Search

Business intelligence and AI services often involve the collection of copious amounts of multidimensional personal data. Since these data usually contain sensitive information of individuals, the direct collection can lead to privacy violations. Local differential privacy (LDP) is currently considered a state-ofthe-art solution for privacy-preserving data collection. However, existing LDP algorithms are not applicable to high-dimensional data; not only because of the increase in computation and communication cost, but also poor data utility.In this paper, we aim at addressing thecurse-of-dimensionalityproblem in LDP-based high-dimensional data collection. Based on the idea of machine learning and data synthesis, we propose DP-Fed-Wae, an efficient privacy-preserving framework for collecting high-dimensional categorical data. With the combination of a generative autoencoder, federated learning, and differential privacy, our framework is capable of privately learning the statistical distributions of local data and generating high utility synthetic data on the server side without revealing users’ private information. We have evaluated the framework in terms of data utility and privacy protection on a number of real-world datasets containing 68–124 classification attributes. We show that our framework outperforms the LDP-based baseline algorithms in capturing joint distributions and correlations of attributes and generating high-utility synthetic data. With a local privacy guarantee ∈ = 8, the machine learning models trained with the synthetic data generated by the baseline algorithm cause an accuracy loss of 10% ~ 30%, whereas the accuracy loss is significantly reduced to less than 3% and at best even less than 1% with our framework. Extensive experimental results demonstrate the capability and efficiency of our framework in synthesizing high-dimensional data while striking a satisfactory utility-privacy balance.

International audience; Over the last years, Remote Sensing Images (RSI) analysis have started resorting to using deep neural networks to solve most of the commonly faced problems, such as detection, land cover classification or segmentation. As far as critical decision making can be based upon the results of RSI analysis, it is important to clearly identify and understand potential security threats occurring in those machine learning algorithms. Notably, it has recently been found that neural networks are particularly sensitive to carefully designed attacks, generally crafted given the full knowledge of the considered deep network. In this paper, we consider the more realistic but challenging case where one wants to generate such attacks in the case of a black-box neural network. In this case, only the prediction score of the network is accessible, given a specific input. Examples that lure away the network's prediction, while being perceptually similar to real images, are called natural or unrestricted adversarial examples. We present an original method to generate such examples, based on a variant of the Wasserstein Generative Adversarial Network. We demonstrate its effectiveness on natural adversarial hyper-spectral image generation and image modification for fooling a state-of-the-art detector. Among others, we also conduct a perceptual evaluation with human annotators to better assess the effectiveness of the proposed method.

This paper presents a teaching experiment in which 3D digital computational models are explored as the representational base to integrate formal, structural, and environmental performance criteria in design. By describing the academic experience, the paper reflects on its methodologies and results, as well as on the relation between human and computer factors in the design process. This assessment is important to make the students aware of the increasingly intelligent design systems offered by digital technologies to support architectural design, as well as of their relationship with precedent digital and analog representational mediums.

We study two state of the art deep generative networks, the Introspective Variational Autoencoder and the Style-Based Generative Adversarial Network, for the generation of new diffusion-weighted magnetic resonance images. We show that high quality, diverse and realistic-looking images, as evaluated by external neuroradiologists blinded to the whole study, can be synthesized using these deep generative models. We evaluate diverse metrics with respect to quality and diversity of the generated synthetic brain images. These findings show that generative models could qualify as a method for data augmentation in the medical field, where access to large image database is in many aspects restricted., Magnetic Resonance Imaging, 81, ISSN:0730-725X, ISSN:1873-5894

Deep generative models have achieved great success in areas such as image, speech, and natural language processing in the past few years. Thanks to the advances in graph-based deep learning, and in particular graph representation learning, deep graph generation methods have recently emerged with new applications ranging from discovering novel molecular structures to modeling social networks. This paper conducts a comprehensive survey on deep learning-based graph generation approaches and classifies them into five broad categories, namely, autoregressive, autoencoder-based, reinforcement learning-based, adversarial, and flow-based graph generators, providing the readers a detailed description of the methods in each class. We also present publicly available source codes, commonly used datasets, and the most widely utilized evaluation metrics. Finally, we review current trends and suggest future research directions based on the existing challenges.

Variational autoencoders (VAEs) are deep latent space generative models that have been immensely successful in multiple exciting applications in biomedical informatics such as molecular design, protein design, medical image classification and segmentation, integrated multi-omics data analyses, and large-scale biological sequence analyses, among others. The fundamental idea in VAEs is to learn the distribution of data in such a way that new meaningful data with more intra-class variations can be generated from the encoded distribution. The ability of VAEs to synthesize new data with more representation variance at state-of-art levels provides hope that the chronic scarcity of labeled data in the biomedical field can be resolved. Furthermore, VAEs have made nonlinear latent variable models tractable for modeling complex distributions. This has allowed for efficient extraction of relevant biomedical information from learned features for biological data sets, referred to as unsupervised feature representation learning. In this article, we review the various recent advancements in the development and application of VAEs for biomedical informatics. We discuss challenges and future opportunities for biomedical research with respect to VAEs.

Video portraits are common in a variety of applications, such as videoconferencing, news broadcasting, and virtual education and training. We present a novel method to synthesize photorealistic video portraits for an input portrait video, automatically driven by a person’s voice. The main challenge in this task is the hallucination of plausible, photorealistic facial expressions from input speech audio. To address this challenge, we employ a parametric 3D face model represented by geometry, facial expression, illumination, etc., and learn a mapping from audio features to model parameters. The input source audio is first represented as a high-dimensional feature, which is used to predict facial expression parameters of the 3D face model. We then replace the expression parameters computed from the original target video with the predicted one, and rerender the reenacted face. Finally, we generate a photorealistic video portrait from the reenacted synthetic face sequence via a neural face renderer. One appealing feature of our approach is the generalization capability for various input speech audio, including synthetic speech audio from text-to-speech software. Extensive experimental results show that our approach outperforms previous general-purpose audio-driven video portrait methods. This includes a user study demonstrating that our results are rated as more realistic than previous methods.

We present a novel method for publishing differentially private synthetic attributed graphs. Our method allows, for the first time, to publish synthetic graphs simultaneously preserving structural properties, user attributes and the community structure of the original graph. Our proposal relies on CAGM, a new community-preserving generative model for attributed graphs. We equip CAGM with efficient methods for attributed graph sampling and parameter estimation. For the latter, we introduce differentially private computation methods, which allow us to release communitypreserving synthetic attributed social graphs with a strong formal privacy guarantee. Through comprehensive experiments, we show that our new model outperforms its most relevant counterparts in synthesising differentially private attributed social graphs that preserve the community structure of the original graph, as well as degree sequences and clustering coefficients.

Objective Due to a complex set of processes involved with the recording of health information in the Electronic Health Records (EHRs), the truthfulness of EHR diagnosis records is questionable. We present a computational approach to estimate the probability that a single diagnosis record in the EHR reflects the true disease. Materials and Methods Using EHR data on 18 diseases from the Mass General Brigham (MGB) Biobank, we develop generative classifiers on a small set of disease-agnostic features from EHRs that aim to represent Patients, pRoviders, and their Interactions within the healthcare SysteM (PRISM features). Results We demonstrate that PRISM features and the generative PRISM classifiers are potent for estimating disease probabilities and exhibit generalizable and transferable distributional characteristics across diseases and patient populations. The joint probabilities we learn about diseases through the PRISM features via PRISM generative models are transferable and generalizable to multiple diseases. Discussion The Generative Transfer Learning (GTL) approach with PRISM classifiers enables the scalable validation of computable phenotypes in EHRs without the need for domain-specific knowledge about specific disease processes. Conclusion Probabilities computed from the generative PRISM classifier can enhance and accelerate applied Machine Learning research and discoveries with EHR data.

Deep learning based disease detection and segmentation algorithms promise to improve many clinical processes. However, such algorithms require vast amounts of annotated training data, which are typically not available in the medical context due to data privacy, legal obstructions, and non-uniform data acquisition protocols. Synthetic databases with annotated pathologies could provide the required amounts of training data. We demonstrate with the example of ischemic stroke that an improvement in lesion segmentation is feasible using deep learning based augmentation. To this end, we train different image-to-image translation models to synthesize magnetic resonance images of brain volumes with and without stroke lesions from semantic segmentation maps. In addition, we train a generative adversarial network to generate synthetic lesion masks. Subsequently, we combine these two components to build a large database of synthetic stroke images. The performance of the various models is evaluated using a U-Net which is trained to segment stroke lesions on a clinical test set. We report a Dice score of 72.8% [70.8±1.0%] for the model with the best performance, which outperforms the model trained on the clinical images alone 67.3% [63.2±1.9%], and is close to the human inter-reader Dice score of 76.9%. Moreover, we show that for a small database of only 10 or 50 clinical cases, synthetic data augmentation yields significant improvement compared to a setting where no synthetic data is used. To the best of our knowledge, this presents the first comparative analysis of synthetic data augmentation based on image-to-image translation, and first application to ischemic stroke., Biomedical Signal Processing and Control, 72 (A), ISSN:1746-8094

International audience; Deep learning, together with the availability of large amounts of data, has transformed the way we process Earth observation (EO) tasks, such as land cover mapping or image registration. Yet, today, new models are needed to push further the revolution and enable new possibilities. This work focuses on a recent framework for generative modeling and explores its applicability to the EO images. The framework learns an energy-based model (EBM) to estimate the underlying joint distribution of the data and the categories, obtaining a neural network that is able to classify and synthesize images. On these two tasks, we show that EBMs reach comparable or better performances than convolutional networks on various public EO datasets and that they are naturally adapted to semisupervised settings, with very few labeled data. Moreover, models of this kind allow us to address high-potential applications, such as out-of-distribution analysis and land cover mapping with confidence estimation.

An artificial intelligence approach based on deep generative neural networks for spectral imaging processing is proposed. The key idea is to treat different spectral image processing operations such as segmentation, regression, and classification as an image-to-image translation task. For the image-to-image translation, the conditional generative adversarial networks were used. As a baseline comparison, the traditional chemometric approach based on pixels wise modelling was demonstrated. The analysis is presented with two real data sets related to fruit property prediction and kernel and shell classification of walnuts. The presented artificial intelligence approach for spectral image processing can provide benefits for any field of science where spectral imaging and processing is widely performed.

Deep learning applications on computer vision involve the use of large-volume and representative data to obtain state-of-the-art results due to the massive number of parameters to optimise in deep models. However, data are limited with asymmetric distributions in industrial applications due to rare cases, legal restrictions, and high image-acquisition costs. Data augmentation based on deep learning generative adversarial networks, such as StyleGAN, has arisen as a way to create training data with symmetric distributions that may improve the generalisation capability of built models. StyleGAN generates highly realistic images in a variety of domains as a data augmentation strategy but requires a large amount of data to build image generators. Thus, transfer learning in conjunction with generative models are used to build models with small datasets. However, there are no reports on the impact of pre-trained generative models, using transfer learning. In this paper, we evaluate a StyleGAN generative model with transfer learning on different application domains—training with paintings, portraits, Pokémon, bedrooms, and cats—to generate target images with different levels of content variability: bean seeds (low variability), faces of subjects between 5 and 19 years old (medium variability), and charcoal (high variability). We used the first version of StyleGAN due to the large number of publicly available pre-trained models. The Fréchet Inception Distance was used for evaluating the quality of synthetic images. We found that StyleGAN with transfer learning produced good quality images, being an alternative for generating realistic synthetic images in the evaluated domains.

Recently, we have released the de novo design platform REINVENT in version 2.0. This improved and extended iteration supports far more features and scoring function components, which allows bespoke and tailor-made protocols to maximize impact in small molecule drug discovery projects. A major obstacle of generative models is producing active compounds, in which predictive (QSAR) models have been applied to enrich target activity. However, QSAR models are inherently limited by their applicability domains. To overcome these limitations, we introduce a structure-based scoring component for REINVENT. DockStream is a flexible, stand-alone molecular docking wrapper that provides access to a collection of ligand embedders and docking backends. Using the benchmarking and analysis workflow provided in DockStream, execution and subsequent analysis of a variety of docking configurations can be automated. Docking algorithms vary greatly in performance depending on the target and the benchmarking and analysis workflow provides a streamlined solution to identifying productive docking configurations. We show that an informative docking configuration can inform the REINVENT agent to optimize towards improving docking scores using public data. With docking activated, REINVENT is able to retain key interactions in the binding site, discard molecules which do not fit the binding cavity, harness unused (sub-)pockets, and improve overall performance in the scaffold-hopping scenario. The code is freely available at https://github.com/MolecularAI/DockStream.

Deep learning, one of the fastest-growing branches of artificial intelligence, has become one of the most relevant research and development areas of the last years, especially since 2012, when a neural network surpassed the most advanced image classification techniques of the time. This spectacular development has not been alien to the world of the arts, as recent advances in generative networks have made possible the artificial creation of high-quality content such as images, movies or music. We believe that these novel generative models propose a great challenge to our current understanding of computational creativity. If a robot can now create music that an expert cannot distinguish from music composed by a human, or create novel musical entities that were not known at training time, or exhibit conceptual leaps, does it mean that the machine is then creative? We believe that the emergence of these generative models clearly signals that much more research needs to be done in this area. We would like to contribute to this debate with two case studies of our own: TimbreNet, a variational auto-encoder network trained to generate audio-based musical chords, and StyleGAN Pianorolls, a generative adversarial network capable of creating short musical excerpts, despite the fact that it was trained with images and not musical data. We discuss and assess these generative models in terms of their creativity and we show that they are in practice capable of learning musical concepts that are not obvious based on the training data, and we hypothesize that these deep models, based on our current understanding of creativity in robots and machines, can be considered, in fact, creative.

Variational Auto-Encoders (VAEs) are deep latent space generative models which have been immensely successful in many applications such as image generation, image captioning, protein design, mutation prediction, and language models among others. The fundamental idea in VAEs is to learn the distribution of data in such a way that new meaningful data can be generated from the encoded distribution. This concept has led to tremendous research and variations in the design of VAEs in the last few years creating a field of its own, referred to as unsupervised representation learning. This paper provides a much-needed comprehensive evaluation of the variations of the VAEs based on their end goals and resulting architectures. It further provides intuition as well as mathematical formulation and quantitative results of each popular variation, presents a concise comparison of these variations, and concludes with challenges and future opportunities for research in VAEs.

This paper proposes a method for calculating a health indicator (HI) for low-speed axial rolling element bearing (REB) health assessment by utilizing the latent representation obtained by variational inference using Variational Autoencoders (VAEs), trained on each speed reference in the dataset. Further, versatility is added by conditioning on the speed, extending the VAE to a conditional VAE (CVAE), thereby incorporating all speeds in a single model. Within the framework, the coefficients of autoregressive (AR) models are used as features. The dimensionality reduction inherent in the proposed method lowers the need of expert knowledge to design good condition indicators. Moreover, the suggested methodology allows for setting the probability of false alarms when encoding new data points to the latent variable space using the trained model. The effectiveness of the proposed method is validated based on two different datasets: from a workshop test of an offshore drilling machine and from an in-house test rig for axial bearings. In both datasets, the HI is exceeding the warning and alarm levels with a probability of false alarm (PFA) of 10 -6 , and the method is most effective at lower shaft speeds. publishedVersion

In this study, we explore building a two-stage framework for enabling users to directly manipulate high-level attributes of a natural scene. The key to our approach is a deep generative network which can hallucinate images of a scene as if they were taken at a different season (e.g. during winter), weather condition (e.g. in a cloudy day) or time of the day (e.g. at sunset). Once the scene is hallucinated with the given attributes, the corresponding look is then transferred to the input image while preserving the semantic details intact, giving a photo-realistic manipulation result. As the proposed framework hallucinates what the scene will look like, it does not require any reference style image as commonly utilized in most of the appearance or style transfer approaches. Moreover, it allows to simultaneously manipulate a given scene according to a diverse set of transient attributes within a single model, eliminating the need of training multiple networks per each translation task. Our comprehensive set of qualitative and quantitative results demonstrate the effectiveness of our approach against the competing methods., Accepted for publication in ACM Transactions on Graphics

Deep generative models have significantly advanced image generation, enabling generation of visually pleasing images with realistic texture. Apart from the texture, it is the shape geometry of objects that strongly dictates their appearance. However, currently available generative models do not incorporate geometric information into the image generation process. This often yields visual objects of degenerated quality. In this work, we propose a regularized Geometry-Aware Generative Adversarial Network (GAGAN) which disentangles appearance and shape in the latent space. This regularized GAGAN enables the generation of images with both realistic texture and shape. Specifically, we condition the generator on a statistical shape prior. The prior is enforced through mapping the generated images onto a canonical coordinate frame using a differentiable geometric transformation. In addition to incorporating geometric information, this constrains the search space and increases the model’s robustness. We show that our approach is versatile, able to generalise across domains (faces, sketches, hands and cats) and sample sizes (from as little as $$\sim \, 200{-}30{,}000$$ to more than 200, 000). We demonstrate superior performance through extensive quantitative and qualitative experiments in a variety of tasks and settings. Finally, we leverage our model to automatically and accurately detect errors or drifting in facial landmarks detection and tracking in-the-wild.

International audience; Deep learning has changed unbelievably the processing of Earth Observation tasks such as land cover mapping or image registration. Yet, today new models are needed to push further the revolution and enable new possibilities. We propose a new framework for generative modelling of Earth Observation images. It learns an energy-based model to estimate the underlying distribution of the data while jointly training a deep neural network for classification. On the varied image types of the EuroSAT benchmark, we show this model obtains classification results on par with state-of-the-art and moreover allows us to tackle a wide range of high-potential applications: image synthesis, out-of-distribution testing for domain adaptation, and image completion or denoising.

Mobile phone data generated in mobile communication networks has the potential to improve current travel demand models and in general, how we plan for better urban transportation systems. However, due to its high-dimensionality, even if anonymised there still exists the possibility to re-identify the users behind the mobile phone traces. This risk makes its usage outside the telecommunication network incompatible with recent data privacy regulations, hampering its adoption in transportation-related applications. To address this issue, we propose a framework designed only with user-aggregated mobile phone data to synthesise realistic daily individual mobility — Digital Twin Travellers. We explore different strategies built around modified Markov models and an adaption of the Rejection Sampling algorithm to recreate realistic daily schedules and locations. We also define a one-day mobility population score to measure the similarity between the population of generated agents and the real mobile phone user population. Ultimately, we show how with a series of histograms provided by the telecommunication service provider (TSP) it is possible and plausible to disaggregate them into new synthetic and useful individual-level information, building in this way a big data travel demand framework that is designed in accordance with current data privacy regulations., Transportation Research Part C: Emerging Technologies, 128, ISSN:0968-090X

Computing and software for big science 5(1), 13 (2021). doi:10.1007/s41781-021-00056-0, Accurate simulation of physical processes is crucial for the success of modern particle physics. However, simulating the development and interaction of particle showers with calorimeter detectors is a time consuming process and drives the computing needs of large experiments at the LHC and future colliders. Recently, generative machine learning models based on deep neural networks have shown promise in speeding up this task by several orders of magnitude. We investigate the use of a new architecture—the Bounded Information Bottleneck Autoencoder—for modelling electromagnetic showers in the central region of the Silicon-Tungsten calorimeter of the proposed International Large Detector. Combined with a novel second post-processing network, this approach achieves an accurate simulation of differential distributions including for the first time the shape of the minimum-ionizing-particle peak compared to a full Geant4 simulation for a high-granularity calorimeter with 27k simulated channels. The results are validated by comparing to established architectures. Our results further strengthen the case of using generative networks for fast simulation and demonstrate that physically relevant differential distributions can be described with high accuracy., Published by Springer International Publishing, Cham, Switzerland

In recent years, machine learning algorithms have been applied widely in various fields such as health, transportation, and the autonomous car. With the rapid developments of deep learning techniques, it is critical to take the security concern into account for the application of the algorithms. While machine learning offers significant advantages in terms of the application of algorithms, the issue of security is ignored. Since it has many applications in the real world, security is a vital part of the algorithms. In this paper, we have proposed a mitigation method for adversarial attacks against machine learning models with an autoencoder model that is one of the generative ones. The main idea behind adversarial attacks against machine learning models is to produce erroneous results by manipulating trained models. We have also presented the performance of autoencoder models to various attack methods from deep neural networks to traditional algorithms by using different methods such as non-targeted and targeted attacks to multi-class logistic regression, a fast gradient sign method, a targeted fast gradient sign method and a basic iterative method attack to neural networks for the MNIST dataset., Comment: 33 pages

This paper reviews the literature concerning technology used for creating and delivering accessible museum and cultural heritage sites experiences. It highlights the importance of the delivery suited for everyone from different areas of expertise, namely interaction design, pedagogical and participatory design, and it presents how recent and future artificial intelligence (AI) developments can be used for this aim, i.e.,improving and widening online and in situ accessibility. From the literature review analysis, we articulate a conceptual framework that incorporates key elements that constitute museum and cultural heritage online experiences and how these elements are related to each other. Concrete opportunities for future directions empirical research for accessibility of cultural heritage contents are suggested and further discussed.

Modelling human mobility is crucial in several areas, from urban planning to epidemic modelling, traffic forecasting, and what-if analysis. Existing generative models focus mainly on reproducing the spatial and temporal dimensions of human mobility, while the social aspect, though it influences human movements significantly, is often neglected. Those models that capture some social perspectives of human mobility utilize trivial and unrealistic spatial and temporal mechanisms. In this paper, we propose the Spatial, Temporal and Social Exploration and Preferential Return model (STS-EPR), which embeds mechanisms to capture the spatial, temporal, and social aspects together. We compare the trajectories produced by STS-EPR with respect to real-world trajectories and synthetic trajectories generated by two state-of-the-art generative models on a set of standard mobility measures. Our experiments conducted on an open dataset show that STS-EPR, overall, outperforms existing spatial-temporal or social models demonstrating the importance of modelling adequately the sociality to capture precisely all the other dimensions of human mobility. We further investigate the impact of the tile shape of the spatial tessellation on the performance of our model. STS-EPR, which is open-source and tested on open data, represents a step towards the design of a mechanistic data-driven model that captures all the aspects of human mobility comprehensively.

Radiomics is an active area of research in medical image analysis, however poor reproducibility of radiomics has hampered its application in clinical practice. This issue is especially prominent when radiomic features are calculated from noisy images, such as low dose computed tomography (CT) scans. In this article, we investigate the possibility of improving the reproducibility of radiomic features calculated on noisy CTs by using generative models for denoising. Our work concerns two types of generative models—encoder–decoder network (EDN) and conditional generative adversarial network (CGAN). We then compared their performance against a more traditional ‘non-local means’ denoising algorithm. We added noise to sinograms of full dose CTs to mimic low dose CTs with two levels of noise: low-noise CT and high-noise CT. Models were trained on high-noise CTs and used to denoise low-noise CTs without re-training. We tested the performance of our model in real data, using a dataset of same-day repeated low dose CTs in order to assess the reproducibility of radiomic features in denoised images. EDN and the CGAN achieved similar improvements on the concordance correlation coefficients (CCC) of radiomic features for low-noise images from 0.87 [95%CI, (0.833, 0.901)] to 0.92 [95%CI, (0.909, 0.935)] and for high-noise images from 0.68 [95%CI, (0.617, 0.745)] to 0.92 [95%CI, (0.909, 0.936)], respectively. The EDN and the CGAN improved the test-retest reliability of radiomic features (mean CCC increased from 0.89 [95%CI, (0.881, 0.914)] to 0.94 [95%CI, (0.927, 0.951)]) based on real low dose CTs. These results show that denoising using EDN and CGANs could be used to improve the reproducibility of radiomic features calculated from noisy CTs. Moreover, images at different noise levels can be denoised to improve the reproducibility using the above models without need for re-training, provided the noise intensity is not excessively greater that of the high-noise CTs. To the authors’ knowledge, this is the first effort to improve the reproducibility of radiomic features calculated on low dose CT scans by applying generative models.

Purpose Given the high level of expertise required for navigation and interpretation of ultrasound images, computational simulations can facilitate the training of such skills in virtual reality. With ray-tracing based simulations, realistic ultrasound images can be generated. However, due to computational constraints for interactivity, image quality typically needs to be compromised. Methods We propose herein to bypass any rendering and simulation process at interactive time, by conducting such simulations during a non-time-critical offline stage and then learning image translation from cross-sectional model slices to such simulated frames. We use a generative adversarial framework with a dedicated generator architecture and input feeding scheme, which both substantially improve image quality without increase in network parameters. Integral attenuation maps derived from cross-sectional model slices, texture-friendly strided convolutions, providing stochastic noise and input maps to intermediate layers in order to preserve locality are all shown herein to greatly facilitate such translation task. Results Given several quality metrics, the proposed method with only tissue maps as input is shown to provide comparable or superior results to a state-of-the-art that uses additional images of low-quality ultrasound renderings. An extensive ablation study shows the need and benefits from the individual contributions utilized in this work, based on qualitative examples and quantitative ultrasound similarity metrics. To that end, a local histogram statistics based error metric is proposed and demonstrated for visualization of local dissimilarities between ultrasound images. Conclusion A deep-learning based direct transformation from interactive tissue slices to likeness of high quality renderings allow to obviate any complex rendering process in real-time, which could enable extremely realistic ultrasound simulations on consumer-hardware by moving the time-intensive processes to a one-time, offline, preprocessing data preparation stage that can be performed on dedicated high-end hardware., International Journal of Computer Assisted Radiology and Surgery, 16 (5), ISSN:1861-6410, ISSN:1861-6429

It is more pressing than ever to reduce the time and costs for the development of lead compounds in the pharmaceutical industry. The co-occurrence of advances in high-throughput screening and the rise of deep learning (DL) have enabled the development of large-scale multimodal predictive models for virtual drug screening. Recently, deep generative models have emerged as a powerful tool to explore the chemical space and raise hopes to expedite the drug discovery process. Following this progress in chemocentric approaches for generative chemistry, the next challenge is to build multimodal conditional generative models that leverage disparate knowledge sources when mapping biochemical properties to target structures. Here, we call the community to bridge drug discovery more closely with systems biology when designing deep generative models. Complementing the plethora of reviews on the role of DL in chemoinformatics, we specifically focus on the interface of predictive and generative modelling for drug discovery. Through a systematic publication keyword search on PubMed and a selection of preprint servers (arXiv, biorXiv, chemRxiv, and medRxiv), we quantify trends in the field and find that molecular graphs and VAEs have become the most widely adopted molecular representations and architectures in generative models, respectively. We discuss progress on DL for toxicity, drug-target affinity, and drug sensitivity prediction and specifically focus on conditional molecular generative models that encompass multimodal prediction models. Moreover, we outline future prospects in the field and identify challenges such as the integration of deep learning systems into experimental workflows in a closed-loop manner or the adoption of federated machine learning techniques to overcome data sharing barriers. Other challenges include, but are not limited to interpretability in generative models, more sophisticated metrics for the evaluation of molecular generative models, and, following up on that, community -accepted benchmarks for both multimodal drug property prediction and property-driven molecular design., Current Medicinal Chemistry, 28 (38), ISSN:0929-8673

Here are presented technical notes and tips on developing graph generative models for molecular design. Although this work stems from the development of GraphINVENT, a Python platform for iterative molecular generation using graph neural networks, this work is relevant to researchers studying other architectures for graph‐based molecular design. In this work, technical details that could be of interest to researchers developing their own molecular generative models are discussed, including an overview of previous work in graph‐based molecular design and strategies for designing new models. Advice on development and debugging tools which are helpful during code development is also provided. Finally, methods that were tested but which ultimately did not lead to promising results in the development of GraphINVENT are described here in the hope that this will help other researchers avoid pitfalls in development and instead focus their efforts on more promising strategies for graph‐based molecular generation.

Variational autoencoders (VAEs), as well as other generative models, have been shown to be efficient and accurate for capturing the latent structure of vast amounts of complex high-dimensional data. However, existing VAEs can still not directly handle data that are heterogenous (mixed continuous and discrete) or incomplete (with missing data at random), which is indeed common in real-world applications. In this paper, we propose a general framework to design VAEs suitable for fitting incomplete heterogenous data. The proposed HI-VAE includes likelihood models for real-valued, positive real valued, interval, categorical, ordinal and count data, and allows accurate estimation (and potentially imputation) of missing data. Furthermore, HI-VAE presents competitive predictive performance in supervised tasks, outperforming supervised models when trained on incomplete data. The authors wish to thank Christopher K. I. Williams, for fruitful discussions and helpful comments to the manuscript. Alfredo Nazabal would like to acknowledge the funding provided by the UK Government’s Defence & Security Programme in support of the Alan Turing Institute, EPSRC Grant EP/N510129/1. The work of Pablo M. Olmos is sup-ported by Spanish government MCI under grant RTI2018-099655-B-100, by Comunidad de Madrid under grants IND2017/TIC-7618, IND2018/TIC-9649, and Y2018/TCS-4705, by BBVA Foundation under the Deep-DARWiNproject, and by the European Union (FEDER and the European Research Council (ERC) through the European Unions Horizon 2020 research and innovation program under Grant 714161). Zoubin Ghahramani acknowledges support from the Alan Turing Institute (EPSRC Grant EP/N510129/1) and EPSRC Grant EP/N014162/1, and donations from Google and Microsoft Research. We also gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan X Pascal GPU used for this research.

Recent Advances in Artificial Intelligence (AI), particularly in the field of compute vision, have been driven by the availability of large public datasets. However, as AI begins to move into embedded devices there will be a growing need for tools to acquire and re-acquire datasets from specific sensing systems to train new device models. In this paper, a roadmap in introduced for a data-acquisition framework that can build the large synthetic datasets required to train AI systems from small seed datasets. A key element to justify such a framework is the validation of the generated dataset and example results are shown from preliminary work on biometric (facial) datasets. non-peer-reviewed

Deep neural networks have recently been used to edit images with great success, in particular for faces. However, they are often limited to only being able to work at a restricted range of resolutions. Many methods are so flexible that face edits can often result in an unwanted loss of identity. This work proposes to learn how to perform semantic image edits through the application of smooth warp fields. Previous approaches that attempted to use warping for semantic edits required paired data, i.e. example images of the same subject with different semantic attributes. In contrast, we employ recent advances in Generative Adversarial Networks that allow our model to be trained with unpaired data. We demonstrate face editing at very high resolutions (4k images) with a single forward pass of a deep network at a lower resolution. We also show that our edits are substantially better at preserving the subject's identity. The robustness of our approach is demonstrated by showing plausible image editing results on the Cub200 birds dataset. To our knowledge this has not been previously accomplished, due the challenging nature of the dataset., CVPR 2020

Purpose Deep learning can be used for improving the performance of computer-aided detection (CADe) in various medical imaging tasks. However, in computed tomographic (CT) colonography, the performance is limited by the relatively small size and the variety of the available training datasets. Our purpose in this study was to develop and evaluate a flow-based generative model for performing 3D data augmentation of colorectal polyps for effective training of deep learning in CADe for CT colonography. Methods We developed a 3D-convolutional neural network (3D CNN) based on a flow-based generative model (3D Glow) for generating synthetic volumes of interest (VOIs) that has characteristics similar to those of the VOIs of its training dataset. The 3D Glow was trained to generate synthetic VOIs of polyps by use of our clinical CT colonography case collection. The evaluation was performed by use of a human observer study with three observers and by use of a CADe-based polyp classification study with a 3D DenseNet. Results The area-under-the-curve values of the receiver operating characteristic analysis of the three observers were not statistically significantly different in distinguishing between real polyps and synthetic polyps. When trained with data augmentation by 3D Glow, the 3D DenseNet yielded a statistically significantly higher polyp classification performance than when it was trained with alternative augmentation methods. Conclusion The 3D Glow-generated synthetic polyps are visually indistinguishable from real colorectal polyps. Their application to data augmentation can substantially improve the performance of 3D CNNs in CADe for CT colonography. Thus, 3D Glow is a promising method for improving the performance of deep learning in CADe for CT colonography.

Conditional Generative Adversarial Networks~(CGAN) are a recent and popular method for generating samples from a probability distribution conditioned on latent information. The latent information often comes in the form of a discrete label from a small set. We propose a novel method for training CGANs which allows us to condition on a sequence of continuous latent distributions $f^{(1)}, \ldots, f^{(K)}$. This training allows CGANs to generate samples from a sequence of distributions. We apply our method to paintings from a sequence of artistic movements, where each movement is considered to be its own distribution. Exploiting the temporal aspect of the data, a vector autoregressive (VAR) model is fitted to the means of the latent distributions that we learn, and used for one-step-ahead forecasting, to predict the latent distribution of a future art movement $f^{{(K+1)}}$. Realisations from this distribution can be used by the CGAN to generate "future" paintings. In experiments, this novel methodology generates accurate predictions of the evolution of art. The training set consists of a large dataset of past paintings. While there is no agreement on exactly what current art period we find ourselves in, we test on plausible candidate sets of present art, and show that the mean distance to our predictions is small., Comment: 15 pages, 6 figures

The processing of facial images is an important task, because it is required for a large number of real-world applications. As deep-learning models evolve, they require a huge number of images for training. In reality, however, the number of images available is limited. Generative adversarial networks (GANs) have thus been utilized for database augmentation, but they suffer from unstable training, low visual quality, and a lack of diversity. In this paper, we propose an auto-encoder-based GAN with an enhanced network structure and training scheme for Database (DB) augmentation and image synthesis. Our generator and decoder are divided into two separate modules that each take input vectors for low-level and high-level features, these input vectors affect all layers within the generator and decoder. The effectiveness of the proposed method is demonstrated by comparing it with baseline methods. In addition, we introduce a new scheme that can combine two existing images without the need for extra networks based on the auto-encoder structure of the discriminator in our model. We add a novel double-constraint loss to make the encoded latent vectors equal to the input vectors.

International audience; Computational Protein Design (CPD) has produced impressive results for engineering new proteins, resulting in a wide variety of applications. In the past few years, various efforts have aimed at replacing or improving existing design methods using Deep Learning technology to leverage the amount of publicly available protein data. Deep Learning (DL) is a very powerful tool to extract patterns from raw data, provided that data are formatted as mathematical objects and the architecture processing them is well suited to the targeted problem. In the case of protein data, specific representations are needed for both the amino acid sequence and the protein structure in order to capture respectively 1D and 3D information. As no consensus has been reached about the most suitable representations, this review describes the representations used so far, discusses their strengths and weaknesses, and details their associated DL architecture for design and related tasks.

Despite the importance of few-shot learning, the lack of labeled training data in the real world makes it extremely challenging for existing machine learning methods because this limited dataset does not well represent the data variance. In this research, we suggest employing a generative approach using variational autoencoders (VAEs), which can be used specifically to optimize few-shot learning tasks by generating new samples with more intra-class variations on the Labeled Faces in the Wild (LFW) dataset. The purpose of our research is to increase the size of the training dataset using various methods to improve the accuracy and robustness of the few-shot face recognition. Specifically, we employ the VAE generator to increase the size of the training dataset, including the basic and the novel sets while utilizing transfer learning as the backend. Based on extensive experimental research, we analyze various data augmentation methods to observe how each method affects the accuracy of face recognition. The face generation method based on VAEs with perceptual loss can effectively improve the recognition accuracy rate to 96.47% using both the base and the novel sets.

Designing a molecule with desired properties is one of the biggest challenges in drug development, as it requires optimization of chemical compound structures with respect to many complex properties. To improve the compound design process, we introduce Mol-CycleGAN—a CycleGAN-based model that generates optimized compounds with high structural similarity to the original ones. Namely, given a molecule our model generates a structurally similar one with an optimized value of the considered property. We evaluate the performance of the model on selected optimization objectives related to structural properties (presence of halogen groups, number of aromatic rings) and to a physicochemical property (penalized logP). In the task of optimization of penalized logP of drug-like molecules our model significantly outperforms previous results.

The prediction and design of photonic features have traditionally been guided by theory-driven computational methods, spanning a wide range of direct solvers and optimization techniques. Motivated by enormous advances in the field of machine learning, there has recently been a growing interest in developing complementary data-driven methods for photonics. Here, we demonstrate several predictive and generative data-driven approaches for the characterization and inverse design of photonic crystals. Concretely, we built a data set of 20,000 two-dimensional photonic crystal unit cells and their associated band structures, enabling the training of supervised learning models. Using these data set, we demonstrate a high-accuracy convolutional neural network for band structure prediction, with orders-of-magnitude speedup compared to conventional theory-driven solvers. Separately, we demonstrate an approach to high-throughput inverse design of photonic crystals via generative adversarial networks, with the design goal of substantial transverse-magnetic band gaps. Our work highlights photonic crystals as a natural application domain and test bed for the development of data-driven tools in photonics and the natural sciences.

International audience; Ethics Guidelines for Trustworthy AI advocate for AI technology that is, among other things, more inclusive. Explainable AI (XAI) aims at making state of the art opaque models more transparent, and defends AI-based outcomes endorsed with a rationale explanation, i.e., an explanation that has as target the non-technical users. XAI and Responsible AI principles defend the fact that the audience expertise should be included in the evaluation of explainable AI systems. However, AI has not yet reached all public and audiences , some of which may need it the most. One example of domain where accessibility has not much been influenced by the latest AI advances is cultural heritage. We propose including minorities as special user and evaluator of the latest XAI techniques. In order to define catalytic scenarios for collaboration and improved user experience, we pose some challenges and research questions yet to address by the latest AI models likely to be involved in such synergy.

Online Social Networks (OSNs) allow users to generate and consume content in an easy and personalized way. Among OSNs, Instagram has seen a surge in popularity, and political actors exploit it to reach people at scale, bypassing traditional media and often triggering harsh debates with and among followers. Uncovering the structural properties and dynamics of such interactions is paramount for understanding the online political debate. This is a challenging task due to both the size of the network and the nature of interactions. In this paper, we define a probabilistic model to extract the backbone of the interaction network among Instagram commenters and, after that, we uncover communities. We apply our model to 10 weeks of comments centered around election times in Brazil and Italy. We monitor both politicians and other categories of influencers, finding persistent commenters, i.e., those who often comment together on Instagram posts. Our methodology allows us to unveil interesting facts: i) commenters’ networks are split into few communities; ii) community structure in politics is weaker than in general profiles, indicating that the political debate is a blur, with some commenters bridging strongly opposed political actors; and iii) communities engaging on political profiles are bigger, more active and more stable during electoral period.

Comunicació presentada a: VISIGRAPP 2019: Computer Vision, Imaging and Computer Graphics Theory and Applications. 14th International Joint Conference, celebrat del 25 al 27 de febrer de 2019 a Praga, República Txeca. Image completion or image inpainting is the task of filling in missing regions of an image. When those areas are large and the missing information is unique such that the information and redundancy available in the image is not useful to guide the completion, the task becomes even more challenging. This paper proposes an automatic semantic inpainting method able to reconstruct corrupted information of an image by semantically interpreting the image itself. It is based on an adversarial strategy followed by an energy-based completion algorithm. First, the data latent space is learned by training a modified Wasserstein generative adversarial network. Second, the learned semantic information is combined with a novel optimization loss able to recover missing regions conditioned by the available information. Moreover, we present an application in the context of face inpainting, where our method is used to generate a new face by integrating desired facial attributes or expressions from a reference face. This is achieved by slightly modifying the objective energy. Quantitative and qualitative top-tier results show the power and realism of the presented method. The authors acknowledge partial support by MICINN/FEDER UE project, reference PGC2018-098625-B-I00 and by H2020-MSCA-RISE-2017 project, reference 777826 NoMADS. We also thank NVIDIA for the Quadro P6000 GPU donation.

Deep probabilistic generative models have achieved incredible success in many fields of application. Among such models, variational autoencoders (VAEs) have proved their ability in modeling a generative process by learning a latent representation of the input. In this paper, we propose a novel VAE defined in the quaternion domain, which exploits the properties of quaternion algebra to improve performance while significantly reducing the number of parameters required by the network. The success of the proposed quaternion VAE with respect to traditional VAEs relies on the ability to leverage the internal relations between quaternion-valued input features and on the properties of second-order statistics which allow to define the latent variables in the augmented quaternion domain. In order to show the advantages due to such properties, we define a plain convolutional VAE in the quaternion domain and we evaluate its performance with respect to its real-valued counterpart on the CelebA face dataset., Comment: Accepted for publication at the 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Comunicació presentada a: ICASSP 2020 IEEE International Conference on Acoustics, Speech and Signal Processing, celebrat en línia del 4 al 8 de maig de 2020. We present a deep neural network-based methodology for synthesising percussive sounds with control over high-level timbral characteristics of the sounds. This approach allows for intuitive control of a synthesizer, enabling the user to shape sounds without extensive knowledge of signal processing. We use a feedforward convolutional neural network-based architecture, which is able to map input parameters to the corresponding waveform. We propose two datasets to evaluate our approach on both a restrictive context, and in one covering a broader spectrum of sounds. The timbral features used as parameters are taken from recent literature in signal processing. We also use these features for evaluation and validation of the presented model, to ensure that changing the input parameters produces a congruent waveform with the desired characteristics. Finally, we evaluate the quality of the output sound using a subjective listening test. We provide sound examples and the system's source code for reproducibility. This work is partially funded by the European Unions Horizon 2020 research and innovation programme under the Marie Skodowska-Curie grant agreement No765068, MIP-Frontiers. This work is partially supported by the Towards Richer Online Music Public-domain Archives (TROMPA) project. The TITANX used for this research was donated by the NVIDIA Corporation.

International audience; Multiple Endmember Spectral Mixture Analysis (MESMA) is one of the leading approaches to perform spectral unmixing (SU) considering variability of the endmembers (EMs). It represents each EM in the image using libraries of spectral signatures acquired a priori. However, existing spectral libraries are often small and unable to properly capture the variability of each EM in practical scenes, which compromises the performance of MESMA. In this paper, we propose a library augmentation strategy to increase the diversity of existing spectral libraries, thus improving their ability to represent the materials in real images. First, we leverage the power of deep generative models to learn the statistical distribution of the EMs based on the spectral signatures available in the existing libraries. Afterwards, new samples can be drawn from the learned EM distributions and used to augment the spectral libraries, improving the overall quality of the SU process. Experimental results using synthetic and real data attest the superior performance of the proposed method even under library mismatch conditions.

We investigate the samples derived from generative adversarial networks (GAN) from a classification perspective. We train a classifier on generated samples and on real data and see how they compared on a held out validation set. We see that recent GAN models which produce visually convincing samples are not yet able to match the training on real data. To analyse this we compare training a classifier on generated samples and various sizes of the real training set. We propose architectural and algorithmic changes to reduce this gap. First, we show that a modification to the GAN architecture is needed, which leads to improve generation of samples. Second, we use multiple GAN models as a way to cover the real data distribution, again leading to improvement in classifier training. We also show that in the case of training on small number of samples, a GAN model provides better compression in terms of storage requirements as compared to the real data.

Advancements on text-to-image synthesis generate remarkable images from textual descriptions. However, these methods are designed to generate only one object with varying attributes. They face difficulties with complex descriptions having multiple arbitrary objects since it would require information on the placement and sizes of each object in the image. Recently, a method that infers object layouts from scene graphs has been proposed as a solution to this problem. However, their method uses only object labels in describing the layout, which fail to capture the appearance of some objects. Moreover, their model is biased towards generating rectangular shaped objects in the absence of ground-truth masks. In this paper, we propose an object encoding module to capture object features and use it as additional information to the image generation network. We also introduce a graph-cuts based segmentation method that can infer the masks of objects from bounding boxes to better model object shapes. Our method produces more discernible images with more realistic shapes as compared to the images generated by the current state-of-the-art method.

Due to the inherent randomness of both solar power generation and residential electrical load, jointly sizing solar panel and storage capacity to meet a given quality-of-service (QoS) constraint is challenging. The challenge is greater when there is limited representative historical data. We therefore propose generating synthetic solar and load traces, corresponding to different realizations of the underlying stochastic processes. Specifically, we compare the effectiveness of three generative models: autoregressive moving-average (ARMA) models, Gaussian mixture models (GMMs), and generative adversarial networks (GANs) -- as well as two direct sampling methods -- for synthetic trace generation. These traces are then used for robust joint sizing by a technique described in recent work. Extensive experiments based on real data show that our approach finds robust sizing with only one year's worth of hourly trace data. Moreover, assuming that solar data are available, given a database of load traces, we demonstrate how to perform robust sizing with access to only twelve data points of load, one for each month of one year., e-Energy '19: The Tenth ACM International Conference on Future Energy Systems, June 25-28, 2019, Phoenix, Arizona

The inclusion of tracking technologies in personal devices opened the doors to the analysis of large sets of mobility data like GPS traces and call detail records. This tutorial presents an overview of both modeling principles of human mobility and machine learning models applicable to specific problems. We review the state of the art of five main aspects in human mobility: (1) human mobility data landscape; (2) key measures of individual and collective mobility; (3) generative models at the level of individual, population and mixture of the two; (4) next location prediction algorithms; (5) applications for social good. For each aspect, we show experiments and simulations using the Python library "scikit-mobility" developed by the presenters of the tutorial.