Publications

AD³: Alternating directions dual decomposition for map inference in graphical models

Martins A. F. T., Figueiredo M. A. T., Aguiar P. M. Q., Smith N. A., Xing E. P.

Journal of Machine Learning Research

2015

Article

Abstract:

We present AD3 a new algorithm for approximate maximum a posteriori (MAP) inference on factor graphs, based on the alternating directions method of multipliers. Like other dual decomposition algorithms, AD3 has a modular architecture, where local subproblems are solved independently, and their solutions are gathered to compute a global update. The key characteristic of AD3 is that each local subproblem has a quadratic regularizer, leading to faster convergence, both theoretically and in practice. We provide closed-form solutions for these AD3 subproblems for binary pairwise factors and factors imposing first-order logic constraints. For arbitrary factors (large or combinatorial), we introduce an active set method which requires only an oracle for computing a local MAP configuration, making AD3 applicable to a wide range of problems. Experiments on synthetic and real-world problems show that AD3 compares favorably with the state-of-the-art. Keywords: MAP inference, graphical models, dual decomposition, alternating directions method of multipliers.

Adaptable Processes (Extended Abstract)

Mario BravettiCinzia Di GiustoJorge A. PérezGianluigi Zavattaro

FMOODS 2011, FORTE 2011

2008

Conference Paper

Abstract:

We propose the concept of adaptable processes as a way of overcoming the limitations that process calculi have for describing patterns of dynamic process evolution. Such patterns rely on direct ways of controlling the behavior and location of running processes, and so they are at the heart of the adaptation capabilities present in many modern concurrent systems. Adaptable processes have a location and are sensible to actions of dynamic update at runtime. This allows to express a wide range of evolvability patterns for processes. We introduce a core calculus of adaptable processes and propose two verification problems for them: bounded and eventual adaptation. While the former ensures that at most k consecutive errors will arise in future states, the latter ensures that if the system enters into an error state then it will eventually reach a correct state. We study the (un)decidability of these two problems in different fragments of the calculus. Rather than a specification language, our calculus intends to be a basis for investigating the fundamental properties of evolvable processes and for developing richer languages with evolvability capabilities.

Adaptive Temporal Triplet-loss for Cross-modal Embedding Learning

Semedo D, Magalhães J.

ACM International Conference on Multimedia

2020

Conference Paper

Abstract:

There are many domains where the temporal dimension is critical to unveil how different modalities, such as images and texts, are correlated. Notably, in the social media domain, information is constantly evolving over time according to the events that take place in the real world. In this work, we seek for highly expressive loss functions that allow the encoding of data temporal traits into cross-modal embedding spaces. To achieve this goal, we propose to steer the learning procedure of such embedding through a set of adaptively enforced temporal constraints. In particular, we propose a new formulation of the triplet loss function, where the traditional static margin is superseded by a novel temporally adaptive maximum margin function. This novel redesign of the static margin formulation, allows the embedding to effectively capture not only the semantic correlations across data modalities, but also data’s fine-grained temporal correlations. Our experiments confirm the effectiveness of our model in structuring different modalities, while organizing data according to temporal correlations. Moreover, we experimentally highlight how can these embeddings be used for multimedia understanding.

Adaptive-Rate Reconstruction of Time-Varying Signals with Application in Compressive Foreground Extraction

Mota J.F.C., Deligiannis N., Sankaranarayanan A.C., Cevher V., Rodrigues M.R.D.

IEEE Transactions on Signal Processing

2016

Article

Abstract:

We propose and analyze an online algorithm for reconstructing a sequence of signals from a limited number of linear measurements. The signals are assumed sparse, with unknown support, and evolve over time according to a generic nonlinear dynamical model. Our algorithm, based on recent theoretical results for l 1 – l 1 minimization, is recursive and computes the number of measurements to be taken at each time on-the-fly. As an example, we apply the algorithm to online compressive video foreground extraction, a problem stated as follows: given a set of measurements of a sequence of images with a static background, simultaneously reconstruct each image while separating its foreground from the background. The performance of our method is illustrated on sequences of real images. We observe that it allows a dramatic reduction in the number of measurements or reconstruction error with respect to state-of-the-art compressive background subtraction schemes.

ADMM for consensus on colored networks

Mota J.F.C., Xavier J.M.F., Aguiar P.M.Q., Puschel M.

Proceedings of the IEEE Conference on Decision and Control

2012

Conference Paper

Abstract:

We propose a novel distributed algorithm for one of the most fundamental problems in networks: the average consensus. We view the average consensus as an optimization problem, which allows us to use recent techniques and results from the optimization area. Based on the assumption that a coloring scheme of the network is available, we derive a decentralized, asynchronous, and communication-efficient algorithm that is based on the Alternating Direction Method of Multipliers (ADMM). Our simulations with other state-of-the-art consensus algorithms show that the proposed algorithm is the one exhibiting the most stable performance across several network models.

Advances on Smart Object Management

Pentikousis K., Agüero R., Timm-Giel A., Sargento S.

Personal and Ubiquitous Computing

2012

Journal

Abstract:

The first part of this issue features four papers that discuss advanced management techniques for Smart Objects. The so-called Internet of Things (IoT) is one of the cornerstones of the Future Internet. One illustrative example of the relevance of IoT in future network development is its growing adoption within the smart city paradigm, as a means to provide enhanced citizen services. In this sense, basic IoT technology is no longer at the purely academic research level, but is starting to be integrated to the the fabric of our daily activities.

Adversarial Multiple Source Domain Adaptation

Zhao H., Zhang S., Wu G., Costeira J.P., Moura J.M.F. , Gordon G.J.

32nd Conference on Neural Information Processing Systems (NIPS 2018), Montreal, Canada.

2018

Conference Paper

Abstract:

While domain adaptation has been actively researched, most algorithms focus on the single-source-single-target adaptation setting. In this paper we propose new generalization bounds and algorithms under both classification and regression settings for unsupervised multiple source domain adaptation. Our theoretical analysisnaturally leads to an efficient learning strategy using adversarial neural networks:we show how to interpret it as learning feature representations that are invariant to the multiple domain shifts while still being discriminative for the learning task. To this end, we propose multisource domain adversarial networks (MDAN) that approach domain adaptation by optimizing task-adaptive generalization bounds. To demonstrate the effectiveness of MDAN, we conduct extensive experiments showing superior adaptation performance on both classification and regression problems: sentiment analysis, digit classification, and vehicle counting.

Æminium: A permission-based concurrent-by-default programming language approach

Stork S., Naden K., Sunshine J., Mohr M., Fonseca A., Marques P., Aldrich J.

ACM Transactions on Programming Languages and Systems

2014

Article

Abstract:

Writing concurrent applications is extremely challenging, not only in terms of producing bug-free and maintainable software, but also for enabling developer productivity. In this article we present the Æminium concurrent-by-default programming language. Using Æminium programmers express data dependencies rather than control flow between instructions. Dependencies are expressed using permissions, which are used by the type system to automatically parallelize the application. The Æminium approach provides a modular and composable mechanism for writing concurrent applications, preventing data races in a provable way. This allows programmers to shift their attention from low-level, error-prone reasoning about thread interleaving and synchronization to focus on the core functionality of their applications. We study the semantics of Æminium through μÆminium, a sound core calculus that leverages permission flow to enable concurrent-by-default execution. After discussing our prototype implementation we present several case studies of our system. Our case studies show up to 6.5X speedup on an eight-core machine when leveraging data group permissions to manage access to shared state, and more than 70% higher throughput in a Web server application.

AHA-3D: A Labelled Dataset for Senior Fitness Exercise Recognition and Segmentation from 3D Skeletal Data

Antunes J., Bernardino A., Smailagic Assim, SiewiorekD.

Workshop, British Machine Vision Conference (BMVC). Newcastle upon Tyne, UK, 3-6 September 2018

2018

Conference Paper

Abstract:

Automated assessment of fitness exercises has important applications in computer and robot-based exercise coaches to deploy at home, gymnasiums or care centers. In this work, we introduce AHA-3D, a labeled dataset of sequences of 3D skeletal data depicting standard fitness tests on young and elderly subjects, for the purpose of automatic fitness exercises assessment. To the best of our knowledge, AHA-3D is the first publicly available dataset featuring multi-generational, male and female subjects, with frame-level labels, allowing for action segmentation as well as the estimation of metrics like risk of fall, and autonomy to perform daily tasks. We present two baseline methods for recognition and one for segmentation. For recognition, we trained models on the positions of the joints achieving 88.2% ± 0.077 accuracy, and on joint positions and velocities, achieving 91% ±0.082 accuracy. Using the Kolmogorov-Smirnov test we determined the model trained on velocities was superior. The segmentation baseline achieved an accuracy of 88.29% in detecting actions at frame level. Our results show promising recognition and detection performance suggesting AHA3D’s potential use in practical applications like exercise performance and correction, elderly fitness level estimation and risk of falling for elders.