Search

International audience; In the context of post-hoc interpretability, this paper addresses the task of explaining the prediction of a classifier, considering the case where no information is available, neither on the classifier itself , nor on the processed data (neither the training nor the test data). It proposes an inverse classification approach whose principle consists in determining the minimal changes needed to alter a prediction: in an instance-based framework, given a data point whose classification must be explained, the proposed method consists in identifying a close neighbor classified differently, where the closeness definition integrates a spar-sity constraint. This principle is implemented using observation generation in the Growing Spheres algorithm. Experimental results on two datasets illustrate the relevance of the proposed approach that can be used to gain knowledge about the classifier.

The nuclear and missile programs of the Democratic People’s Republic of Korea (DPRK) have attracted considerable attention on the part of the international community in recent years. This is not only due to the issue of nonproliferation, as it also concerns East Asian regional security and the international relationships among the DPRK, the United States (US), China, the Republic of Korea (ROK), and others. As a close neighbor, China has its own interests and concerns vis-a-vis the DPRK and has played its unique role in handling the DPRK’s nuclear and missile programs.

In this paper, we examine the factors that contribute to the variability in download time of a self-organizing P2P file distribution application such as BitTorrent. We conducted a series of side-by-side live experiments, involving two clients running on the same machine downloading the same file at the same time. We found that the download latency varied significantly, sometimes by a factor of 2. Surprisingly, the main contributing factor isn't the network bandwidths of the set of neighbors that a client is given. Rather, it has to do with the frequency of turn-overs in "close" neighbors, i.e. those that are in a stable data-exchange relationship with the client. Analysis of the log data shows that a client obtains over 90% of the file from a small set of close neighbors, and if a close neighbor leaves the network, it takes the client a long time, over half an hour , to find another one. This suggests that self-organizing P2P file distributions indeed need external help in order to provide QoS guarantees, but such guarantees are achievable with proper enhancements to the P2P network.

It is shown that 5n/4 plus-minus lower order terms comparisons on average are necessary and sufficient to solve the problem of finding the values of ranks immediately above and below a specified element x in a set X of size n>1. When x turns out to be the median of X, 1.5n+√πn/8+O(lg n) comparisons are proven to be sufficient. n+min(k, n−k)+3 ln n+O(1) comparisons are sufficient if k, the rank of x in X, differs from n/2 by Θ(n).