This repository contains the core services of the FfDL (Fabric for Deep Learning) platform. FfDL is an operating system "fabric" for Deep Learning Once installed, use the command make minikube to start Minikube and set up local network routes. The minimum recommended capacity for FfDL is 4GB Memory and 2 CPUs. If you already have a FfDL deployment up and running, you can jump to FfDL User Guide to use FfDL for training your deep learning models. If you are getting started and want to setup your own FfDL deployment, please follow the steps below.

Friday, March 30, 2018 The Linux Foundation has launched the LF Deep Learning Foundation, an umbrella organization that will support and sustain open source innovation in artificial intelligence, machine learning, and deep learning while striving to make these critical new technologies available to developers and data scientists everywhere. Founding members of LF Deep Learning include Amdocs, AT&T, B.Yond, Baidu, Huawei, Nokia, Tech Mahindra, Tencent, Univa, and ZTE. With LF Deep Learning, members are working to create a neutral space where makers and sustainers of tools and infrastructure can interact and harmonize their efforts and accelerate the broad adoption of deep learning technologies. "We are excited to offer a deep learning foundation that can drive long-term strategy and support for a host of projects in the AI, machine learning, and deep learning ecosystems," said Jim Zemlin, executive director of The Linux Foundation. "With LF Deep Learning, we are launching the Acumos AI Project, a comprehensive platform for AI model discovery, development and sharing. In addition, we are pleased to announce that Baidu and Tencent each intend to contribute projects to LF Deep Learning. LF Deep Learning enables the open source community to support entire ecosystems of projects in these spaces, and we invit e the open source community to join us in this effort."

More companies are buying into the hype of artificial intelligence every day, though reaping the rewards can be a tricky game. After all, it is more often than not that the practical, not the sexy, AI application boosts a company's bottom-line. Balancing the influx of new information, from apocalyptic musings on the future of the technology to highly technical research reports -- incomprehensible to this writer and many of the most eager computer scientists -- is a challenge, but that's a problem Twitter can actually help with. Position: Co-founder at Coursera; adjunct professor at Stanford University; board of directors at drive.ai; chairman of the board at Woebot Ng is a staple in the AI sphere and helped build up some of the most prominent AI insititutions today, such as Google Brain and Baidu. He is also a driving force in AI education, and through his online learning program Coursera, Ng is trying to teach millions of new AI experts.

It has brought a paradigm shift to artificial intelligence and many cross-disciplinary areas. Despite its great success in applications, the theoretical explanation and mathematical framework are still open problems for DNN. There have been substantial efforts on the mathematical analysis for DNN, such as using the wavelet framework to understand DNN [12], applying the framework of function approximation [10], enumerating the linear regions after the nonlinear mapping of DNN [14] and analyzing the shape deformation in the transient chaos of DNN [15]. Although these excellent studies have made significant progress on a deeper understanding of DNN, they are still insufficient to fully describe the behavior, quantitatively analyze the performance and thus systematically design DNNs. In this paper, we propose a systematic framework to analyze DNN by considering it as a dynamical system.

Advances in deep generative networks have led to impressive results in recent years. Nevertheless, such models can often waste their capacity on the minutiae of datasets, presumably due to weak inductive biases in their decoders. This is where graphics engines may come in handy since they abstract away low-level details and represent images as high-level programs. Current methods that combine deep learning and renderers are limited by hand-crafted likelihood or distance functions, a need for large amounts of supervision, or difficulties in scaling their inference algorithms to richer datasets. To mitigate these issues, we present SPIRAL, an adversarially trained agent that generates a program which is executed by a graphics engine to interpret and sample images. The goal of this agent is to fool a discriminator network that distinguishes between real and rendered data, trained with a distributed reinforcement learning setup without any supervision. A surprising finding is that using the discriminator's output as a reward signal is the key to allow the agent to make meaningful progress at matching the desired output rendering. To the best of our knowledge, this is the first demonstration of an end-to-end, unsupervised and adversarial inverse graphics agent on challenging real world (MNIST, Omniglot, CelebA) and synthetic 3D datasets.

This paper proposes a method for generating speech from filterbank mel frequency cepstral coefficients (MFCC), which are widely used in speech applications, such as ASR, but are generally considered unusable for speech synthesis. First, we predict fundamental frequency and voicing information from MFCCs with an autoregressive recurrent neural net. Second, the spectral envelope information contained in MFCCs is converted to all-pole filters, and a pitch-synchronous excitation model matched to these filters is trained. Finally, we introduce a generative adversarial network -based noise model to add a realistic high-frequency stochastic component to the modeled excitation signal. The results show that high quality speech reconstruction can be obtained, given only MFCC information at test time.

Neurons in the visual cortex are correlated in their variability. The presence of correlation impacts cortical processing because noise cannot be averaged out over many neurons. In an effort to understand the functional purpose of correlated variability, we implement and evaluate correlated noise models in deep convolutional neural networks. Inspired by the cortex, correlation is defined as a function of the distance between neurons and their selectivity. We show how to sample from high-dimensional correlated distributions while keeping the procedure differentiable, so that back-propagation can proceed as usual. The impact of correlated variability is evaluated on the classification of occluded and non-occluded images with and without the presence of other regularization techniques, such as dropout. More work is needed to understand the effects of correlations in various conditions, however in 10/12 of the cases we studied, the best performance on occluded images was obtained from a model with correlated noise.

Variational auto-encoders (VAEs) are a popular and powerful deep generative model. Previous works on VAEs have assumed a factorised likelihood model, whereby the output uncertainty of each pixel is assumed to be independent. This approximation is clearly limited as demonstrated by observing a residual image from a VAE reconstruction, which often possess a high level of structure. This paper demonstrates a novel scheme to incorporate a structured Gaussian likelihood prediction network within the VAE that allows the residual correlations to be modelled. Our novel architecture, with minimal increase in complexity, incorporates the covariance matrix prediction within the VAE. We also propose a new mechanism for allowing structured uncertainty on color images. Furthermore, we provide a scheme for effectively training this model, and include some suggestions for improving performance in terms of efficiency or modelling longer range correlations. The advantage of our approach is illustrated on the CelebA face data and the LSUN outdoor churches dataset, with substantial improvements in terms of samples over traditional VAE and better reconstructions.

Can an algorithm create original and compelling fashion designs to serve as an inspirational assistant? To help answer this question, we design and investigate different image generation models associated with different loss functions to boost creativity in fashion generation. The dimensions of our explorations include: (i) different Generative Adversarial Networks architectures that start from noise vectors to generate fashion items, (ii) a new loss function that encourages creativity, and (iii) a generation process following the key elements of fashion design (disentangling shape and texture makers). A key challenge of this study is the evaluation of generated designs and the retrieval of best ones, hence we put together an evaluation protocol associating automatic metrics and human experimental studies that we hope will help ease future research. We show that our proposed creativity loss yields better overall appreciation than the one employed in Creative Adversarial Networks. In the end, about 61% of our images are thought to be created by human designers rather than by a computer while also being considered original per our human subject experiments, and our proposed loss scores the highest compared to existing losses in both novelty and likability.

The Variational Auto-Encoder (VAE) is one of the most used unsupervised machine learning models. But although the default choice of a Gaussian distribution for both the prior and posterior represents a mathematically convenient distribution often leading to competitive results, we show that this parameterization fails to model data with a latent hyperspherical structure. To address this issue we propose using a von Mises-Fisher (vMF) distribution instead, leading to a hyperspherical latent space. Through a series of experiments we show how such a hyperspherical VAE, or $\mathcal{S}$-VAE, is more suitable for capturing data with a hyperspherical latent structure, while outperforming a normal, $\mathcal{N}$-VAE, in low dimensions on other data types.