Asynchronous distributed stochastic gradient descent methods have trouble converging because of stale gradients. A gradient update sent to a parameter server by a client is stale if the parameters used to calculate that gradient have since been updated on the server. Approaches have been proposed to circumvent this problem that quantify staleness in terms of the number of elapsed updates. In th...

Updating the materialized views stored in data warehouses usually implies making the warehouse unavailable to users. We propose MAUVE , a new algorithm for online incremental view updates that uses timestamps and allows consistent read-only access to the warehouse while it being updated. The algorithm propagates the updates to the views more often than the typical once a day in order to reduce ...

Distributed Stochastic Gradient Descent (SGD) when run in a synchronous manner, suffers from delays in waiting for the slowest learners (stragglers). Asynchronous methods can alleviate stragglers, but cause gradient staleness that can adversely affect convergence. In this work we present the first theoretical characterization of the speed-up offered by asynchronous methods by analyzing the trad...

This thesis considers a novel problem of obtaining global channel state information (CSI) at every node in an ad-hoc wireless network. A class of protocols for dissemination and estimation are developed which attempt to minimize the staleness of the estimates throughout the network. This thesis also provides an optimal protocol for CSI dissemination in networks with complete graph topology and ...

of prior cloud work: The PIs have built Depot, a cloud storage system that minimizes trust assumptions. Depot tolerates buggy or malicious behavior by any number of clients or servers, yet it provides safety and liveness guarantees to correct clients. Depot provides these guarantees using a two-layer architecture. First, Depot ensures that the updates observed by correct nodes are consistently ...

Single-writer k-quorum protocols achieve high availability without incurring the risk of read operations returning arbitrarily stale values: in particular, they guarantee that, even in the presence of an adversarial scheduler, any read operation will return the value written by one of the last k writes. In this paper, we expand our understanding of k-quorums in two directions: first, we present...

UpStream is a framework for load management over data streams with update semantics. It provides a novel storage manager architecture that can be plugged into data stream processing engines for serving streaming applications that require low-staleness results over real-time continuous queries. We propose to demonstrate the key aspects of the UpStream architecture as well as its performance usin...

The paper describes the design, implementation, and evaluation of Depot, a cloud storage system that minimizes trust assumptions. Depot tolerates buggy or malicious behavior by any number of clients or servers, yet it provides safety and liveness guarantees to correct clients. Depot provides these guarantees using a two-layer architecture. First, Depot ensures that the updates observed by corre...

It is widely agreed that overtraining should be employed in order to achieve peak performance but it is also recognised that overtraining can actually produce decrements in performance. The challenge appears to be one of monitoring stress indicators in the athlete in order to titrate the training stimulus and prevent the onset of staleness. The present paper summarises a ten-year research effor...