Abstract: Cold-start is a notoriously difficult problem which
can occur in recommendation systems, and arises when there is
insufficient information to draw inferences for users or items. To
address this challenge, a contextual bandit algorithm – the Fast
Approximate Bayesian Contextual Cold Start Learning algorithm
(FAB-COST) – is proposed, which is designed to provide improved
accuracy compared to the traditionally used Laplace approximation
in the logistic contextual bandit, while controlling both algorithmic
complexity and computational cost. To this end, FAB-COST uses
a combination of two moment projection variational methods:
Expectation Propagation (EP), which performs well at the cold
start, but becomes slow as the amount of data increases; and
Assumed Density Filtering (ADF), which has slower growth of
computational cost with data size but requires more data to obtain an
acceptable level of accuracy. By switching from EP to ADF when
the dataset becomes large, it is able to exploit their complementary
strengths. The empirical justification for FAB-COST is presented, and
systematically compared to other approaches on simulated data. In a
benchmark against the Laplace approximation on real data consisting
of over 670, 000 impressions from autotrader.co.uk, FAB-COST
demonstrates at one point increase of over 16% in user clicks. On
the basis of these results, it is argued that FAB-COST is likely to
be an attractive approach to cold-start recommendation systems in a
variety of contexts.
Abstract: In this paper, we propose the variational EM inference
algorithm for the multi-class Gaussian process classification model
that can be used in the field of human behavior recognition. This
algorithm can drive simultaneously both a posterior distribution of a
latent function and estimators of hyper-parameters in a Gaussian
process classification model with multiclass. Our algorithm is based
on the Laplace approximation (LA) technique and variational EM
framework. This is performed in two steps: called expectation and
maximization steps. First, in the expectation step, using the Bayesian
formula and LA technique, we derive approximately the posterior
distribution of the latent function indicating the possibility that each
observation belongs to a certain class in the Gaussian process
classification model. Second, in the maximization step, using a derived
posterior distribution of latent function, we compute the maximum
likelihood estimator for hyper-parameters of a covariance matrix
necessary to define prior distribution for latent function. These two
steps iteratively repeat until a convergence condition satisfies.
Moreover, we apply the proposed algorithm with human action
classification problem using a public database, namely, the KTH
human action data set. Experimental results reveal that the proposed
algorithm shows good performance on this data set.