Graph autoencoders (AE) and variational autoencoders (VAE) recently emerged as powerful node embedding methods. In particular, graph AE and VAE were successfully leveraged to tackle the challenging link prediction problem, aiming at figuring out whether some pairs of nodes from a graph are connected by unobserved edges. However, these models focus on undirected graphs and therefore ignore the potential direction of the link, which is limiting for numerous real-life applications. In this paper, we extend the graph AE and VAE frameworks to address link prediction in directed graphs. We present a new gravity-inspired decoder scheme that can effectively reconstruct directed graphs from a node embedding. We empirically evaluate our method on three different directed link prediction tasks, for which standard graph AE and VAE perform poorly. We achieve competitive results on three real-world graphs, outperforming several popular baselines.

Distance metric learning based on triplet loss has been applied with success in a wide range of applications such as face recognition, image retrieval, speaker change detection and recently recommendation with the Collaborative Metric Learning (CML) model. However, as we show in this article, CML requires large batches to work reasonably well because of a too simplistic uniform negative sampling strategy for selecting triplets. Due to memory limitations, this makes it difficult to scale in high-dimensional scenarios. To alleviate this problem, we propose here a 2-stage negative sampling strategy which finds triplets that are highly informative for learning. Our strategy allows CML to work effectively in terms of accuracy and popularity bias, even when the batch size is an order of magnitude smaller than what would be needed with the default uniform sampling. We demonstrate the suitability of the proposed strategy for recommendation and exhibit consistent positive results across various datasets.

Exploiting a tissue-conductive sensor -a stethoscopic microphone -the system developed at NAIST which converts Non-Audible Murmur (NAM) to audible speech by GMM-based statistical mapping is a very promising technique. The quality of the converted speech is however still insufficient for computer-mediated communication, notably because of the poor estimation of F 0 from unvoiced speech and because of impoverished phonetic contrasts. This paper presents our investigations to improve the intelligibility and naturalness of the synthesized speech and first objective and subjective evaluations of the resulting system. The first improvement concerns voicing and F 0 estimation. Instead of using a single GMM for both, we estimate a continuous F 0 using a GMM, trained on target voiced segments only. The continuous F 0 estimation is filtered by a voicing decision computed by a neural network. The objective and subjective improvement is significant. The second improvement concerns the input time window and its dimensionality reduction: we show that the precision of F 0 estimation is also significantly improved by extending the input time window from 90 to 450ms and by using a Linear Discriminant Analysis (LDA) instead of the original Principal Component Analysis (PCA). Estimation of spectral envelope is also slightly improved with LDA but is degraded with larger time windows. A third improvement consists in adding visual parameters both as input and output parameters. The positive contribution ACCEPTED MANUSCRIPT of this information is confirmed by a subjective test. Finally, HMM-based conversion is compared with GMM-based conversion.

No abstract