Conference Papers

Understanding and capturing people’s privacy policies in a mobile social networking application

Sadeh N., Hong J., Cranor L., Fette I., Kelley P., Prabaker M., Rao J.

Personal and Ubiquitous Computing

2008

Abstract:

A number of mobile applications have emerged that allow users to locate one another. However, people have expressed concerns about the privacy implications associated with this class of software, suggesting that broad adoption may only happen to the extent that these concerns are adequately addressed. In this article, we report on our work on PEOPLEFINDER, an application that enables cell phone and laptop users to selectively share their locations with others (e.g. friends, family, and colleagues). The objective of our work has been to better understand people’s attitudes and behaviors towards privacy as they interact with such an application, and to explore technologies that empower users to more effectively and efficiently specify their privacy preferences (or “policies”). These technologies include user interfaces for specifying rules and auditing disclosures, as well as machine learning techniques to refine user policies based on their feedback. We present evaluations of these technologies in the context of one laboratory study and three field studies.

Understanding sequential decisions via inverse reinforcement learning

Liu S., Araujo M., Brunskill E., Rossetti R., Barros J., Krishnan R.

Proceedings - IEEE International Conference on Mobile Data Management

2013

Abstract:

The execution of an agent’s complex activities, comprising sequences of simpler actions, sometimes leads to the clash of conflicting functions that must be optimized. These functions represent satisfaction, short-term as well as long-term objectives, costs and individual preferences. The way that these functions are weighted is usually unknown even to the decision maker. But if we were able to understand the individual motivations and compare such motivations among individuals, then we would be able to actively change the environment so as to increase satisfaction and/or improve performance. In this work, we approach the problem of providing highlevel and intelligible descriptions of the motivations of an agent, based on observations of such an agent during the fulfillment of a series of complex activities (called sequential decisions in our work). A novel algorithm for the analysis of observational records is proposed. We also present a methodology that allows researchers to converge towards a summary description of an agent’s behaviors, through the minimization of an error measure between the current description and the observed behaviors. This work was validated using not only a synthetic dataset representing the motivations of a passenger in a public transportation network, but also real taxi drivers’ behaviors from their trips in an urban network. Our results show that our method is not only useful, but also performs much better than the previous methods, in terms of accuracy, efficiency and scalability.

Understanding the Mechanics of SPIGOT: Surrogate Gradients for Latent Structure Learning

Mihaylova T., Niculae V., . Martins A

Proc. of EMNLP 2020

2020

Abstract:

Latent structure models are a powerful tool for modeling language data: they can mitigate the error propagation and annotation bottleneck in pipeline systems, while simultaneously uncovering linguistic insights about the data. One challenge with end-to-end training of these models is the argmax operation, which has null gradient. In this paper, we focus on surrogate gradients, a popular strategy to deal with this problem. We explore latent structure learning through the angle of pulling back the downstream learning objective. In this paradigm, we discover a principled motivation for both the straight-through estimator (STE) as well as the recently-proposed SPIGOT – a variant of STE for structured models. Our perspective leads to new algorithms in the same family. We empirically compare the known and the novel pulled-back estimators against the popular alternatives, yielding new insight for practitioners and revealing intriguing failure cases.

Understanding the practical issues of deploying energy monitoring and eco-feedback technology in the wild: Lesson learned from three long-term deployments

Pereira L., Nunes N.

Energy Reports

2020

Abstract:

This paper reports on the different engineering, social and financial challenges behind the building and deploying electric energy monitoring and eco-feedback technology in real-world scenarios, which despite being relevant to the research community are seldom reported in the literature. The objectives of this paper are two-fold: First, discuss the technical and social constraints of real-world deployments. This includes, for example, hardware and software requirements, and issues related to security and intrusiveness of the monitoring solutions. Second, identify and understand the costs associated with developing and deploying such systems. These include hardware costs and consumed energy. To this end, we rely on over five years of experience developing and improving a non-intrusive energy monitoring research platform to enable the deployment of long and short-term studies of eco-feedback technology. During this time, two versions of that platform were deployed in 50 homes for periods that lasted between 6 and 18 consecutive months. By iteratively developing and deploying our sensing and eco-feedback infrastructures, we managed to build upon previous findings and lessons learned to understand how to create, deploy, and maintain such systems. Concurrently, we gained insights regarding what are some of the most relevant costs associated with running such experiments.

Understanding the role of social networks on labor market outcomes using a large dataset from a mobile network

Reis F., Ferreira P.

2015 International Conference on Information Systems: Exploring the Information Frontier, ICIS 2015

2015

Abstract:

We use a new and unique dataset combining social network data from Call Detail Records with employment information on mobile phone subscribers to study the role of information networks on job market outcomes. The novel contribution of our work is to focus on the effect of actual social connections beyond that associated to living in the same neighborhood. We find that the propensity to work together is two orders of magnitude greater for a pair of neighbors who call each other than that for a pair of neighbors who do not, suggesting that actual social ties play a significant role in learning about job opportunities. We also find that social networks play a stronger role in less privileged neighborhoods, which provides some evidence that social networks may be unable to mitigate the insulation problems of such neighborhoods. Keywords: Job Information Networks; Call Detail Records

Unicast communication in vehicular ad hoc networks: A reality check

Boban M., Tonguz O., Barros J.

IEEE Communications Letters

2009

Abstract:

We characterize the unicast performance available to applications in infrastructureless vehicular ad hoc networks (VANETs) in terms of connection duration, packet delivery ratio, end-to-end delay, and jitter in both highway and urban VANET environments. The results show the existence of several stringent QoS constraints for unicast applications in infrastructureless VANETs

Unifying nuclear norm and bilinear factorization approaches for low-rank matrix decomposition

Cabral R., Torre F.D.L., Costeira J.P., Bernardino A.

Proceedings of the IEEE International Conference on Computer Vision

2013

Abstract:

Low rank models have been widely used for the representation of shape, appearance or motion in computer vision problems. Traditional approaches to fit low rank models make use of an explicit bilinear factorization. These approaches benefit from fast numerical methods for optimization and easy kernelization. However, they suffer from serious local minima problems depending on the loss function and the amount/type of missing data. Recently, these low-rank models have alternatively been formulated as convex problems using the nuclear norm regularizer, unlike factorization methods, their numerical solvers are slow and it is unclear how to kernelize them or to impose a rank a priori. This paper proposes a unified approach to bilinear factorization and nuclear norm regularization, that inherits the benefits of both. We analyze the conditions under which these approaches are equivalent. Moreover, based on this analysis, we propose a new optimization algorithm and a “rank continuation” strategy that outperform state-of-the-art approaches for Robust PCA, Structure from Motion and Photometric Stereo with outliers and missing data.

Universal grapheme-based speech synthesis

Sitaram S., Parlikar A., Anumanchipalli G.K., Black A.W.

Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

2015

Abstract:

Grapheme-to-phoneme conversion follows the text processing step in speech synthesis. Typically, lexicons or Letter-to-Sound rules are used to map graphemes to phonemes. However, in some languages, such resources may not be readily available. In this paper, we describe a universal front end that supports using grapheme information alone to build usable speech synthesis systems. This work takes advantage of an explicit mapping of Unicode characters from a wide range of scripts to a single phoneset to create support for building speech synthesizers for most languages in the world. We compare the efficacy of this front end to the baseline approach of treating every single grapheme as a separate phoneme for synthesis by building voices for twelve languages across several language families and to front ends with linguistic knowledge in languages with higher resources. In addition, we improve our models by using Random Forests as opposed to using single Classification and Regression Trees. We find that the common universal front end performs better than the raw graphemes in general. We also find that using Random Forests lead to a significant improvement in synthesis quality, which is better than the quality of the knowledge based front end in many cases.

Unsupervised image segmentation using comparative reasoning and random walks

Kulkarni A., Condessa F., Kovacevic J.

2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015

2016

Abstract:

An image segmentation method that does not need training data can provide faster results than methods using complex optimization. Motivated by this idea, we present an unsupervised image segmentation method that combines comparative reasoning with graph-based clustering. Comparative reasoning enables fast similarity search on the image, and these search results are used with the Random Walks algorithm, which is used for clustering and calculating class probabilities. Our method is validated on diverse image modalities such as biomedical images, natural images and texture images. The performance of the method is measured through cluster purity based on available ground truth. Our results are compared to existing segmentation methods using Global Consistency Error scores.