Dario Garcia Gasulla

Applying COAALAS to SPiDer

Artificial Intelligence techniques and tools have been applied to Assistive Technologies (AT) in order to support elder or impeded people on their daily activities. A common application are intelligent pill dispensers and reminders that help the patient comply with his medication. This has become even more important, as patients suffering from multiple pathologies are prescribed cocktails of drugs that require strict compliance in order to achieve a successful treatment. Existing intelligent pill dispensers tend to focus in the user-tool interaction, neglecting user’s connection with its social environment and the possibility to monitor patient’s behaviour, effectively adapting to a dynamic environment and providing early response to potentially dangerous situations by detecting unexpected or undesired patterns of behaviour. In previous work we have presented COAALAS, an intelligent social and norm-aware device for elder people that is able to autonomously organize, reorganize and interact with the different actors involved in elder-care, either human actors or other devices. In this paper, we present SPiDer an intelligent pill dispenser integrated with the COAALAS architecture.

On the representativeness of convolutional neural networks layers

Convolutional Neural Networks (CNN) are the most popular of deep network models due to their applicability and success in image processing. Although plenty of effort has been made in designing and training better discriminative CNNs, little is yet known about the internal features these models learn. Questions like, what specific knowledge is coded within CNN layers, and how can it be used for other purposes besides discrimination, remain to be answered. To advance in the resolution of these questions, in this work we extract features from CNN layers, building vector representations from CNN activations. The resultant vector embedding is used to represent first images and then known image classes. On those representations we perform an unsupervised clustering process, with the goal of studying the hidden semantics captured in the embedding space. Several abstract entities untaught to the network emerge in this process, effectively defining a taxonomy of knowledge as perceived by the CNN. We evaluate and interpret these sets using WordNet, while studying the different behaviours exhibited by the layers of a CNN model according to their depth. Our results indicate that, while top (i.e., deeper) layers provide the most representative space, low layers also define descriptive dimensions.

Evaluating link prediction on large graphs

Exploiting network data (i.e., graphs) is a rather particular case of data mining. The size and relevance of network domains justifies research on graph mining, but also brings forth severe complications. Computational aspects like scalability and parallelism have to be reevaluated, and well as certain aspects of the data mining process. One of those are the methodologies used to evaluate graph mining methods, particularly when processing large graphs. In this paper we focus on the evaluation of a graph mining task known as Link Prediction. First we explore the available solutions in traditional data mining for that purpose, discussing which methods are most appropriate. Once those are identified, we argue about their capabilities and limitations for producing a faithful and useful evaluation. Finally, we introduce a novel modification to a traditional evaluation methodology with the goal of adapting it to the problem of Link Prediction on large graphs.Exploiting network data (i.e., graphs) is a rather particular case of data mining. The size and relevance of network domains justifies research on graph mining, but also brings forth severe complications. Computational aspects like scalability and parallelism have to be reevaluated, and well as certain aspects of the data mining process. One of those are the methodologies used to evaluate graph mining methods, particularly when processing large graphs. In this paper we focus on the evaluation of a graph mining task known as Link Prediction. First we explore the available solutions in traditional data mining for that purpose, discussing which methods are most appropriate. Once those are identified, we argue about their capabilities and limitations for producing a faithful and useful evaluation. Finally, we introduce a novel modification to a traditional evaluation methodology with the goal of adapting it to the problem of Link Prediction on large graphs.

Discovery of spatio-temporal patterns from location based social networks

Location Based Social Networks (LBSN) like Twitter or Instagram are a good source for user spatio-temporal behavior. These networks collect data from users in such a way that they can be seen as a set of collective and distributed sensors of a geographical area. A low rate sampling of user’s location information can be obtained during large intervals of time that can be used to discover complex patterns, including mobility profiles, points of interest or unusual events. These patterns can be used as the elements of a knowledge base for different applications in different domains like mobility route planning, touristic recommendation systems or city planning. The aim of this paper is twofold, first to analyze the frequent spatio-temporal patterns that users share when living and visiting a city. This behavior is studied by means of frequent itemsets algorithms in order to establish some associations among visits that can be interpreted as interesting routes or spatio-temporal connections. Second, to analyze how the spatio-temporal behavior of a large number of users can be segmented in different profiles. These behavioral profiles are obtained by means of clustering algorithms that show the different patterns of behavior of visitors and citizens. The data analyzed was obtained from the public data feeds of Twitter and Instagram within an area surrounding the cities of Barcelona and Milan for a period of several months. The analysis of these data shows that these kind of algorithms can be successfully applied to data from any city (or general area) to discover useful patterns that can be interpreted on terms of singular places and areas and their temporal relationships.Location Based Social Networks (LBSN) have become an interesting source for mining user behavior. These networks (e.g. Twitter, Instagram or Foursquare) collect spatio-temporal data from users in a way that they can be seen as a set of collective and distributed sensors on a geographical area. Processing this information in different ways could result in patterns useful for several application domains. These patterns include simple or complex user visits to places in a city or groups of users that can be described by a common behavior. The domains of application range from the recommendation of points of interest to visit and route planning for touristic recommender systems to city analysis and planning. This paper presents the analysis of data collected for several months from such LBSN inside the geographical area of two large cities. The goal is to obtain by means of unsupervised data mining methods sets of patterns that describe groups of users in terms of routes, mobility patterns and behavior profiles that can be useful for city analysis and mobility decisions.