Sorry, you need to enable JavaScript to visit this website.

Convolutional neural networks (CNN) have emerged as a powerful tool for solving computational imaging reconstruction problems. However, CNNs are generally difficult-to-understand black-boxes. Accordingly, it is challenging to know when they will work and, more importantly, when they will fail.

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.

In several applications, including imaging of deformable objects while in motion, simultaneous localization and mapping, and unlabeled sensing, we encounter the problem of recovering a signal that is measured subject to unknown permutations. In this paper we take a fresh look at this problem through the lens of optimal transport (OT). In particular, we recognize that in most practical applications the unknown permutations are not arbitrary but some are more likely to occur than others.

Sound Event Detection and Audio Classification tasks are traditionally addressed through time-frequency representations of audio signals such as spectrograms. However, the emergence of deep neural networks as efficient feature extractors has enabled the direct use of audio signals for classification purposes. In this paper, we attempt to recognize musical instruments in polyphonic audio by only feeding their raw waveforms into deep learning models.

The recognition of music genre and the discrimination between music and speech are important components of modern digital music systems. Depending on the acquisition conditions, such as background environment, these signals may come from different probability distributions, making the learning problem complicated. In this context, domain adaptation is a key theory to improve performance. Considering data coming from various background conditions, the adaptation scenario is called multi-source.

In this paper, we present a simple but novel framework to train a non-parallel many-to-many voice conversion (VC) model based on the encoder-decoder architecture. It is observed that an encoder-decoder text-to-speech (TTS) model and an encoder-decoder VC model have the same structure.

The typical problem like insufficient training instances in time series classification task demands for novel deep neural architecture to warrant consistent and accurate performance. Deep Residual Network (ResNet) learns through H(x)=F(x)+x, where F(x) is a nonlinear function. We propose Blend-Res2Net that blends two different representation spaces: H^1 (x)=F(x)+Trans(x) and H^2 (x)=F(Trans(x))+x with the intention of learning over richer representation by capturing the temporal as well as the spectral signatures (Trans(∙) represents the transformation function).