One way to model this is by requiring that for each cluster, the contained samples should be close to their mean.Īccording to this model, a good clustering is one in which the sum of the squared distances of the points in each cluster to their corresponding cluster center is small. Intuitively, the clustering problem can be described as finding groups of points which are similar to each other but different from the members of other groups. means clusteringīefore we dive into the details of our optimized algorithms, let’s go one step back and briefly review standard -means clustering.
The following results were developed at Avira in collaboration with University of Ulm and were recently presented at ICML. Our goal is to decrease the computational time while guaranteeing the same results as the standard -means algorithm. The main idea will be to come up with a way to accelerate a computationally expensive aspect of the -means algorithm involving the repeated computation of Euclidean distances to cluster centers. This is the case at Avira, where the data consists of several thousand features extracted for our samples of malicious files. We are especially interested in the case where one is dealing with a high amount of high-dimensional sparse data and the goal is to find a large number of clusters.
#Avira download 2016 how to#
In this post we will talk about how to speed-up the popular -means clustering algorithm. However, in our daily work we often face the situation that standard techniques are not suitable to handle the sheer amount of data we are dealing with.įor this reason one has to come up with ways to compute the solutions of these algorithms more efficiently.
Thus, it is of crucial importance that this task can be done as fast as possible. The Avira Protection Labs maintain databases containing several hundred millions of malware samples which are used to provide up-to-date protection to our customers.īeing able to automatically cluster these huge amounts of data into meaningful groups is an essential task both for data analysis and as a preprocessing step for our machine learning engines.
0 kommentar(er)
