Generative Adversarial Networks (GANs) are classified within the group of generative models. That means that they're able to produce, i.e generate completely new, "valid" data. By valid data, we mean that the network's output should be something that we would deem acceptable for our target. To illustrate, consider an example where we wish to generate some new images for training an image classification network. Of course, for such an application we want our training data to be as realistic as possible, perhaps quite similar in style to other image classification training data sets.

Every month we'll share updates from the School of AI HQ and have some topics and/or demonstrations about AI, Machine Learning, Deep Learning, and other topics in AI. This month we have a guest speaker from IBM to talk about PowerAI Vision and other AI services and tools IBM has to offer. If you'd like to present at a meeting, please reach out to us through the Meetup messages or @cincyai on twitter and we'll get you scheduled.

Uber's services require real-world coordination between a wide range of customers, including driver-partners, riders, restaurants, and eaters. Accurately forecasting things like rider demand and ETAs enables this coordination, which makes our services work as seamlessly as possible. In an effort to constantly optimize our operations, serve our customers, and train our systems to perform better and better, we leverage machine learning (ML). In addition, we make many of our ML tools open source, sharing them with the community to advance the state of the art. In this spirit, members of our Seattle Engineering team shared their work at an April 2019 meetup on ML and AI at Uber.

Deep learning is an emerging field currently gaining significant attention in the security industry, especially in the surveillance market. By using powerful algorithms, devices are now capable of understanding and adapting to their environment by storing and analyzing large data sets, all without human intervention. Considered the next Industrial Revolution, the deep learning of artificial intelligence is making technology more efficient than ever thought possible. This intelligence is a key component to the development of sophisticated, data-rich information that enables surveillance systems to search through hours or days of video with a single click to locate an individual person or vehicle across a single site or multiple locations. Deep learning provides the necessary building blocks to clearly identify a person and know if that person is walking, running or crawling -- distinctions that video analytics was not previously capable of doing.

In the past few weeks, two important developments in artificial intelligence research have gone largely unheralded. Both hint at just how earth shaking – or at least industry-shattering – A.I.'s potential really is. The first item was news that a Hong Kong-based biotechnology startup, InSilico Medicine, working with researchers from the University of Toronto, had used machine learning to create a potential new drug to prevent tissue scarring. What's eye-popping here is the timescale: just 46 days from molecular design to animal testing in mice. Considering that, on average, it takes more than a decade and costs $350 million to $2.7 billion to bring a new drug to market, depending on which study one believes, the potential impact on the pharmaceutical industry is huge.

At the recent ApacheCon North America, Denis Magda spoke on continuous machine learning with Apache Ignite, an in-memory data grid. Ignite simplifies the machine-learning pipeline by performing training and hosting models in the same cluster that stores the data, and can perform "online" training to incrementally improve models when new data is available. Magda, vice-president of product management at GridGain, began by describing some of the pain points of machine learning on large datasets, in particular the latency involved in moving data across the network from its storage location to the processors that perform training. Models also have to be deployed into a production system after they are trained, and retrained periodically after new data is collected. Because Ignite runs code on the same computers that host data, it can train, deploy, and update a machine-learning model without a time-consuming extract-transform-load (ETL) step.

Apple is seeking highly skilled and creative students passionate about advancing speech and audio technologies on the audio algorithms team. As a member of the team, you will research multi-modal deep learning approaches to demonstrate video, audio, and other signals for improving speech and audio acquisition. You will work alongside a group of researchers developing machine learning and multichannel signal processing technologies for speech improvement including source separation, adaptive beamforming, dereverberation, and source localization. You'll collaborate with multiple teams across product development, software, and Siri to integrate your ideas into products and build future technologies that will impact millions of Apple customers.

Apple is seeking highly skilled and creative students passionate about advancing speech and audio technologies on the audio algorithms team. As a member of the team, you will research multi-modal deep learning approaches to demonstrate video, audio, and other signals for improving speech and audio acquisition. You will work alongside a group of researchers developing machine learning and multichannel signal processing technologies for speech improvement including source separation, adaptive beamforming, dereverberation, and source localization. You'll collaborate with multiple teams across product development, software, and Siri to integrate your ideas into products and build future technologies that will impact millions of Apple customers.

This paper presents an efficient neural network model to generate robotic grasps with high resolution images. The proposed model uses fully convolution neural network to generate robotic grasps for each pixel using 400 $\times$ 400 high resolution RGB-D images. It first down-sample the images to get features and then up-sample those features to the original size of the input as well as combines local and global features from different feature maps. Compared to other regression or classification methods for detecting robotic grasps, our method looks more like the segmentation methods which solves the problem through pixel-wise ways. We use Cornell Grasp Dataset to train and evaluate the model and get high accuracy about 94.42% for image-wise and 91.02% for object-wise and fast prediction time about 8ms. We also demonstrate that without training on the multiple objects dataset, our model can directly output robotic grasps candidates for different objects because of the pixel wise implementation.

This paper takes a step towards theoretical analysis of the relationship between word embeddings and context embeddings in models such as word2vec. We start from basic probabilistic assumptions on the nature of word vectors, context vectors, and text generation. These assumptions are supported either empirically or theoretically by the existing literature. Next, we show that under these assumptions the widely-used word-word PMI matrix is approximately a random symmetric Gaussian ensemble. This, in turn, implies that context vectors are reflections of word vectors in approximately half the dimensions. As a direct application of our result, we suggest a theoretically grounded way of tying weights in the SGNS model.