Bayesian dynamic modelling of network flows

Duration: 36 mins 13 secs

Share this media item:

Embed this media item:

<iframe width="_width_" height="_height_" src="https://sms.cam.ac.uk/media/2289936/embed" frameborder="0" scrolling="no" allowfullscreen></iframe>

Choose size:

About this item

Available Formats

About this item

Description:	West, M (Duke University) Wednesday 27th July 2016 - 10:00 to 10:30


Created:	2016-07-28 15:23
Collection:	Theoretical Foundations for Statistical Network Analysis
Publisher:	Isaac Newton Institute
Copyright:	West, M
Language:	eng (English)


Abstract:	I discuss Bayesian dynamic modelling for sequential analysis of network flow count data, linking two classes of models which allow fast, scalable and interpretable Bayesian inference. The first class involves sets of "decoupled" univariate state-space models for streaming count data, able to adaptively characterize and quantify network dynamics in real-time. These are then "recoupled" to define "emulation" of a second class of more structured, time-varying gravity models that allow closer and formal dissection of network dynamics and interactions among network nodes. Evolving internet flows on a defined network of web domains in e-commerce applications provide context, data and examples. Bayesian model monitoring theory defines a strategy for sequential model assessment and adaptation in cases of signaled departures of network flow data from model-based predictions. This work builds on the more general concepts and strategy of "decouple/recouple" for Bayesian model emulation. That is, we use decoupled, parallel and scalable analyses of a set of simpler and computationally efficient univariate models, then recouple- on a sound theoretical basis- to rebuild the larger multivariate process for more formal inferences. Co-authors: Xi Chen (Duke University), Kaoru Irie (University of Tokyo), David Banks (Duke University), Jewell Thomas (MaxPoint Interactive Inc.), and Rob Haslinger (The Sync Project).

Abstract:

I discuss Bayesian dynamic modelling for sequential analysis of network flow count data, linking two classes of models which allow fast, scalable and interpretable Bayesian inference. The first class involves sets of "decoupled" univariate state-space models for streaming count data, able to adaptively characterize and quantify network dynamics in real-time. These are then "recoupled" to define "emulation" of a second class of more structured, time-varying gravity models that allow closer and formal dissection of network dynamics and interactions among network nodes. Evolving internet flows on a defined network of web domains in e-commerce applications provide context, data and examples. Bayesian model monitoring theory defines a strategy for sequential model assessment and adaptation in cases of signaled departures of network flow data from model-based predictions.

This work builds on the more general concepts and strategy of "decouple/recouple" for Bayesian model emulation. That is, we use decoupled, parallel and scalable analyses of a set of simpler and computationally efficient univariate models, then recouple- on a sound theoretical basis- to rebuild the larger multivariate process for more formal inferences.

Co-authors: Xi Chen (Duke University), Kaoru Irie (University of Tokyo), David Banks (Duke University), Jewell Thomas (MaxPoint Interactive Inc.), and Rob Haslinger (The Sync Project).

Available Formats

Format	Quality	Bitrate	Size
MPEG-4 Video	640x360	1.94 Mbits/sec	527.15 MB	View	Download
WebM	640x360	448.47 kbits/sec	119.02 MB	View	Download
iPod Video	480x270	522.06 kbits/sec	138.48 MB	View	Download
MP3	44100 Hz	249.8 kbits/sec	66.32 MB	Listen	Download
Auto *	(Allows browser to choose a format it supports)