arXiv subjects

Zohar Yakhini

Publications and source records attributed to Zohar Yakhini.

23 records · Page 2Linked to original sources

arXiv2020

Autoencoder Image Interpolation by Shaping the Latent Space

Autoencoders represent an effective approach for computing the underlying factors characterizing datasets of different types. The latent representation of autoencoders have been studied in the context of enabling interpolation between data points by decoding convex combinations of latent vectors. This interpolation, however, often leads to artifacts or produces unrealistic results during reconstruction. We argue that these incongruities are due to the structure of the latent space and because such naively interpolated latent vectors deviate from the data manifold. In this paper, we propose a regularization technique that shapes the latent representation to follow a manifold that is consistent with the training images and that drives the manifold to be smooth and locally convex. This regularization not only enables faithful interpolation between data points, as we show herein, but can also be used as a general regularization technique to avoid overfitting or to produce new samples for data augmentation.

cs.LG↗

arXiv2019

IoT or NoT: Identifying IoT Devices in a ShortTime Scale

In recent years the number of IoT devices in home networks has increased dramatically. Whenever a new device connects to the network, it must be quickly managed and secured using the relevant security mechanism or QoS policy. Thus a key challenge is to distinguish between IoT and NoT devices in a matter of minutes. Unfortunately, there is no clear indication of whether a device in a network is an IoT. In this paper, we propose different classifiers that identify a device as IoT or non-IoT, in a short time scale, and with high accuracy. Our classifiers were constructed using machine learning techniques on a seen (training) dataset and were tested on an unseen (test) dataset. They successfully classified devices that were not in the seen dataset with accuracy above 95%. The first classifier is a logistic regression classifier based on traffic features. The second classifier is based on features we retrieve from DHCP packets. Finally, we present a unified classifier that leverages the advantages of the other two classifiers. We focus on the home-network environment, but our classifiers are also applicable to enterprise networks.

cs.NI↗

arXiv2013

Mutual Enrichment in Ranked Lists and the Statistical Assessment of Position Weight Matrix Motifs

Statistics in ranked lists is important in analyzing molecular biology measurement data, such as ChIP-seq, which yields ranked lists of genomic sequences. State of the art methods study fixed motifs in ranked lists. More flexible models such as position weight matrix (PWM) motifs are not addressed in this context. To assess the enrichment of a PWM motif in a ranked list we use a PWM induced second ranking on the same set of elements. Possible orders of one ranked list relative to the other are modeled by permutations. Due to sample space complexity, it is difficult to characterize tail distributions in the group of permutations. In this paper we develop tight upper bounds on tail distributions of the size of the intersection of the top of two uniformly and independently drawn permutations and demonstrate advantages of this approach using our software implementation, mmHG-Finder, to study PWMs in several datasets.

q-bio.QM↗

arXiv2013

On the Sample Complexity of Learning Bayesian Networks

In recent years there has been an increasing interest in learning Bayesian networks from data. One of the most effective methods for learning such networks is based on the minimum description length (MDL) principle. Previous work has shown that this learning procedure is asymptotically successful: with probability one, it will converge to the target distribution, given a sufficient number of samples. However, the rate of this convergence has been hitherto unknown. In this work we examine the sample complexity of MDL based learning procedures for Bayesian networks. We show that the number of samples needed to learn an epsilon-close approximation (in terms of entropy distance) with confidence delta is O((1/epsilon)^(4/3)log(1/epsilon)log(1/delta)loglog (1/delta)). This means that the sample complexity is a low-order polynomial in the error threshold and sub-linear in the confidence bound. We also discuss how the constants in this term depend on the complexity of the target distribution. Finally, we address questions of asymptotic minimality and propose a method for using the sample complexity results to speed up the learning process.

cs.LG↗

arXiv2012

A Distance-Based Branch and Bound Feature Selection Algorithm

There is no known efficient method for selecting k Gaussian features from n which achieve the lowest Bayesian classification error. We show an example of how greedy algorithms faced with this task are led to give results that are not optimal. This motivates us to propose a more robust approach. We present a Branch and Bound algorithm for finding a subset of k independent Gaussian features which minimizes the naive Bayesian classification error. Our algorithm uses additive monotonic distance measures to produce bounds for the Bayesian classification error in order to exclude many feature subsets from evaluation, while still returning an optimal solution. We test our method on synthetic data as well as data obtained from gene expression profiling.

cs.LG↗