In theory, the variational auto-encoder (VAE) is not suitable for recommendation tasks, although it has been successfully utilized for collaborative filtering (CF) models. In this paper, we propose a Gaussian Copula-V...
详细信息
ISBN:
(纸本)9781450384469
In theory, the variational auto-encoder (VAE) is not suitable for recommendation tasks, although it has been successfully utilized for collaborative filtering (CF) models. In this paper, we propose a Gaussian Copula-Vector Quantized autoencoder (GC-VQAE) model that differs prior arts in two key ways: (1) Gaussian Copula helps to model the dependencies among latent variables which are used to construct a more complex distribution compared with the meanfield theory;and (2) by incorporating a vector quantisation method into encoders our model can learn discrete representations which are consistent with the observed data rather than directly sampling from the simple Gaussian distributions. Our approach is able to circumvent the "posterior collapse" issue and break the prior constraint to improve the flexibility of latent vector encoding and learning ability. Empirically, GC-VQAE can significantly improve the recommendation performance compared to existing state-of-the-art methods.
Automatic segmentation of brain abnormalities is challenging, as they vary considerably from one pathology to another. Current methods are supervised and require numerous annotated images for each pathology, a strenuo...
详细信息
ISBN:
(纸本)9781665412469
Automatic segmentation of brain abnormalities is challenging, as they vary considerably from one pathology to another. Current methods are supervised and require numerous annotated images for each pathology, a strenuous task. To tackle anatomical variability, Unsupervised Anomaly Detection (UAD) methods are proposed, detecting anomalies as outliers of a healthy model learned using a variational autoencoder (VAE). Previous work on UAD adopted 2D approaches, meaning that MRIs are processed as a collection of independent slices. Yet, it does not fully exploit the spatial information contained in MRI. Here, we propose to perform UAD in a 3D fashion and compare 2D and 3D VAEs. As a side contribution, we present a new loss function guarantying a robust training. Learning is performed using a multicentric dataset of healthy brain MRIs, and segmentation performances are estimated on White-Matter Hyperintensities and tumors lesions. Experiments demonstrate the interest of 3D methods which outperform their 2D counterparts.
We present a multi-module framework based on Conditional variational autoencoder (CVAE) to detect anomalies in the power signals coming from multiple High Voltage Converter Modulators (HVCMs). We condition the model w...
详细信息
We present a multi-module framework based on Conditional variational autoencoder (CVAE) to detect anomalies in the power signals coming from multiple High Voltage Converter Modulators (HVCMs). We condition the model with the specific modulator type to capture different representations of the normal waveforms and to improve the sensitivity of the model to identify a specific type of fault when we have limited samples for a given module type. We studied several Artificial Neural Network (ANN) architectures for our CVAE model and evaluated the model performance by looking at their loss landscape for stability and generalization. Our results for the Spallation Neutron Source (SNS) experimental data show that the trained model generalizes well to detecting multiple fault types for several HVCM module types. The results of this study can be used to improve the HVCM reliability and overall SNS uptime.
Recommender systems have become indispensable for several Web sites, helping users deal with big amounts of data. They are capable of analyzing user/item interactions taking place on-line, and provide each user with a...
详细信息
ISBN:
(纸本)9781450362382
Recommender systems have become indispensable for several Web sites, helping users deal with big amounts of data. They are capable of analyzing user/item interactions taking place on-line, and provide each user with a list of suggestions sorted by relevance. Items with the same or very close relevance, however, may occupy different positions in the ranking and may be exposed to completely different levels of attention. This promotes unfair treatment and can only be addressed by a long term strategy. variational autoencoders (VAEs) were recently proposed as the state-of-the-art for collaborative filtering recommendations, but as every other approach, they generate homogeneous prediction scores among the highest positions. In this paper, we propose incorporating randomness in the regular operation of VAEs in order to increase the fairness (mitigate the position bias) in multiple rounds of recommendation. We argue that adding a noise component when sampling values from VAE's latent representation provides long term fairness, despite of a tolerable decrease in ranking quality (NDCG). We calculate the trade-off between unfairness and NDCG when introducing 4 different noise distributions. The solution has proved to be a very practical one and the results point for a clear positive effect of turning recommendation far more fair, despite some small NDCG loss in Movie Lens, Netflix and MSD datasets. In our best scenario, the unfairness was reduced by 76% despite a decrease of 5% in the quality of ranking.
Traditional approaches to terrain heightmap generation rely on geometric methods to generate a matrix of elevation values using noise functions. More advanced approaches attempt to model the natural processes that sha...
详细信息
ISBN:
(纸本)9780738133669
Traditional approaches to terrain heightmap generation rely on geometric methods to generate a matrix of elevation values using noise functions. More advanced approaches attempt to model the natural processes that shape landmasses in the real world, such as wind, moisture, and rainfall. This survey leverages recent advancements in image generation using generative machine learning models in order to evaluate their effectiveness in this problem space. A variational autoencoder (VAE), generative adversarial network (GAN), and PixelCNN network are trained on real-world island heightmap data and produce realistic island terrain. The author compares the results in terms of image quality and "closeness" to the original real images, and evaluates the trade-off between quality and training/generation speed.
Besides its linguistic content, our speech is rich in biometric information that can be inferred by classifiers. Learning privacy-preserving representations for speech signals enables downstream tasks without sharing ...
详细信息
ISBN:
(纸本)9781713836902
Besides its linguistic content, our speech is rich in biometric information that can be inferred by classifiers. Learning privacy-preserving representations for speech signals enables downstream tasks without sharing unnecessary, private information about an individual. In this paper, we show that protecting gender information in speech is more effective than modelling speaker-identity information only when generating a nonsensitive representation of speech. Our method relies on reconstructing speech by decoding linguistic content along with gender information using a variational autoencoder. Specifically, we exploit disentangled representation learning to encode information about different attributes into separate subspaces that can be factorised independently. We present a novel way to encode gender information and disentangle two sensitive biometric identifiers, namely gender and identity, in a privacyprotecting setting. Experiments on the LibriSpeech dataset show that gender recognition and speaker verification can be reduced to a random guess, protecting against classification-based attacks.
Although the main structures of cortical folding are present in each human brain, the folding pattern is unique to each individual. Because of this large normal variability, the identification of abnormal patterns ass...
详细信息
ISBN:
(纸本)9783030875862;9783030875855
Although the main structures of cortical folding are present in each human brain, the folding pattern is unique to each individual. Because of this large normal variability, the identification of abnormal patterns associated to developmental disorders is a complex open challenge. In this paper, we tackle this problem as an anomaly detection task and explore the potential of deep generative models using benchmarks made up of synthetic anomalies. To focus learning on the folding geometry, brain MRI are preprocessed first to deal only with a skeleton-based negative cast of the cortex. A variational auto-encoder is trained to get a representation of the regional variability of the folding pattern of the general population. Then several synthetic benchmark datasets of abnormalities are designed. The latent space expressivity is assessed through classification experiments between control's and abnormal's latent codes. Finally, the properties encoded in the latent space are analyzed through perturbation of specific latent dimensions and observation of the resulting modification of the reconstructed images. The results have shown that the latent representation is rich enough to distinguish subtle differences like asymmetries between the right and left hemispheres.
To enjoy fishing indoors, we study a hardware-type fishing simulator that employs a real fishing rod. In this paper, as the first step of our research, we develop a pull force acquisition system and a winding system t...
详细信息
ISBN:
(纸本)9781665449588
To enjoy fishing indoors, we study a hardware-type fishing simulator that employs a real fishing rod. In this paper, as the first step of our research, we develop a pull force acquisition system and a winding system that consists of a motor, reel, and controller. The pull force acquisition system obtains an actual fish pull force. The time variation of the pull force represents the pull pattern of the fish. We show that the winding system can reproduce a pull pattern similar to the original pull pattern obtained at the pull force acquisition system. A lot of pull pattern of fish has to be acquired to represent a specific pull pattern to the fish species. It is inefficient to obtain them with fieldwork. We use a variational autoencoder (VAE) to generate multiple pull patterns similar to the original pull pattern. Here we assume that the fish species-specific pull pattern maintains its rough shape of movements. Simulation results showed that VAE generated multiple pull patterns roughly maintaining the original shape.
When designing variational autoencoders (VAEs) or other types of latent space models, the dimensionality of the latent space is typically defined upfront. In this process, it is possible that the number of dimensions ...
详细信息
ISBN:
(纸本)9780738133669
When designing variational autoencoders (VAEs) or other types of latent space models, the dimensionality of the latent space is typically defined upfront. In this process, it is possible that the number of dimensions is under- or overprovisioned for the application at hand. In case the dimensionality is not predefined, this parameter is usually determined using time- and resource-consuming cross-validation. For these reasons we have developed a technique to shrink the latent space dimensionality of VAEs automatically and on-the-fly during training using Generalized ELBO with Constrained Optimization (GECO) and the L-0-Augment-REINFORCE-Merge (L-0-ARM) gradient estimator. The GECO optimizer ensures that we are not violating a predefined upper bound on the reconstruction error. This paper presents the algorithmic details of our method along with experimental results on five different datasets. We find that our training procedure is stable and that the latent space can be pruned effectively without violating the GECO constraints.
Traffic matrices (TMs) contain information that is essential for network management, traffic engineering, and anomaly detection. However, constructing a TM through direct traffic measurements has a high administrative...
详细信息
ISBN:
(纸本)9780738113302
Traffic matrices (TMs) contain information that is essential for network management, traffic engineering, and anomaly detection. However, constructing a TM through direct traffic measurements has a high administrative and computational cost. A more feasible approach is to estimate the TM from the easily obtainable link load measurements. In this paper, we address the issue of traffic matrix estimation (TME) from link loads using a deep generative model - namely, a variational autoencoder (VAE) - to solve the respective ill-posed inverse problem. In particular, we train the VAE with historical data (previously observed TMs) and we leverage the trained decoder to transform TME into a minimization problem in the latent space, which in turn can be solved by employing a gradient-based optimizer. Furthermore, the trained decoder can be used for traffic matrix synthesis, i.e., for generating synthetic TM examples that have "similar" properties to the samples of the training set. Finally, we explore the incremental optimization of the sequence of objectives constructed from the sequence of decoders that we obtain at different stages of the VAE training. The performance of the proposed methods is evaluated using a publicly available dataset of actual traffic matrices recorded in a real backbone network.
暂无评论