques.

Autossociative memories (AMs) are models inspired by the human brain ability to store and recall information. They should be able to retrieve a stored information upon presentation of a partial or corrupted item. An AM that projects the input onto a linear subspace is called subspace projection autoassociative memory (SPAM). The recall phase of a SPAM model is equivalent to a multi-linear regression problem. In particular, the optimal linear autoassociative memory (OLAM) corresponds to the SPAM model obtained by considering traditional least squares regression in the recall phase. In this paper, we present a novel class of SPAM models obtained by considering linear support vector regression (SVR). Precisely, we introduce three SPAM models based on primal, dual, and bi-level formulations of the linear e-support vector regression. A simple example is used throughout the paper to illustrate the noise tolerance of the proposed memory models.

The recurrent exponential fuzzy associative memory (RE-FAM) can be viewed as a recurrent neural network that employs a fuzzy similarity measure in its hidden layer. This paper introduces the generalized recurrent exponential fuzzy associative memory (GRE-FAM). In contrast to the RE-FAM, the GREFAM is equipped with a second hidden layer that is geared to avoiding crosstalk. Apart from theoretical results, this paper includes some computational experiments concerning the reconstruction of corrupted gray-scale images.

Associative neural memories are models of biological phenomena that allow for the storage of pattern associations and the retrieval of the desired output pattern upon presentation of a possibly noisy or incomplete version of an input pattern. In this paper, we introduce implicative fuzzy associative memories (IFAMs), a class of associative neural memories based on fuzzy set theory. An IFAM consists of a network of completely interconnected Pedrycz logic neurons with threshold whose connection weights are determined by the minimum of implications of presynaptic and postsynaptic activations. We present a series of results for autoassociative models including one pass convergence, unlimited storage capacity and tolerance with respect to eroded patterns. Finally, we present some results on fixed points and discuss the relationship between implicative fuzzy associative memories and morphological associative memories

In this paper, we generalize the bipolar recurrent correlation neural networks (RCNNs) of Chiueh and Goodman for patterns whose components are in the complex unit circle. The novel networks, referred to as complex-valued RCNNs (CV-RCNNs), are characterized by a possible nonlinear function, which is applied on the real part of the scalar product of the current state and the original patterns. We show that the CV-RCNNs always converge to a stationary state. Thus, they have potential application as associative memories. In this context, we provide sufficient conditions for the retrieval of a memorized vector. Furthermore, computational experiments concerning the reconstruction of corrupted grayscale images reveal that certain CV-RCNNs exhibit an excellent noise tolerance.

The biomedical community has shown a continued interest in automated detection of Diabetic Retinopathy (DR), with new imaging techniques, evolving diagnostic criteria, and advancing computing methods. Existing state of the art for detecting DR-related lesions tends to emphasize different, specific approaches for each type of lesion. However, recent research has aimed at general frameworks adaptable for large classes of lesions. In this paper, we follow this latter trend by exploring a very flexible framework, based upon two-tiered feature extraction (low-level and mid-level) from images and Support Vector Machines. The main contribution of this work is the evaluation of BossaNova, a recent and powerful mid-level image characterization technique, which we contrast with previous art based upon classical Bag of Visual Words (BoVW). The new technique using BossaNova achieves a detection performance (measured by area under the curve - AUC) of 96.4% for hard exudates, and 93.5% for red lesions using a cross-dataset training/testing protocol.

According to the American Cancer Society, one person dies of melanoma every 57 minutes, although it is the most curable type of cancer if detected early. Thus, computeraided diagnosis for melanoma screening has been a topic of active research. Much of the existing art is based on the Bag-of-Visual-Words (BoVW) model, combined with color and texture descriptors. However, recent advances in the BoVW model, as well as the evaluation of the importance of the many different factors affecting the BoVW model were yet to be explored, thus motivating our work. We show that a new approach for melanoma screening, based upon the state-of-the-art BossaNova descriptors, shows very promising results for screening, reaching an AUC of up to 93.7%. An important contribution of this work is an evaluation of the factors that affect the performance of the two-layered BoVW model. Our results show that the low-level layer has a major impact on the accuracy of the model, but that the codebook size on the mid-level layer is also important. Those results may guide future works on melanoma screening.