Label propagation is a powerful and flexible semi-supervised learning technique on graphs. Neural networks, on the other hand, have proven track records in many supervised learning tasks. In this work, we propose a training framework with a graph-regularised objective, namely Neural Graph Machines, that can combine the power of neural networks and label propagation. This work generalises previous literature on graph-augmented training of neural networks, enabling it to be applied to multiple neural architectures (Feed-forward NNs, CNNs and LSTM RNNs) and a wide range of graphs. The new objective allows the neural networks to harness both labeled and unlabeled data by: (a)~allowing the network to train using labeled data as in the supervised setting, (b)~biasing the network to learn similar hidden representations for neighboring nodes on a graph, in the same vein as label propagation. Such architectures with the proposed objective can be trained efficiently using stochastic gradient descent and scaled to large graphs, with a runtime that is linear in the number of edges. The proposed joint training approach convincingly outperforms many existing methods on a wide range of tasks (multi-label classification on social graphs, news categorization, document classification and semantic intent classification), with multiple forms of graph inputs (including graphs with and without node-level features) and using different types of neural networks.

We introduce the problem of automatically suggesting conversational responses to photos and present an intelligent assistant called PhotoReply that solves the problem in the context of a messaging application. For example, when a user receives a photo showing a dog, PhotoReply suggests responses such as "Aww»» and "Cute terrier!»» This simplifies composing responses on constrained mobile keyboards, and delights users with uncanny insights into the photos they receive. PhotoReply is an integral part of the Allo chat application, being its predictive assistance feature with highest click-through rate. We formalize the problem of suggesting responses to images as an instance of multimodal learning which is akin to caption generation models, a topic that has recently received significant attention \citeKarpathy2015,Vinyals2015,Xu2015. We then present a system that "translates»» image pixels to text responses. The system includes a conditioned language model, based on an LSTM, which, given an embedding of image pixels and the previous predicted words, calculates the probability of all words in a vocabulary of being the next word in the generated response and a triggering model trained with image embeddings and concept labels from a large concept taxonomy. We describe training of the models and a thorough experimental evaluation based on crowdsourced datasets and live traffic.

In this paper we propose and investigate a novel end-to-end method for automatically generating short email responses, called Smart Reply. It generates semantically diverse suggestions that can be used as complete email responses with just one tap on mobile. The system is currently used in Inbox by Gmail and is responsible for assisting with 10% of all mobile responses. It is designed to work at very high throughput and process hundreds of millions of messages daily. The system exploits state-of-the-art, large-scale deep learning.