No abstract

No abstract

Federated Learning enables mobile devices to collaboratively learn a shared inference model while keeping all the training data on a user's device, decoupling the ability to do machine learning from the need to store the data in the cloud. Existing work on federated learning with limited communication demonstrates how random rotation can enable users' model updates to be quantized much more efficiently, reducing the communication cost between users and the server. Meanwhile, secure aggregation enables the server to learn an aggregate of at least a threshold number of device's model contributions without observing any individual device's contribution in unaggregated form. In this paper, we highlight some of the challenges of setting the parameters for secure aggregation to achieve communication efficiency, especially in the context of the aggressively quantized inputs enabled by random rotation. We then develop a recipe for auto-tuning communication-efficient secure aggregation, based on specific properties of random rotation and secure aggregation -namely, the predictable distribution of vector entries postrotation and the modular wrapping inherent in secure aggregation. We present both theoretical results and initial experiments.

No abstract

Learning of new words is assisted by contextual information. This context can come in several forms, including observations in nonlinguistic semantic domains, as well as the linguistic context in which the new word was presented. We outline a general architecture for word learning, in which structural alignment coordinates this contextual information in order to restrict the possible interpretations of unknown words. We identify spatial relations as an applicable semantic domain, and describe a system-in-progress for implementing the general architecture using video sequences as our non-linguistic input. For example, when the complete system is presented with "The bird dove to the rock," with a video sequence of a bird flying from a tree to a rock, and with the meanings for all the words except the preposition "to," the system will register the unknown "to" with the corresponding aspect of the bird's trajectory.