Action recognition plays an important role in various applications such as video monitoring, automatic video indexing, crowd analysis, human-machine interaction, smart homes and personal assistive robotics. In this paper, we propose improvements to some methods for human action recognition from videos that work with data represented in the form of skeleton poses. These methods are based on the most widely used techniques for this problem—Graph Convolutional Networks (GCNs), Temporal Convolutional Networks (TCNs) and Recurrent Neural Networks (RNNs). Initially, the paper explores and compares different ways to extract the most relevant spatial and temporal characteristics for a sequence of frames describing an action. Based on this comparative analysis, we show how a TCN type unit can be extended to work even on the characteristics extracted from the spatial domain. To validate our approach, we test it against a benchmark often used for human action recognition problems and we show that our solution obtains comparable results to the state-of-the-art, but with a significant increase in the inference speed.

Artificial Intelligence is currently the new technological revolution. AI influences all domains of science, technology, industry, and society. The paper shortly discusses two views on the relationship between AI and education: one on how can AI enhance education, help personalize the learning experience, help teachers in their endeavor, and its impact on e-learning, and the other on how education in AI has to be conceived to create the work force required to face this new technological revolution.

Recent studies in social robotics show that it can provide economic efficiency and growth in domains such as retail, entertainment, and active and assisted living (AAL). Recent work also highlights that users have the expectation of affordable social robotics platforms, providing focused and specific assistance in a robust manner. In this paper, we present the AMIRO social robotics framework, designed in a modular and robust way for assistive care scenarios. The framework includes robotic services for navigation, person detection and recognition, multi-lingual natural language interaction and dialogue management, as well as activity recognition and general behavior composition. We present AMIRO platform independent implementation based on a Robot Operating System (ROS). We focus on quantitative evaluations of each functionality module, providing discussions on their performance in different settings and the possible improvements. We showcase the deployment of the AMIRO framework on a popular social robotics platform—the Pepper robot—and present the experience of developing a complex user interaction scenario, employing all available functionality modules within AMIRO.

International audienceApplications envisioned for the Internet of Things (IoT) would generally have to fulfill their design goals by mashing up devices and digital services in a manner that is both flexible, such that they can adapt to dynamic environments, and responsive, such that they can react to sensor and user input in a timely fashion. Most existing approaches for the development of IoT applications rely on precompiled mashups that are highly responsive, but inflexible due to their static nature. At the other end of the spectrum, fully automatic composition of services results in IoT mashups that are highly flexible, but responsive only for small numbers of IoT services. This paper presents a middle ground approach: goal-driven software agents are equipped with precompiled mashups and cooperate with one another to compose their mashups at runtime in pursuit of their goals. Agents are interconnected via relations that enable them to discover and interact with one another in a flexible manner. To support our approach, we provide an open-source platform that facilitates application development. We used this platform to implement a realistic IoT application that achieves its design goal by mashing-up multiple heterogeneous devices at runtime. Evaluation results suggest that applications remain responsive when scaling to many devices and for relatively large mashup compositions

Sentiment analysis plays an important role in automatically finding the polarity and insights of users with regards to a specific subject, events, and entity. In this article, we propose a new topic-document embedding (TOPICDOC2VEC) for detecting the polarity of a text. The TOPICDOC2VEC is constructed as the concatenation between a document embedding (DOC2VEC) and a topic embedding (TOPIC2VEC). A DOC2VEC is computed using word embeddings weighted by TF-IDF, while a TOPIC2VEC is computed using the word embeddings for the relevant terms belonging to the topic. To determine the quality of classification using the new proposed topic-document embeddings, we apply two algorithms, i.e., Logistic Regression and Long Short-Term Memory, for the task of sentiment analysis. We build the TOPICDOC2VEC using five word embedding models, i.e., Word2Vec with CBOW and SkipGram, FastText with CBOW and Skip-Gram, and GloVe. The classification task is evaluated using accuracy, precision, and recall. For our empirical validation, we use a large labeled corpus containing game reviews. The experimental results prove that the TOPICDOC2VEC embeddings outperform the DOC2VEC embeddings for detecting the polarity of a document.