Sense representations for Portuguese: experiments with sense embeddings and deep neural language models

Sense representations have gone beyond word representations like Word2Vec, GloVe and FastText and achieved innovative performance on a wide range of natural language processing tasks. Although very useful in many applications, the traditional approaches for generating word embeddings have a strict drawback: they produce a single vector representation for a given word ignoring the fact that ambiguous words can assume different meanings. In this paper, we explore unsupervised sense representations which, different from traditional word embeddings, are able to induce different senses of a word by analyzing its contextual semantics in a text. The unsupervised sense representations investigated in this paper are: sense embeddings and deep neural language models. We present the first experiments carried out for generating sense embeddings for Portuguese. Our experiments show that the sense embedding model (Sense2vec) outperformed traditional word embeddings in syntactic and semantic analogies task, proving that the language resource generated here can improve the performance of NLP tasks in Portuguese. We also evaluated the performance of pre-trained deep neural language models (ELMo and BERT) in two transfer learning approaches: feature based and fine-tuning, in the semantic textual similarity task. Our experiments indicate that the fine tuned Multilingual and Portuguese BERT language models were able to achieve better accuracy than the ELMo model and baselines.

     

Mid‐infrared spectroscopy as a tool for real‐time monitoring of ethanol absorption in glycols

Fast, simple, accurate, and inexpensive methods for obtaining analyte concentration data are desirable in the industrial sector. In the present study, the use of Fourier transform mid‐infrared (FT‐MIR) spectroscopy, combined with partial least squares (PLS) regression, was investigated as a tool for real‐time monitoring of processes of ethanol absorption in glycols. Calibration was performed using simple synthetic samples containing ethanol, water, and monoethylene glycol (MEG) or diethylene glycol (DEG). The PLS models presented excellent performance, with correlation coefficients (R²) close to unity and root‐mean‐square errors of cross‐validation (RMSECV) and prediction (RMSEP) lower than 2% of the calibration data ranges for both analytes (ethanol and water) in both absorbents (MEG and DEG). The monitoring technique developed has potential to be applied in absorption processes and could also be used in other large‐scale unit operations, providing information in real time and enhancing process control.     

Evaluation of the removal of n-butanol vapor by the poly(lactic acid)-zeolite-TiO2 composite and formation of by-products

The use of polymeric films incorporated with zeolite-TiO₂ composites associated with UV radiation can be an alternative in the removal of volatile organic compounds (VOCs) through the adsorption and photodegradation processes. This study produced poly(lactic acid) (PLA) films incorporated with 13× zeolite, TiO₂, and 13×-TiO₂ zeolite composite to remove n-butanol and evaluate the by-products generated in the process. The results showed that 13× zeolite and TiO₂ added individually or as a composite to PLA, gave the polymer matrix a significant increase in the removal capacity of n-butanol. The best performance was presented by the zeolite-TiO₂, composite, confirming a synergistic effect. However, the formation of CO and CO₂ exceeded the expected values, with the verification that the polymeric matrix underwent photodegradation action by TiO₂. The polymeric film only containing zeolite is the most suitable for the removal of VOCs, as it did not present degradation of the PLA, generating a lower concentration of by-products.     

Modelling and analysis of non-cooperative peer-assisted VoD streaming in managed networks

The growing popularity of the Video on Demand service in the Internet Protocol Television environments and the demand for increased quality of the offered videos are becoming a serious threat for the service providers because the high amounts of video traffic are causing congestion in the delivery networks. One of the most acceptable approaches to solve this issue is the peer-assisted streaming, where the peers participate in the streaming process in order to alleviate the load on the streaming servers and in the core of the network. Although the reliability of the Peer-to-Peer service is considerably improved in the managed networks because of the control that the operators have over the clients’ Set-Top Boxes, the failures of the peers still cannot be completely eliminated. The operator can take advantage of the streaming and storage resources of the clients and use them for peer-assisted streaming only while they are watching a video, but not after they finish the streaming session because they may turn off their receiving devices until the next session. In this chapter, we address the issue of the failures of the peers in such environments and their influence on the traffic requested from the servers for providing uninterrupted video experience. For that purpose, we propose a precise mathematical tool for modelling a peer-assisted system for Video on Demand streaming in managed networks with non-cooperative peers, which may decide not to share their resources while they are not active. This tool calculates the performance of the system taking into consideration large variety of system parameters, including the failure probability and the time the peers spend until they decide to turn on the STB and join the network. As the results from the simulations verify the correctness of the mathematical model, we use it to analyse how the failures of the peers are affecting the system’s performance for different system parameters.     

Augmented reality system to assist inexperienced pool players

Computational Visual Media - Pool and billiards are amongst a family of games played on a table with six pockets along the rails. This paper presents an augmented reality tool designed to assist...     

AmbiKraf

This paper presents, AmbiKraf, a non-emissive fabric display that subtly animates patterns on common fabrics. We use thermochromic inks and peltier semiconductor elements to achieve this technology. With this technology we have produced numerous prototypes from animated wall paintings to pixilated fabric displays. The ability of this technology to subtly and ubiquitously change the color of the fabric itself has made us able to merge different fields and technologies with AmbiKraf. In addition, with an animated room divider screen, Ambikraf merged its technology with Japanese Byobu art to tighten the gap between traditional arts and contemporary technologies. Through this AmbiKraf Byobu art installation and other installations, we discuss the impact of this technology as a ubiquitous fabric display. With focus to improvements of some limitations of the existing system, we present our future vision that enables us to merge this technology into more applications fields thus making this technology a platform for ubiquitous interactions on our daily peripherals.     

Variable Structure Control with Chattering Reduction of a Generalized T‐S Model

In this paper, a fuzzy logic controller (FLC) based variable structure control (VSC) is presented. The main objective is to obtain an improved performance of highly non‐linear unstable systems. New functions for chattering reduction and error convergence without sacrificing invariant properties are proposed. The main feature of the proposed method is that the switching function is added as an additional fuzzy variable and will be introduced in the premise part of the fuzzy rules; together with the state variables. In this work, a tuning of the well known weighting parameters approach is proposed to optimize local and global approximation and modelling capability of the Takagi‐Sugeno (T‐S) fuzzy model to improve the choice of the performance index and minimize it. The main problem encountered is that the T‐S identification method can not be applied when the membership functions are overlapped by pairs. This in turn restricts the application of the T‐S method because this type of membership function has been widely used in control applications. The approach developed here can be considered as a generalized version of the T‐S method. An inverted pendulum mounted on a cart is chosen to evaluate the robustness, effectiveness, accuracy and remarkable performance of the proposed estimation approach in comparison with the original T‐S model. Simulation results indicate the potential, simplicity and generality of the estimation method and the robustness of the chattering reduction algorithm. In this paper, we prove that the proposed estimation algorithm converge the very fast, thereby making it very practical to use. The application of the proposed FLC‐VSC shows that both alleviation of chattering and robust performance are achieved.     

A neural network-based scheme coupled with the RPV model inversion package

A hybrid scheme for the inversion of the Rahman-Pinty-Verstraete (RPV) model is presented. It combines the inversion technique described by Lavergne et al. (Lavergne, T., Kaminski, T., Pinty, B., Taberner, M., Gobron, N., Verstraete, M.M., Vossbeck, M., Widlowski, J.L., Giering, R. (2007). Application to MISR land products of an RPV model inversion package using adjoint and Hessian codes. Remote Sensing of Environment, 107, 362-375.) and a multilayer backpropagation feedforward neural network. The RPV inversion package is applied to a sample set of pixels within the satellite scene. Subsequently the pairs of bidirectional reflectance factors (BRF) and model parameters estimated from the sample set of pixels are used to train the neural network. Since the mathematical formulation of the RPV model is embedded in these training data variables, the neural network can efficiently retrieve the model parameters for the whole satellite scene. This scheme has been tested for a MISR L2 BRF scene, MISR L1B2-derived BRF data corresponding to two different dates and a mosaic of MISR L2 BRF scenes acquired over Southern Africa covering a large extent of Miombo woodland. The results show this strategy retrieves the RPV model parameters and uncertainties with high accuracy and considerable speed over large areas.