Bar-Ilan University

About: Bar-Ilan University is a education organization based out in Ramat Gan, Israel. It is known for research contribution in the topics: Population & Poison control. The organization has 12835 authors who have published 34964 publications receiving 995648 citations. The organization is also known as: Bar Ilan University & BIU.

Adversarial Removal of Demographic Attributes from Text Data

Abstract: Recent advances in Representation Learning and Adversarial Training seem to succeed in removing unwanted features from the learned representation. We show that demographic information of authors is encoded in—and can be recovered from—the intermediate representations learned by text-based neural classifiers. The implication is that decisions of classifiers trained on textual data are not agnostic to—and likely condition on—demographic attributes. When attempting to remove such demographic information using adversarial training, we find that while the adversarial component achieves chance-level development-set accuracy during training, a post-hoc classifier, trained on the encoded sentences from the first part, still manages to reach substantially higher classification accuracies on the same data. This behavior is consistent across several tasks, demographic properties and datasets. We explore several techniques to improve the effectiveness of the adversarial component. Our main conclusion is a cautionary one: do not rely on the adversarial training to achieve invariant representation to sensitive features.

Di-tert-butyl dicarbonate and 4-(dimethylamino)pyridine revisited. Their reactions with amines and alcohols

Abstract: The reaction of BOC2O in the presence and absence of DMAP was examined with some amines, alcohols, diols, amino alcohols, and aminothiols. Often, unusual products were observed depending on the ratio of reagents, reaction time, polarity of solvent, pKa of alcohols, or type of amine (primary or secondary). In reactions of aliphatic alcohols with BOC2O/DMAP, we isolated for the first time carbonic-carbonic anhydride intermediates; this helps explain the formation of symmetrical carbonates in addition to the O-BOC products. In the case of secondary amines, we succeeded to isolate unstable carbamic-carbonic anhydride intermediates that in the presence of DMAP led to the final N-BOC product. The effect of N-methylimidazole in place of DMAP was also examined.

Scaling theory of transport in complex biological networks.

Abstract: Transport is an important function in many network systems and understanding its behavior on biological, social, and technological networks is crucial for a wide range of applications. However, it is a property that is not well understood in these systems, probably because of the lack of a general theoretical framework. Here, based on the finding that renormalization can be applied to bionetworks, we develop a scaling theory of transport in self-similar networks. We demonstrate the networks invariance under length scale renormalization, and we show that the problem of transport can be characterized in terms of a set of critical exponents. The scaling theory allows us to determine the influence of the modular structure on transport in metabolic and protein-interaction networks. We also generalize our theory by presenting and verifying scaling arguments for the dependence of transport on microscopic features, such as the degree of the nodes and the distance between them. Using transport concepts such as diffusion and resistance, we exploit this invariance, and we are able to explain, based on the topology of the network, recent experimental results on the broad flow distribution in metabolic networks.

Leading by example: the case of leader OCB.

Abstract: The importance of leading by personal example or role modeling for effective leadership has been recognized in many leadership theories. However, leaders' ability to influence group behavior through exemplary behavior has received little attention in empirical work. This study explores leading by example through theoretical development and empirical testing of a moderated mediation model of the potential effects of leader organizational citizenship behavior (OCB). This model suggests that a leader's OCB may promote group OCB directly and indirectly by enhancing the group's belief that OCB is worthy. It also specifies the moderators of the direct and indirect effects of leader OCB on group OCB. Data from 683 members of 67 intact work groups, 67 group managers, and their supervisors support the hypothesized model. The theoretical and practical implications of these findings are discussed.

A Little Is Enough: Circumventing Defenses For Distributed Learning

Abstract: Distributed learning is central for large-scale training of deep-learning models. However, it is exposed to a security threat in which Byzantine participants can interrupt or control the learning process. Previous attack models assume that the rogue participants (a) are omniscient (know the data of all other participants), and (b) introduce large changes to the parameters. Accordingly, most defense mechanisms make a similar assumption and attempt to use statistically robust methods to identify and discard values whose reported gradients are far from the population mean. We observe that if the empirical variance between the gradients of workers is high enough, an attacker could take advantage of this and launch a non-omniscient attack that operates within the population variance. We show that the variance is indeed high enough even for simple datasets such as MNIST, allowing an attack that is not only undetected by existing defenses, but also uses their power against them, causing those defense mechanisms to consistently select the byzantine workers while discarding legitimate ones. We demonstrate our attack method works not only for preventing convergence but also for repurposing of the model behavior (``backdooring''). We show that less than 25\% of colluding workers are sufficient to degrade the accuracy of models trained on MNIST, CIFAR10 and CIFAR100 by 50\%, as well as to introduce backdoors without hurting the accuracy for MNIST and CIFAR10 datasets, but with a degradation for CIFAR100.