Conference Papers

Towards improving the naturalness of social conversations with dialogue systems

Marge M., Miranda J., Black A.W., Rudnicky A.I.

Proceedings of the SIGDIAL 2010 Conference: 11th Annual Meeting of the Special Interest Group onDiscourse and Dialogue

2010

Abstract:

We describe an approach to improving the naturalness of a social dialogue system, Talkie, by adding disfluencies and other content-independent enhancements to synthesized conversations. We investigated whether listeners perceive conversations with these improvements as natural (i.e., human-like) as human-human conversations. We also assessed their ability to correctly identify these conversations as between humans or computers. We find that these enhancements can improve the perceived naturalness of conversations for observers “overhearing” the dialogues.

Towards Independent Stress Detection: A Dependent Model Using Facial Action Units

Viegas C., Lau S.H., Maxion R., Hauptmann A.

2018 International Conference on Content-Based Multimedia Indexing (CBMI)

2018

Abstract:

Our society is increasingly more susceptible to chronic stress. Reasons are daily worries, workload, and the wish to fulfil a myriad of expectations. Unfortunately, long-exposure to stress leads to physical and mental health problems. To avoid the described consequences, mobile applications have been studied to track stress in combination with wearables. However, wearables need to be worn all day long and can be costly. Given that most laptops have inbuilt cameras, using video data for personal tracking of stress levels could be a more affordable alternative. In previous work, videos have been used to detect cognitive stress during driving by measuring the presence of anger or fear through a limited number of facial expressions. In contrast, we propose the use of 17 facial action units (AUs) not solely restricted to those emotions. We used five one-hour long videos from the dataset collected by Lau [1]. The videos show subjects while typing, resting, and exposed to a stressor, being a multitasking exercise combined with social evaluation. We performed binary classification using several simple classifiers on AUs extracted in each video frame and were able to achieve an accuracy of up to 74% in subject independent classification and 91% in subject dependent classification. These preliminary results indicate that the AUs most relevant for stress detection are not consistently the same for all 5 subjects. Also in previous work, using facial cues, a strong person-specific component was found during classification.

Towards natural handshakes for social robots: human-aware hand grasps using tactile sensors

Avelino J., Paulino T., Cardoso C., Nunes R., Moreno P., Bernardino A.

Paladyn, Journal of Behavioral Robotics

2018

Abstract:

Handshaking is a fundamental part of human physical interaction that is transversal to various cultural backgrounds. It is also a very challenging task in the field of Physical Human-Robot Interaction (pHRI), requiring compliant force control in order to plan the arm’s motion and for a confident, but at the same time pleasant grasp of the human user’s hand. In this paper,we focus on the study of the hand grip strength for comfortable handshakes and perform three sets of physical interaction experiments between twenty human subjects in the first experiment, thirty-five human subjects in the second one, and thirty-eight human subjects in the third one. Tests are made with a social robot whose hands are instrumented with tactile sensors that provide skin-like sensation. From these experiments, we: (i) learn the preferred grip closure according to each user group; (ii) analyze the tactile feedback provided by the sensors for each closure; (iii) develop and evaluate the hand grip controller based on previous data. In addition to the robot-human interactions, we also learn about the robot executed handshake interactions with inanimate objects, in order to detect if it is shaking hands with a human or an inanimate object. This work adds physical human-robot interaction to the repertory of social skills of our robot, fulfilling a demand previously identified by many users of the robot.

Towards the benchmarking of question generation: introducing the Monserrate corpus

Rodrigues H., Nyberg E., Coheur L.

Language Resources and Evaluation

2021

Abstract:

Despite the growing interest in Question Generation, evaluating these systems remains notably difficult. Many authors rely on metrics like BLEU or ROUGE instead of relying on manual evaluations, as their computation is mostly free. However, corpora generally used as reference is very incomplete, containing just a couple of hypotheses per source sentence. In this paper, we propose MONSERRATE corpus, a dataset specifically built to evaluate Question Generation systems, with, on average, 26 questions associated to each source sentence, attempting to be an “exhaustive” reference. With MONSERRATE we study the impact of the reference size in evaluating Question Generation systems. Several evaluation metrics are used, from more traditional lexical ones to metrics based on word embeddings, and we conclude that these are still a limiting evaluation factor, as they lead to different outcomes. Finally, with MONSERRATE, we benchmark three different Question Generation systems, representing different approaches to this task.

Towards the Verification of Adaptable Processes

Mario BravettiCinzia Di GiustoJorge A. PérezGianluigi Zavattaro

ISoLA 2012

2012

Abstract:

In prior work, with the aim of formally modeling and analyzing the behavior of concurrent processes with forms of dynamic evolution, we have proposed a process calculus of adaptable processes. Our proposal addressed the (un)decidability of two safety properties related to error occurrence. In order to allow for a more comprehensive verification framework for adaptable processes, the ability to express general properties is most desirable. In this paper we address this important issue: we explain how the proof techniques for (un)decidability results for adaptable processes generalize to a simple yet expressive temporal logic over adaptable processes. We provide examples of the expressiveness of the logic and its significance in relation with the calculus of adaptable processes.

Traffic flow from a low frame rate city camera

Toropov E., Gui L., Zhang S., Kottur S., Moura J.M.F.

Proceedings - International Conference on Image Processing, ICIP

2015

Abstract:

Traffic flow in a city is a rich source of information about the city. Cities are being instrumented with video cameras. They can potentially generate continuously large datasets to be processed (big data). This paper reports on our current work to detect traffic flow from an on-line low quality, low frame rate city video camera. The paper details a pipeline of four main steps – background subtraction, scene geometry, car detection, and car counting, and it illustrates results obtained with processing video from a single camera.

Transactive Community Microgrids to Share Energy Storage Resources in Portugal

Moura P., Sriram U., Mohammadi J.

2020

Abstract:

The contribution of renewable energy sources to Portugal’s energy generation portfolio is significant and on the path to achieving 100% renewable generation by 2050. Most of the new renewable generation capacity will be procured from distributed photovoltaic (PV) generation installed at buildings. The inherent intermittence of PV output combined with a mismatch with demand profile are challenging the operation and resiliency of the electrical grid. Addressing these issues requires leveraging spatio-temporal flexibility of controllable energy resources such as batteries and Electric Vehicles (EV). This need is recognized by regulators in Portugal and the recent renewable generation self-consumption legislation enables generation-surplus trading in communities. Implementing intra-community trading and utilizing the potentials of renewable generation requires oversight and coordination at the community level in the context of transactive energy systems. This paper focuses on addressing energy sharing through a transactive energy market in community microgrids. The proposed framework considers public and commercial buildings with on-site battery storage and numerous EV charging stations as the main source of flexibility. The formulation is tested using real data from a community of buildings on a Portuguese University campus. The results showcase the achieved increase in renewable self-consumption at building and community levels, as well as the reduction in electricity costs.

Transition-based dependency parsing with stack long short-term memory

Dyer C., Ballesteros M., Ling W., Matthews A., Smith N.A.

ACL-IJCNLP 2015 - 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing of the Asian Federation of Natural Language Processing, Proceedings of the Conference

2015

Abstract:

We propose a technique for learning representations of parser states in transitionbased dependency parsers. Our primary innovation is a new control structure for sequence-to-sequence neural networks— the stack LSTM. Like the conventional stack data structures used in transitionbased parsing, elements can be pushed to or popped from the top of the stack in constant time, but, in addition, an LSTM maintains a continuous space embedding of the stack contents. This lets us formulate an efficient parsing model that captures three facets of a parser’s state: (i) unbounded look-ahead into the buffer of incoming words, (ii) the complete history of actions taken by the parser, and (iii) the complete contents of the stack of partially built tree fragments, including their internal structures. Standard backpropagation techniques are used for training and yield state-of-the-art parsing performance.

TrimTuner: Efficient Optimization of Machine Learning Jobs in the Cloud via Sub-Sampling

Mendes P., Casimiro M., Romano P., Garlan D.

IEEE MASCOTS’2020

2020

Abstract:

This work introduces TrimTuner, the first system for optimizing machine learning jobs in the cloud to exploit sub-sampling techniques to reduce the cost of the optimization process while keeping into account user-specified constraints. TrimTuner jointly optimizes the cloud and application-specific parameters and, unlike state of the art works for cloud optimization, eschews the need to train the model with the full training set every time a new configuration is sampled. Indeed, by leveraging sub-sampling techniques and data-sets that are up to 60x smaller than the original one, we show that TrimTuner can reduce the cost of the optimization process by up to 50x. Further, TrimTuner speeds-up the recommendation process by 65x with respect to state of the art techniques for hyper-parameter optimization that use sub-sampling techniques. The reasons for this improvement are twofold: i) a novel domain specific heuristic that reduces the number of configurations for which the acquisition function has to be evaluated; ii) the adoption of an ensemble of decision trees that enables boosting the speed of the recommendation process by one additional order of magnitude.