IEEE 2019 Accepted for the 2019 International Joint Conference on Neural Networks Not All Adversarial Examples Require a Complex Defense: Identifying Over-optimized Adversarial Examples with IQR-based Logit Thresholding Utku Ozbulak 1, 3 Arnout V an Messem 2, 3 Wesley De Neve 1, 3 1 Department of Electronics and Information Systems, Ghent University, Belgium 2 Department of Applied Mathematics, Computer Science and Statistics, Ghent University, Belgium 3 Center for Biotech Data Science, Ghent University Global Campus, Republic of Korea {utku.ozbulak,arnout.vanmessem,wesley.deneve} Abstract --Detecting adversarial examples currently stands as one of the biggest challenges in the field of deep learning. Adversarial attacks, which produce adversarial examples, increase the prediction likelihood of a target class for a particular data point. During this process, the adversarial example can be further optimized, even when it has already been wrongly classified with 100% confidence, thus making the adversarial example even more difficult to detect. For this kind of adversarial examples, which we refer to as over-optimized adversarial examples, we discovered that the logits of the model provide solid clues on whether the data point at hand is adversarial or genuine. In this context, we first discuss the masking effect of the softmax function for the prediction made and explain why the logits of the model are more useful in detecting over-optimized adversarial examples. T o identify this type of adversarial examples in practice, we propose a nonparametric and computationally efficient method which relies on interquartile range, with this method becoming more effective as the image resolution increases. We support our observations throughout the paper with detailed experiments for different datasets (MNIST, CIF AR-10, and ImageNet) and several architectures.

The detection of interesting patterns in large high-dimensional datasets is difficult because of their dimensionality and pattern complexity. Therefore, analysts require automated support for the extraction of relevant patterns. In this paper, we present FDive, a visual active learning system that helps to create visually explorable relevance models, assisted by learning a pattern-based similarity. We use a small set of user-provided labels to rank similarity measures, consisting of feature descriptor and distance function combinations, by their ability to distinguish relevant from irrelevant data. Based on the best-ranked similarity measure, the system calculates an interactive Self-Organizing Map-based relevance model, which classifies data according to the cluster affiliation. It also automatically prompts further relevance feedback to improve its accuracy. Uncertain areas, especially near the decision boundaries, are highlighted and can be refined by the user. We evaluate our approach by comparison to state-of-the-art feature selection techniques and demonstrate the usefulness of our approach by a case study classifying electron microscopy images of brain cells. The results show that FDive enhances both the quality and understanding of relevance models and can thus lead to new insights for brain research.

In this work we present STEVE - Soccer TEam VEctors, a principled approach for learning real valued vectors for soccer teams where similar teams are close to each other in the resulting vector space. STEVE only relies on freely available information about the matches teams played in the past. These vectors can serve as input to various machine learning tasks. Evaluating on the task of team market value estimation, STEVE outperforms all its competitors. Moreover, we use STEVE for similarity search and to rank soccer teams.

In an image recognition application, the raw input may be a matrix of pixels; the first representational layer may abstract the pixels and encode edges; the second layer may compose and encode arrangements of edges; the third layer may encode a nose and eyes; and the fourth layer may recognize that the image contains a face. Importantly, a deep learning process can learn which features to optimally place in which level on its own.

In this post, I'll take a look at how you can compare regression models. Comparing regression models is perhaps one of the trickiest tasks to complete in the "comparing models" arena; The reason is that there are literally dozens of statistics you can calculate to compare regression models, including: This list isn't exhaustive--there are many other tools, tests and plots at your disposal. Rather than discuss the statistics in detail, I chose to focus this post on comparing a few of the most popular regression model evaluation techniques and discuss when you might want to use them (or when you might not want to). The techniques listed below tend to be on the "easier to use and understand" end of the spectrum, so if you're new to model comparison it's a good place to start. The first question you should be asking is: How well do I know my data?

Stochastic gradient descent (SGD) has been the dominant optimization method for training deep neural networks due to its many desirable properties. One of the more remarkable and least understood quality of SGD is that it generalizes relatively well on unseen data even when the neural network has millions of parameters. In this work, we show that SGD is an extreme case of deep gradient boosting (DGB) and as such is intrinsically regularized. The key idea of DGB is that back-propagated gradients calculated using the chain rule can be viewed as pseudo-residual targets. Thus at each layer the weight update is calculated by solving the corresponding gradient boosting problem. We hypothesize that some learning tasks can benefit from a more lax regularization requirement and this approach provides a way to control that. We tested this hypothesis on a number of benchmark data sets and show that indeed in a subset of cases DGB outperforms SGD and under-performs on tasks that are more prone to over-fitting, such as image recognition.

We propose a framework for interactive and explainable machine learning that enables users to (1) understand machine learning models; (2) diagnose model limitations using different explainable AI methods; as well as (3) refine and optimize the models. Our framework combines an iterative XAI pipeline with eight global monitoring and steering mechanisms, including quality monitoring, provenance tracking, model comparison, and trust building. To operationalize the framework, we present explAIner, a visual analytics system for interactive and explainable machine learning that instantiates all phases of the suggested pipeline within the commonly used TensorBoard environment. We performed a user-study with nine participants across different expertise levels to examine their perception of our workflow and to collect suggestions to fill the gap between our system and framework. The evaluation confirms that our tightly integrated system leads to an informed machine learning process while disclosing opportunities for further extensions.

Zeroth-order (gradient-free) method is a class of powerful optimization tool for many machine learning problems because it only needs function values (not gradient) in the optimization. In particular, zeroth-order method is very suitable for many complex problems such as black-box attacks and bandit feedback, whose explicit gradients are difficult or infeasible to obtain. Recently, although many zeroth-order methods have been developed, these approaches still exist two main drawbacks: 1) high function query complexity; 2) not being well suitable for solving the problems with complex penalties and constraints. To address these challenging drawbacks, in this paper, we propose a novel fast zeroth-order stochastic alternating direction method of multipliers (ADMM) method (\emph{i.e.}, ZO-SPIDER-ADMM) with lower function query complexity for solving nonconvex problems with multiple nonsmooth penalties. Moreover, we prove that our ZO-SPIDER-ADMM has the optimal function query complexity of $O(dn + dn^{\frac{1}{2}}\epsilon^{-1})$ for finding an $\epsilon$-approximate local solution, where $n$ and $d$ denote the sample size and dimension of data, respectively. In particular, the ZO-SPIDER-ADMM improves the existing best nonconvex zeroth-order ADMM methods by a factor of $O(d^{\frac{1}{3}}n^{\frac{1}{6}})$. Moreover, we propose a fast online ZO-SPIDER-ADMM (\emph{i.e.,} ZOO-SPIDER-ADMM). Our theoretical analysis shows that the ZOO-SPIDER-ADMM has the function query complexity of $O(d\epsilon^{-\frac{3}{2}})$, which improves the existing best result by a factor of $O(\epsilon^{-\frac{1}{2}})$. Finally, we utilize a task of structured adversarial attack on black-box deep neural networks to demonstrate the efficiency of our algorithms.

M. E. J. Newman, 1, 2 George T. Cantwell, 1 and Jean Gabriel Young 2 1 Department of Physics, University of Michigan, Ann Arbor, Michigan, USA 2 Center for the Study of Complex Systems, University of Michigan, Ann Arbor, Michigan, USA The information theoretic quantity known as mutual information finds wide use in classification and community detection analyses to compare two classifications of the same set of objects into groups. In the context of classification algorithms, for instance, it is often used to compare discovered classes to known ground truth and hence to quantify algorithm performance. Here we argue that the standard mutual information, as commonly defined, omits a crucial term which can become large under real-world conditions, producing results that can be substantially in error. We demonstrate how to correct this error and define a mutual information that works in all cases. We discuss practical implementation of the new measure and give some example applications. I. INTRODUCTION Mutual information is widely used in physics, statistics, and machine learning as a tool for comparing different labelings of a set of objects [1]. For instance, within physics it is used in statistical mechanics for comparing states of spin models [2] and particularly in network science for comparing partitions of networks into communities, mutual information being perhaps the standard measure for quantifying the performance of community detection algorithms [3, 4]: it tells us the extent to which the set of communities found by an algorithm agree with a given set of ground-truth communities. In machine learning and statistics, mutual information is similarly used in classification problems to quantify the similarity of different labelings of sets of objects [5]. For instance, we might attempt to deduce characteristics of a set of users of an online service, such as their age group or gender, and then calibrate our algorithm by using mutual information to compare our results against known characteristics of a test set of users. Imagine then that we have some set of individuals or objects, such as people, documents, email messages, or aerial photographs, among many other possibilities. Each object can be classified or labeled as belonging to one of several types, groups, or communities. People could be labeled by sex, race, or blood type for instance; documents by topic; aerial photographs by type of terrain, and so forth. Now imagine we have two different sets of labels for our objects, one inferred by some algorithm and the other assigned for instance by human experts. The mutual information of the two labelings represents the amount of information that the first labeling gives us about the second--in effect, how good the algorithm is at mimicking the human experts.

Pricing a rental property on Airbnb is a challenging task for the owner as it determines the number of customers for the place. On the other hand, customers have to evaluate an offered price with minimal knowledge of an optimal value for the property. This paper aims to develop a reliable price prediction model using machine learning, deep learning, and natural language processing techniques to aid both the property owners and the customers with price evaluation given minimal available information about the property. Features of the rentals, owner characteristics, and the customer reviews will comprise the predictors, and a range of methods from linear regression to tree-based models, support-vector regression (SVR), K-means Clustering (KMC), and neural networks (NNs) will be used for creating the prediction model.