Archive

This talk will present the challenges for the research area known as aspect-based sentiment analysis. Sentiment Analysis (or Opinion Mining) is a field of study in Natural Language Processing that aims to extract and classify the sentiment orientation present in texts. Aspect-based Sentiment Analysis (ABSA) is a particular study of sentiment analysis focusing in the extraction of opinions directed to a specific target or aspect. ABSA...

Sign languages are the main way of communication in the Deaf community and with the listening population. There are about 70 million deaf people and over 200 distinct sign languages in the world. Unfortunately, not all deaf know a sign language, and many cannot read or write in a spoken language. Moreover, when there is prelingual deafness, a sign language becomes the children's native language,...

Water utilities can operate their systems at higher efficiencies when they are able to predict water consumption in the short and long terms. Baseform develops forward-thinking software for networked water infrastructures, including applications that use real time flow monitoring records to analyse demand patterns and forecast their behaviours. In this talk, we show how statistical methodologies can be a crucial aid in understanding human behaviour. Several...

In this seminar I will present my work on trajectory optimization for robot motion planning in Reproducing Kernel Hilbert Spaces (RKHSs). Functional gradient algorithms are a popular choice for motion planning in complex many-degree-of-freedom robots. They work by directly optimizing a continuous trajectory that avoids obstacles while maintaining geometric properties such as smoothness. We exploit this fact and propose a functional gradient based method under RKHSs. This...

The convex envelope of a function is its best convex approximation. Convex envelopes are important because they permit to approximate hard nonconvex optimization problems by easy convex ones. However, creating convex envelopes is more of an art than a science. In this talk, I reveal a trick for creating convex envelopes and illustrate with examples in signal reconstruction with sparse, gaussian, and markov priors. Those...

This thesis's purpose is to generate a movie tribute in the form of a videoclip for a given movie and music. A videoclip is considered to be a video containing the most important clips from the movie and a music playing in the background. The most important clips are collected by summarizing the movie subtitles with Support Sets algorithm. It is important that the artifact...

Scientific talks range from conference presentations to lectures and documentaries. Even though some formats have high-grade content, such as lectures, their presentation is not always the most appealing. On the other hand, documentaries and other entertainment-driven programs can present consolidated information benefiting from aesthetically assured visual and audible effects. Motivated by these aspects, we illustrate the topics of a physics lecture through the use of...

Every day we generate large amounts of data. Just by going online or using our cell phones, we leave informative traces, the so called “digital breadcrumbs”, which can give away a lot of information about our individual actions, with obvious threats to privacy. However, when this individual data is gathered and analyzed collectively, it can be very revealing of global behavior patterns and, at the...

In natural language processing problems, we often encounter the situation where we don't have enough data in the language we're interested to build a model, but there is data for the same problem in another language. There has been a line of research into devising multilingual models, transferring information from resource-rich languages into resource-poor languages. One way to do this cross-lingual transfer would be to...

In some real-world applications of machine learning, despite the investments in the training of classification systems and in feature selection, misclassifications occur and their effects are critical. This is common in ill-posed classification problems, where overlapping classes, small or incomplete training sets, and unknown classes are prevalent. We can mitigate misclassifications and their effects by adapting the behavior of the classifier on samples with high...