In this paper, we introduce Recipe1M+, a new large-scale, structured corpus of over one million cooking recipes and 13 million food images. As the largest publicly available collection of recipe data, Recipe1M+ affords the ability to train high-capacity models on aligned, multi-modal data. Using these data, we train a neural network to learn a joint embedding of recipes and images that yields impressive results on an image-recipe retrieval task. Moreover, we demonstrate that regularization via the addition of a high-level classification objective both improves retrieval performance to rival that of humans and enables semantic vector arithmetic. We postulate that these embeddings will provide a basis for further exploration of the Recipe1M+ dataset and food and cooking in general. Code, data and models are publicly available.

We are interested in modeling the total expected data collection time for a set of cars (or other mobile nodes) to aggregate the information that they are each carrying to a single vehicle. Data collection processes between cars -- or other mobile nodes -- can be modeled as random walks of particles on a graph, and in this model, data is transferred between data-possessing cars once they encounter each other on the same node of the graph. We analyze this process in the special case in which this graph is a square grid torus graph. Recursion occurs in the system because once an encounter between multiple cars that possess data occurs, the system becomes analogous to simpler systems with the same graph and with fewer cars that possess data. We make approximations for the total data collection time using this recursive property.