Results 1 
2 of
2
Fast variational inference for largescale internet diagnosis
 ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 20
"... Web servers on the Internet need to maintain high reliability, but the cause of intermittent failures of web transactions is nonobvious. We use approximate Bayesian inference to diagnose problems with web services. This diagnosis problem is far larger than any previously attempted: it requires infe ..."
Abstract

Cited by 2 (0 self)
 Add to MetaCart
Web servers on the Internet need to maintain high reliability, but the cause of intermittent failures of web transactions is nonobvious. We use approximate Bayesian inference to diagnose problems with web services. This diagnosis problem is far larger than any previously attempted: it requires inference of 10 4 possible faults from 10 5 observations. Further, such inference must be performed in less than a second. Inference can be done at this speed by combining a meanfield variational approximation and the use of stochastic gradient descent to optimize a variational cost function. We use this fast inference to diagnose a time series of anomalous HTTP requests taken from a real web service. The inference is fast enough to analyze network logs with billions of entries in a matter of hours.
Observation Biases in Diagnostic Inference
, 2001
"... In realworld statistical inference problems, such as the diagnosis of diseases from the results of medical tests and procedures, there are systematic observation biases as to which test results are available. Ignoring these biases leads to catastrophically poor inference. We have previously sho ..."
Abstract
 Add to MetaCart
In realworld statistical inference problems, such as the diagnosis of diseases from the results of medical tests and procedures, there are systematic observation biases as to which test results are available. Ignoring these biases leads to catastrophically poor inference. We have previously shown that pretrained recognition networks are one of the statistically and computationally best ways of performing inference in the largest scale statistical medical diagnosis belief network (QMRDT). In this paper, we show how to use a gating process to integrate the outputs of a library of recognition networks, each trained on a di#erent observation bias. The resulting network performs accurate inference even for previously unseen biases.