Towards Artificial Entities that Learn Through Dialog: Intention-Based Knowledge Extraction and Lear
Artificial entities able to interact naturally using human language and adapt to different situations by learning from those interactions have several potential applications. To advance some steps towards their development, we propose to adapt a dialog system, allowing it to extract knowledge from the segments in a dialog and use it to update its capabilities. We start by exploring generic intention recognition, which allows the identification of segments that are able to provide knowledge and the kind of knowledge that they are able to provide. Then, we propose a semantic frame induction approach, which allows the identification of domains, the disambiguation of concepts according to the context, and the identification of related concepts. After that, we propose an intention-based knowledge extraction approach, which is able to extract open-domain knowledge from the dialog by applying specialized approaches according to the intention behind the segments. The extracted knowledge can then be used to update a semantic network by linking the referred concepts with their representation in the network, creating new nodes and solving conflicts if necessary. To bridge the gap between the conceptual knowledge present in the semantic network and its observation in the environment, we propose a grounding approach that includes the identification of references to observable concepts in the dialog, the pairing of those references with the corresponding observations, and the incremental learning of concept models. Finally, we define three evaluation scenarios, which involve human interaction and cover multiple aspects of the learning process.
access_time October 12, 2020 at 03:30PM
place https://videoconf-colibri.zoom.us/j/99170561241 Password: 338296
local_offer CAT exam
person Candidato: Eugénio Alves Ribeiro
supervisor_account Orientador: Prof. David Matos
Personality-based Persuasion by an Interactive Social Robot Storyteller
Research has shown that it is possible to improve storytelling techniques, making them more personalised and efficient. By using a person’s preferences, one can make a task and a story more pleasurable, attractive, encouraging and even more challenging for each participant. The use of a persuasive strategy based on the person's personality traits tends to achieve better results as it acts in accordance with the person’s preferences. This work proposes a methodology to classify the preferences and the virtual behaviour of a person into a decision-making task. It also combines a persuasion model that can be used by a social robot as a persuasive mechanism in a scenario for interactive storytelling. The persuasion model is based on personality traits and can help to motivate the person's behaviour, attitudes, actions and knowledge in a natural way. Further, this work also reports on a study performed so far with an Interactive Storytelling scenario. It is explained how this scenario can be used to classify the decisions made into the story using a personality theory, allowing to identify user's personality traits. Plus, how through those decisions it was possible to identify virtual behaviours (patterns of choice) among the participants.
access_time October 09, 2020 at 01:30PM
local_offer Doctoral exam
person Candidato: Raul Benites Paradeda
supervisor_account Orientador 1: Prof. Ana Paiva
supervisor_account Orientador 2: Prof. Carlos Martinho
Anomaly detection in Maritime Vessels Tracks using Multivariate Temporal Data
From cybersecurity to life sciences anomaly detection in temporal data is considered crucial as it often enables the detection of relevant semantic information that can help to prevent and detect events such as cyber attacks or patients heart-attacks. While temporal data is commonly studied from an univariate perspective where only one data dimension changes over time, few works have been dedicated to multi-variate temporal data anomaly detection. In real world scenarios data series are typically multivariate, noisy, of high dimensionality, non stationary and irregular. In this work we provide a brief overview on the main definitions and approaches supporting our study case context which is characterized by geographical referenced and highly dynamic data series. After identifying knowledge issues we raise questions that drive our research mainly concentrating in the identification of unusual behaviour, semantic information retrieval and contextual data usage.
access_time September 17, 2020 at 09:00AM
place https://videoconf-colibri.zoom.us/j/96083267690 Password: 376978
local_offer CAT exam
person Candidato: Rui Maia
supervisor_account Orientador: Prof. Cláudia Antunes
Best of All Worlds for Low-resource Natural Language Processing
The state of the art of several Natural Language Processing (NLP) tasks in the latest years has been dominated by deep learning systems, whose performance is enabled by the availability of massive amounts of labeled data, especially for English. Annotation is a costly process, often requiring linguistic or domain expertise, which leads many languages to lack large annotated corpora for certain domains and/or tasks. Thus, it would be desirable to enable NLP applications based on small annotated corpora (e.g., with only a few hundred sentences or less), or even none at all, not only for training purposes, but also for the evaluation of concurrent approaches. In this thesis, we address the problem of deploying the best approach for an NLP low-resource scenario by taking advantage of already existing human feedback, without having to evaluate concurrent approaches on large enough (annotated) test set. We tackle this as a problem of prediction with expert advice, making it our goal to dynamically converge to the performance of the best approach across different scenarios. We are currently applying this proposal to two distinct scenarios. The first scenario is that of using Active Learning to query the most useful sentences to be annotated by a human in sequence labeling tasks. In this scenario, the goal is to converge towards the best individual query strategy. The second scenario is a conversational agent that aims at retrieving an appropriate answer to a user request, from a collection of answers. Here, our goal is to converge to the best criterion for selecting an answer. In preliminary experiments for both scenarios, we observed convergence towards the best performing individual approaches. As our next steps, we intend to extend the experiments made so far, both in terms of scenarios evaluated and evaluation methodology.
access_time September 16, 2020 at 02:00PM
place https://videoconf-colibri.zoom.us/j/99137153363 Password: 820632
local_offer CAT exam
person Candidato: Vânia Mendonça
supervisor_account Orientador: Prof. Luísa Coheur
Algorithms of Cooperation
One hundred and fifty years following the publication of Darwin’s “The Origin of Species”, the emergence of cooperative action remains one of the biggest challenges that science is facing at present, recently classified as one of the top scientific problems for the 21st century by Science’s invited panel of Scientists. From the emergence of multi-cellularity, to the evolution of social behaviour in humans, climate action and managing of global pandemics, many problems may be formulated as challenges of cooperation, fascinating mathematicians, philosophers, biologists, computer scientists, and economists alike, to name a few. Adopting the terminology resulting from the seminal work of Hamilton, Trivers, and Wilson, cooperation can be defined as a costly behavior that provides a greater benefit to another individual (or group of individuals). In this context, even for the simplest living organisms, the pervasiveness of cooperation is a theoretical paradox difficult to explain, although all so often observed in reality. Darwin even referred cooperation as “one special difficulty, which first appeared to me insuperable, and actually fatal to my theory”. If evolution is characterized by competition and the survival of the fittest, why should selfish unrelated individuals cooperate with each other? The same question arises among humans. Why humans share resources, engage in joint enterprises, or should nurture the generations to come? In all cases, the immediate advantage of free riding, can drive the population into the tragedy of the commons, the famous Hardin’s doomsday scenario of widespread defection. Thus, it is not only challenging to understand the mechanisms underlying the emergence of cooperation in nature and societies, but also daring to apply our current understanding of Human cooperation to foster pro-sociality in situations in which cooperation remains astray. The last decades have witnessed the discovery of several core mechanisms leading to the emergence, promotion and maintenance of cooperation at different levels of organization. Many species (e.g., social insects) rely on genetic and group ties to portray admirable levels of cooperation. By contrast, Humans found their way to cooperation beyond related individuals, developing complex mechanisms of reciprocity, social networking, signaling, commitments, sanctions and (sanctioning) Institutions, and social norms, among others. This talent for cooperation forms one of the cornerstones of human society, and is, as such, also largely responsible for Human’s unprecedented success.
access_time September 09, 2020 at 01:00PM
local_offer Habilitation exam
person Candidato: Doutor Francisco João Duarte Cordeiro Correia dos Santos