No abstract

Many studies focused on detecting and measuring the security and privacy risks associated with the integration of advertising libraries in mobile apps. These studies consistently demonstrate the abuses of existing ad libraries. However, to fully assess the risks of an app that uses an advertising library, we need to take into account not only the current behaviors but all of the allowed behaviors that could result in the compromise of user data confidentiality. Ad libraries on Android have potential for greater data collection through at least four major channels: using unprotected APIs to learn other apps' information on the phone (e.g., app names); using protected APIs via permissions inherited from the host app to access sensitive information (e.g. Google and Facebook account information, geo locations); gaining access to files which the host app stores in its own protection domain; and observing user inputs into the host app.In this work, we systematically explore the potential reach of advertising libraries through these channels. We design a framework called Pluto that can be leveraged to analyze an app and discover whether it exposes targeted user data-such as contact information, interests, demographics, medical conditions and so on--to an opportunistic ad library. We present a prototype implementation of Pluto, that embodies novel strategies for using natural language processing to illustrate what targeted data can potentially be learned from an ad network using files and user inputs. Pluto also leverages machine learning and data mining models to reveal what advertising networks can learn from the list of installed apps. We validate Pluto with a collection of apps for which we have determined ground truth about targeted data they may reveal, together with a data set derived from a survey we conducted that gives ground truth for targeted data and corresponding lists of installed apps for about 300 users. We use these to show that Pluto, and hence also opportunistic ad networks, can achieve 75% recall and 80% precision for selected targeted data coming from app files and inputs, and even better results for certain targeted data based on the list of installed apps. Pluto is the first tool that estimates the risk associated with integrating advertising in apps based on the four available channels and arbitrary sets of targeted data.

No abstract

This paper tackles the problem of reading comprehension over long narratives where documents easily span over thousands of tokens. We propose a curriculum learning (CL) based Pointer-Generator framework for reading/sampling over large documents, enabling diverse training of the neural model based on the notion of alternating contextual difficulty. This can be interpreted as a form of domain randomization and/or generative pretraining during training. To this end, the usage of the Pointer-Generator softens the requirement of having the answer within the context, enabling us to construct diverse training samples for learning. Additionally, we propose a new Introspective Alignment Layer (IAL), which reasons over decomposed alignments using block-based self-attention. We evaluate our proposed method on the NarrativeQA reading comprehension benchmark, achieving state-of-the-art performance, improving existing baselines by 51% relative improvement on BLEU-4 and 17% relative improvement on Rouge-L. Extensive ablations confirm the effectiveness of our proposed IAL and CL components.

No abstract