A Geometric Understanding of Deep Learning

This work introduces an optimal transportation(OT)view of generative adversarial networks(GANs).Natural datasets have intrinsic patterns,which can be summarized as the manifold distribution principle:the distribution of a class of data is close to a low-dimensional manifold.GANs mainly accomplish two tasks:manifold learning and probability distribution transformation.The latter can be carried out using the classical OT method.From the OT perspective,the generator computes the OT map,while the discriminator computes the Wasserstein distance between the generated data distribution and the real data distribution;both can be reduced to a convex geometric optimization process.Furthermore,OT theory discovers the intrinsic collaborative-instead of competitive-relation between the generator and the discriminator,and the fundamental reason for mode collapse.We also propose a novel generative model,which uses an autoencoder(AE)for manifold learning and OT map for probability distribution transformation.This AE–OT model improves the theoretical rigor and transparency,as well as the computational stability and efficiency;in particular,it eliminates the mode collapse.The experimental results validate our hypothesis,and demonstrate the advantages of our proposed model.

Quantum Generative Adversarial Network: A Survey

Generative adversarial network(GAN)is one of the most promising methods for unsupervised learning in recent years.GAN works via adversarial training concept and has shown excellent performance in the fields image synthesis,image super-resolution,video generation,image translation,etc.Compared with classical algorithms,quantum algorithms have their unique advantages in dealing with complex tasks,quantum machine learning(QML)is one of the most promising quantum algorithms with the rapid development of quantum technology.Specifically,Quantum generative adversarial network(QGAN)has shown the potential exponential quantum speedups in terms of performance.Meanwhile,QGAN also exhibits some problems,such as barren plateaus,unstable gradient,model collapse,absent complete scientific evaluation system,etc.How to improve the theory of QGAN and apply it that have attracted some researcher.In this paper,we comprehensively and deeply review recently proposed GAN and QAGN models and their applications,and we discuss the existing problems and future research trends of QGAN.

Autoencoder-based conditional optimal transport generative adversarial network for medical image generation

Medical image generation has recently garnered significant interest among researchers.However,the primary generative models,such as Generative Adversarial Networks(GANs),often encounter challenges during training,including mode collapse.To address these issues,we proposed the AECOT-GAN model(Autoencoder-based Conditional Optimal Transport Generative Adversarial Network)for the generation of medical images belonging to specific categories.The training process of our model comprises three fundamental components.The training process of our model encompasses three fundamental components.First,we employ an autoencoder model to obtain a low-dimensional manifold representation of real images.Second,we apply extended semi-discrete optimal transport to map Gaussian noise distribution to the latent space distribution and obtain corresponding labels effectively.This procedure leads to the generation of new latent codes with known labels.Finally,we integrate a GAN to train the decoder further to generate medical images.To evaluate the performance of the AE-COT-GAN model,we conducted experiments on two medical image datasets,namely DermaMNIST and BloodMNIST.The model’s performance was compared with state-of-the-art generative models.Results show that the AE-COT-GAN model had excellent performance in generating medical images.Moreover,it effectively addressed the common issues associated with traditional GANs.

Facial Expression Recognition Based on PCD-CNN with Pose and Expression

In order to achieve high recognition rate,most facial expression recognition(FER)methods generate sufficient labeled facial images based on generative adversarial networks(GAN)to train model.However,these methods do not estimate the facial pose before passing the images to the generator,which affects the quality of generated images.And mode collapse is prone to occur during the training process,leading to generate a single-style facial images.To solve these problems,a FER model is proposed based on pose conditioned dendritic convolution neural network(PCD-CNN)with pose and expression.Before passing the facial images to the generator,PCD-CNN was used to process facial images,effectively estimating the facial landmarks to detect face and disentangle the pose.In order to accelerate the training speed of the model,PCD-CNN was based on the ShuffleNet-v2 framework.Every landmark of facial image was modeled by a separate ShuffleNet-DeconvNet,maintaining better performance with fewer parameters.To solve the mode collapse during image generation,we theoretically analyzed the causes,and implemented mini-batch processing on the discriminator in the model and directly calculated the statistical characteristics of the mini-batch samples.Experiments were carried out on the Multi-PIE and BU-3DFE facial expression datasets.Compared with current advanced methods,our method achieves higher accuracy 93.08%,and the training process is more stable.

Research on failure scenarios of domes based on form vulnerability

In this paper, form vulnerability theory was applied to the analysis of the failure mechanisms of single-layer latticed spherical shells subjected to seismic excitations. Three 1/10 scale testing models were designed with characteristics as follows： Model 1 possesses overall uniform stiffness and is expected to collapse in the strength failure mode as some members become plastic; Model 2 possesses six man-made weak parts located on six radial main rib zones and is expected to collapse in the dynamic in- stability mode with all members still in the elastic stage; Model 3 strengthens the six weak zones of Model 2, and therefore, its stiffness is uniform. Model 3 is proposed to collapse in the strength failure mode when the members are still in the elastic stage By increasing the peak ground accelerations of seismic waves gradually, the shaking table tests were carried out until all three models collapsed （or locally collapsed）. On the basis of form vulnerability theory, topological hierarchy models of the test models were established through a clustering process, and various failure scenarios, including overall collapse scenarios and partial collapse scenarios, were identified by unzipping corresponding hierarchical models. By comparison of the failure scenarios based on theoretical analysis and experiments, it was found that vulnerability theory could effectively reflect the weak- ness zones in topological relations of the structures from the perspective of internal causes. The intemal mechanisms of the distinct failure characteristics of reticulated shells subjected to seismic excitations were also revealed in this process. The well-formedness of structural clusters, Q, is closely related to the collapse modes, i.e., uniform changes of Q indicate a uniform distribution of overall structural stiffness, which indicates that strength failure is likely to happen; conversely, non-uniform changes of Q indicate that weak zones exist in the structure, and dynamic instability is likely to occur.