We propose a new method for approaching novelty/anomaly detection. Novelty detection is needed when you want to distinguish between inlier and outlier samples. However, you have training data only for inliers, while for outliers training data is not available. This is a very frequent scenario for real-world problems when outliers are infrequent and may have unknown distribution.

Deep hashing approaches are widely applied to approximate nearest neighbor search for large-scale image retrieval. We propose Spherical Deep Supervised Hashing (SDSH), a new supervised deep hashing approach to learn compact binary codes. The goal of SDSH is to go beyond learning similarity preserving codes, by encouraging them to also be balanced and to maximize the mean average precision. This is enabled by advocating the use of a different relaxation method, allowing the learning of a spherical embedding, which overcomes the challenge of maintaining the learning problem well-posed without the need to add extra binarizing priors. This allows the formulation of a general triplet loss framework, with the introduction of the spring loss for learning balanced codes, and of the ability to learn an embedding quantization that maximizes the mean average precision. Extensive experiments demonstrate that the approach compares favorably with the state-of-the-art while providing significant performance increase at more compact code sizes.

The quest for deeper understanding of biological systems has driven the acquisition of increasingly larger multi-dimensional image datasets. Inspecting and manipulating data of this complexity is very challenging in traditional visualization systems. We developed syGlass, a software package capable of visualizing large-scale volumetric data with inexpensive virtual reality head-mounted display technology. This allows leveraging stereoscopic vision to significantly improve perception of complex 3D structures, and provides immersive interaction with data directly in 3D. We accomplished this by developing highly optimized data flow and volume rendering pipelines, tested on datasets up to 16TB in size, as well as tools available in a virtual reality GUI to support advanced data exploration, annotation, and cataloging.

Abstract: Autoencoder networks are unsupervised approaches aiming at combining generative and representational properties by learning simultaneously an encoder-generator map. Although studied extensively, the issues of whether they have the same generative power of GANs, or learn disentangled representations, have not been fully addressed. We introduce an autoencoder that tackles these issues jointly, which we call Adversarial Latent Autoencoder (ALAE). It is a general architecture that can leverage recent improvements on GAN training procedures. We designed two autoencoders: one based on a MLP encoder, and another based on a StyleGAN generator, which we call StyleALAE. We verify the disentanglement properties of both architectures. We show that StyleALAE can not only generate 1024x1024 face images with comparable quality of StyleGAN, but at the same resolution can also produce face reconstructions and manipulations based on real images. This makes ALAE the first autoencoder able to compare with, and go beyond the capabilities of a generator-only type of architecture.