From sekhar.tatikonda at yale.edu Tue Apr 2 09:07:08 2019 From: sekhar.tatikonda at yale.edu (Sekhar Tatikonda) Date: Tue, 2 Apr 2019 09:07:08 -0400 Subject: [YPNG] YPNG Friday 5 April 2019 In-Reply-To: References: Message-ID: Hi Folks, Dominic Richards (visiting from Oxford) will talk this Friday about: Title: Optimal Statistical Rates for Non-parametric Decentralised Regression with Distributed Gradient Descent Abstract: Due to bandwidth limitations, privacy concerns or network instability, it is often required to fit statistical models on data sets stored across multiple computers without a central server to coordinate computation and disseminate information i.e. star topology. This has motivated the investigation of decentralised methods which solve the problem in a more robust manner by not relying on a single computer. In this work we investigate the statistical performance of a simple synchronous decentralised iterative gradient descent method (Distributed Gradient Descent) in the homogeneous distributed non-parametric regression setting i.e. computers hold samples from the same distribution. By utilising the concentration of quantities held by individual computers, we show there are a number of settings where computers can save on computational and communication costs without any loss in statistical accuracy. Given computers hold sufficiently many samples with respect to the network topology, we show that Distributed Gradient Descent yields optimal statistical rates with the same numbers of iterations as Centralised algorithm. (Joint work with P. Rebeschini) See you Friday at 11am in the Stat's classroom. Regards, sekhar -------------- next part -------------- An HTML attachment was scrubbed... URL: From sekhar.tatikonda at yale.edu Mon Apr 8 18:28:53 2019 From: sekhar.tatikonda at yale.edu (Sekhar Tatikonda) Date: Mon, 8 Apr 2019 18:28:53 -0400 Subject: [YPNG] YPNG Friday 12 April 2019 In-Reply-To: References: Message-ID: Hi Everyone, John Hartigan will talk this Friday about: The mean shift method leads to clusters in which the mean of the data within a sphere lies at the center of the sphere. The DP ( please suggest acronym) test for the existence of a population mode within the sphere is based on the variance within the sphere being small enough. The null hypothesis is taken from the spherical tree distribution -- a mixture of uniform distributions over spheres which have the tree property that each pair of spheres are disjoint, or one includes the other. The test is applied to the distribution of orbital radius and inclination of 500000 asteroids of greater than 1km diameter -- some of which could crash into the earth at any moment, so you had better come and check it out before they do! A version of the test applies to k-means clusters also. See you Friday at 11am in the Stat's classroom (unless the asteroids see us first.) Regards, sekhar -------------- next part -------------- An HTML attachment was scrubbed... URL: