ECG signal generation based on conditional generative models

A conditional variational autoencoding generative adversarial networks with self-modulation for rolling bearing fault diagnosis

In this paper, we tackle the well-known problem of dataset construction from the point of its generation using generative adversarial networks (GAN). As semantic information of the dataset should have a proper alignment with images, controlling the image generation process of GAN comes to the first position. Considering this, we focus on conditioning the generative process by solely utilizing conditional information to achieve reliable control over the image generation. Unlike the existing works that consider the input (noise or image) in conjunction with conditions, our work considers transforming the input directly to the conditional space by utilizing the given conditions only. By doing so, we reveal the relations between conditions to determine their distinct and reliable feature space without the impact of input information. To fully leverage the conditional information, we propose a novel architectural framework (i.e., conditional transformation) that aims to learn features only from a set of conditions for guiding a generative model by transforming the input to the generator. Such an approach enables controlling the generator by setting its inputs according to the specific conditions necessary for semantically correct image generation. Given that the framework operates at the initial stage of generation, it can be plugged into any existing generative models and trained in an end-to-end manner together with the generator. Extensive experiments on various tasks, such as novel image synthesis and image-to-image translation, demonstrate that the conditional transformation of inputs facilitates solid control over the image generation process and thus shows its applicability for use in dataset construction.

Research on fault diagnosis of industrial robots based on generative adversarial network

PO-04-175 RECONSTRUCTING TWELVE-LEAD ELECTROCARDIOGRAM SIGNALS FROM A SINGLE-LEAD SIGNAL USING GENERATIVE ADVERSARIAL NETWORKS

Recent advancements in conditional generative models, e.g., Conditional Variational AutoEncoder (CVAE) and Conditional Denoising Diffusion Probabilistic Model (CDDPM), which utilize class-specific embeddings for generating images in specific classes, have demonstrated exceptional capabilities in producing high-quality images on balanced datasets. However, their performance decreases with imbalanced real-world data, affecting the fidelity and diversity of the generated images. This paper discusses the reasons and solutions for the imbalance issue in CVAE and CDDPM. By selectively reweighting the gradients of the embedding layer and main network, we identify the embedding layer’s bias as a critical bottleneck in these models. Motivated by this, we propose Embedding Pretraining and Regularization (PREmbed), a training methodology for addressing biased learning in the embedding layer of these models. PREmbed consists of two key components: (i) Masked Autoencoder pretraining on the original imbalanced dataset to obtain a robust initial embedding, and (ii) an embedding regularization loss to preserve class-level distances during the training phase to improve imbalanced embedding learning. Implementing PREmbed leads to the enhanced convergence of a robust embedding layer. Our experiments on imbalanced CIFAR-10 and CUB-200 datasets demonstrate that PREmbed enhances the performance of CVAE and CDDPM; it outperforms the baseline CVAE by 22.3% (FID 18.63) and the CDDPM by 30.3% (FID 12.70) in imbalanced CIFAR-10; 33.7% (FID 24.08) and 26.5% (FID 13.90) in CUB-200; and 16.4% (FID 49.41) and 15.2% (FID 33.05) in ImageNet-LT (which is where it achieves the state-of-the-art performance in imbalanced conditional image generation task).

Multi-modal generative adversarial networks for synthesizing time-series structural impact responses

Electrocardiogram (ECG) signal processing aims basically 1) at artifact reduction to make the ECG signals cleaner and better interpretable by human or machine observers, 2) at revealing aspects not immediately observable in plain measured ECG signals even after artifact reduction, or 3) at diagnostics decision support and automated ECG signal interpretation, including classification of ECG signals into different classes associated with normal or pathological heart function. Thus, sports related applications aside, body surface ECG signal processing aims at enhancing ECG based diagnostics. In this Chapter, we review and demonstrate a statistical signal processing approach, independent component analysis (ICA), which is inherently very suitable for ECG signal processing regarding the aims 1) and 2) above, and also equally applicable as a component in systems aimed at accomplishing the aim 3). For more on general ECG signal processing, the reader is directed to the textbook written by Sornmo & Laguna (2005), and for a thorough treatment of ICA to the Hyvarinen’s book (Hyvarinen et al., 2001). A concise review on ICA in ECG signal processing has been presented by Castells et al. (2007a). In this Chapter, we describe and illustrate several widely adopted applications of ICA in ECG signal processing, and discuss associated practical aspects, some of which are not generally found in the literature. The treatment of the matter is aimed at conceptual and practical understanding, leaving the mathematical derivations and proofs far mostly for the interested reader to find in the references. ICA (Castells et al., 2007a; Comon, 1994; Hyvarinen et al., 2001; Hyvarinen & Oja, 2000; Naik & Kumar, 2011) is a statistical signal processing method for decomposing a set of signals into a set of mutually independent component signals. In general, in the applications of ICA, including in ECG signal processing, the objective is that the resulting independent component signals are the original source signals. Since ICA operates purely based on the input signals and a few assumptions, ICA belongs to the class of methods called blind source separation methods. For ICA, a source signal is called an independent component (IC). The terms ‘IC’ and ‘source signal’ are here used interchangeably. For ECG, the source signals are the bioelectrical signals generated by the heart, and all the possible artifact signals. Generally, ICA input signals are the observed signals, which may be measurement time series, such as sampled voltage values in time as in the case of ECGs, image pixel values, or basically any sets of values fulfilling the assumptions of ICA. In the sequel, the term

In the field of deep learning, the generative model did not attract much attention until GANs (generative adversarial networks) appeared. In 2014, Google’s Ian Goodfellow proposed a generative model called GANs. GANs use different structures and objective functions from the existing generative model. For example, GANs use two neural networks: a generator that creates a realistic image, and a discriminator that distinguishes whether the input is real or synthetic. If there are no problems in the training process, GANs can generate images that are difficult even for experts to distinguish in terms of authenticity. Currently, GANs are the most researched subject in the field of computer vision, which deals with the technology of image style translation, synthesis, and generation, and various models have been unveiled. The issues raised are also improving one by one. In image synthesis, BEGAN (Boundary Equilibrium Generative Adversarial Network), which outperforms the previously announced GANs, learns the latent space of the image, while balancing the generator and discriminator. Nonetheless, BEGAN also has a mode collapse wherein the generator generates only a few images or a single one. Although BEGAN-CS (Boundary Equilibrium Generative Adversarial Network with Constrained Space), which was improved in terms of loss function, was introduced, it did not solve the mode collapse. The discriminator structure of BEGAN-CS is AE (AutoEncoder), which cannot create a particularly useful or structured latent space. Compression performance is not good either. In this paper, this characteristic of AE is considered to be related to the occurrence of mode collapse. Thus, we used VAE (Variational AutoEncoder), which added statistical techniques to AE. As a result of the experiment, the proposed model did not cause mode collapse but converged to a better state than BEGAN-CS.

Regular sampled seismic data is important for seismic data processing. However, seismic data is often missing due to natural or economic reasons. Especially, when encountering big obstacles, the seismic data will be missing in big gaps, which is more difficult to be reconstructed. Conditional generative adversarial networks (cGANs) are deep-learning models learning the characteristics of the seismic data to reconstruct the missing data. In this paper, we use a conditional Wasserstein generative adversarial network (cWGAN) to interpolate the missing seismic data in big gaps. We use the Wasserstein loss function to train the network and use a gradient penalty in the WGAN (cWGAN-GP) to enforce the Lipschitz constraint. We use a pre-stack seismic dataset to assess the performance. The interpolated results and the calculated recovered signal-to-noise ratios indicate that the cWGAN-GP can recover the missing seismic traces in portions or the entire regions, and the cWGAN-GP based interpolation is more accurate than the cGAN.

A tropical cyclone intensity prediction model using conditional generative adversarial network

The recent success of Generative Adversarial Networks (GAN) is a result of their ability to generate high quality images given samples from a latent space. One of the applications of GANs is to generate images from a text description, where the text is first encoded and further used for the conditioning in the generative model. In addition to text, conditional generative models often use label information for conditioning. Hence, the structure of the meta-data and the ontology of the labels is important for such models. In this paper, we propose Ontology Generative Adversarial Networks (O-GANs) to handle the complexities of the data with label ontology. We evaluate our model on a dataset of fashion images with hierarchical label structure. Our results suggest that the incorporation of the ontology, leads to better image quality as measured by Frechet Inception Distance and Inception Score. Additionally, we show that the O-GAN better matches the generated images to their conditioning text, compared to models that do not incorporate the label ontology.

Metamaterials are subwavelength-sized artificial structures with the ability to manipulate incident waves in such a way that affects how the energy propagates throughout the medium. In acoustics, particularly placed scattering elements can reduce the total scattering cross section (TSCS) response. We propose a method to inversely design acoustic metamaterial configurations using deep learning and generative modeling. Using our proprietary multiple scattering solver with MATLAB optimization toolbox, we generate a dataset of optimal configurations with minimized TSCS within a discrete range of wavenumbers. We use this dataset to train a Conditional Wasserstein Generative Adversarial Network (cWGAN) to generate similar metacluster designs corresponding to specified input TSCS. To improve the coordinate recognition ability of the cWGAN, we include the novel CoordConv layer in the generator and critic. After training, the cWGAN can produce a variety of optimal configurations given an expected TSCS. Evaluating TSCS of generated configurations shows that the model is capable of proposing scatterer configurations that are comparable or better than the dataset within the optimized range.

Generative AI enables realistic data generation, making it essential for tasks such as text, image, and video synthesis. In robotics, predicting unknown map regions as an image generation problem is key to improving exploration and navigation. This study uniquely compares the performance of three deep generative models—Conditional Variational Autoencoder (CVAE), Vision Transformer (ViT), and Conditional Generative Adversarial Network (CGAN)—for map completion. Unlike prior studies focusing on individual models, this work provides a direct performance comparison for global map prediction. A custom dataset, derived from HouseExpo and enriched with structured time-series exploration data in Gazebo, was developed for this purpose. Experimental results demonstrate that CGAN achieves superior performance in completing unknown regions, making it more effective for robotic exploration. Additionally, we introduce a novel dataset generation methodology leveraging ROS, Gazebo, and Voronoi-based mapping. Future research will focus on scaling the dataset and integrating these models into exploration systems, further advancing generative AI applications in robotics.

The Deepfake phenomenon has become very popular nowadays thanks to the possibility to create incredibly realistic images using deep learning tools, based mainly on ad-hoc Generative Adversarial Networks (GAN). In this work we focus on the analysis of Deepfakes of human faces with the objective of creating a new detection method able to detect a forensics trace hidden in images: a sort of fingerprint left in the image generation process. The proposed technique, by means of an Expectation Maximization (EM) algorithm, extracts a set of local features specifically addressed to model the underlying convolutional generative process. Ad-hoc validation has been employed through experimental tests with naive classifiers on five different architectures (GDWCT, STARGAN, ATTGAN, STYLEGAN, STYLEGAN2) against the CELEBA dataset as ground-truth for non-fakes. Results demonstrated the effectiveness of the technique in distinguishing the different architectures and the corresponding generation process.

Hierarchical deep learning with Generative Adversarial Network for automatic cardiac diagnosis from ECG signals