INESC-ID

About: INESC-ID is a nonprofit organization based out in Lisbon, Portugal. It is known for research contribution in the topics: Field-programmable gate array & Control theory. The organization has 932 authors who have published 2618 publications receiving 37658 citations.

L2F/INESC-ID at SemEval-2019 Task 2: Unsupervised Lexical Semantic Frame Induction using Contextualized Word Representations

Abstract: Building large datasets annotated with semantic information, such as FrameNet, is an expensive process. Consequently, such resources are unavailable for many languages and specific domains. This problem can be alleviated by using unsupervised approaches to induce the frames evoked by a collection of documents. That is the objective of the second task of SemEval 2019, which comprises three subtasks: clustering of verbs that evoke the same frame and clustering of arguments into both frame-specific slots and semantic roles. We approach all the subtasks by applying a graph clustering algorithm on contextualized embedding representations of the verbs and arguments. Using such representations is appropriate in the context of this task, since they provide cues for word-sense disambiguation. Thus, they can be used to identify different frames evoked by the same words. Using this approach we were able to outperform all of the baselines reported for the task on the test set in terms of Purity F1, as well as in terms of BCubed F1 in most cases.

Cascaded H-bridge multilevel inverter with a fault detection scheme based on the statistic moments indexes

Abstract: Multilevel inverters allow to generate AC voltages with low total harmonic distortion (THD) but requires an increased number of power switches. One of the disadvantages of that is the increased probability of a fault in one of the power switches. Thus in order to improve the reliability of the converter a fast and robust fault detection scheme must be used. In this context this paper presents a new fault detection scheme based on the AC voltages of the inverter. The proposed scheme uses fault factors that are based on the statistic moment method. This method was used for the identification of image patterns. Due to this the proposed scheme allows for a robust detection to load or voltage variations. To test the method was used a cascaded H-bridge inverter with five levels. Several tests were performed through numerical results. The use of a laboratory prototype was also used to confirm the proposed scheme.

Cache-Oblivious parallel SIMD Viterbi decoding for sequence search in HMMER

Abstract: HMMER is a commonly used bioinformatics tool based on Hidden Markov Models (HMMs) to analyze and process biological sequences. One of its main homology engines is based on the Viterbi decoding algorithm, which was already highly parallelized and optimized using Farrar’s striped processing pattern with Intel SSE2 instruction set extension. A new SIMD vectorization of the Viterbi decoding algorithm is proposed, based on an SSE2 inter-task parallelization approach similar to the DNA alignment algorithm proposed by Rognes. Besides this alternative vectorization scheme, the proposed implementation also introduces a new partitioning of the Markov model that allows a significantly more efficient exploitation of the cache locality. Such optimization, together with an improved loading of the emission scores, allows the achievement of a constant processing throughput, regardless of the innermost-cache size and of the dimension of the considered model. The proposed optimized vectorization of the Viterbi decoding algorithm was extensively evaluated and compared with the HMMER3 decoder to process DNA and protein datasets, proving to be a rather competitive alternative implementation. Being always faster than the already highly optimized ViterbiFilter implementation of HMMER3, the proposed Cache-Oblivious Parallel SIMD Viterbi (COPS) implementation provides a constant throughput and offers a processing speedup as high as two times faster, depending on the model’s size.

Optimization of Area and Delay at Gate-Level in Multiple Constant Multiplications

Abstract: Although many efficient high-level algorithms have been proposed for the realization of Multiple Constant Multiplications (MCM) using the fewest number of addition and subtraction operations, they do not consider the low-level implementation issues that directly affect the area, delay, and power dissipation of the MCM design. In this paper, we initially present area efficient addition and subtraction architectures used in the design of the MCM operation. Then, we propose an algorithm that searches an MCM design with the smallest area taking into account the cost of each operation at gate-level. To address the area and delay tradeoff in MCM design, the proposed algorithm is improved to find the smallest area solution under a delay constraint. The experimental results show that the proposed algorithms yield low-complexity and high-speed MCM designs with respect to those obtained by the prominent algorithms designed for the optimization of the number of operations and the optimization of area at gate-level.

Global meta-analysis of transcriptomics studies.

Abstract: Transcriptomics meta-analysis aims at re-using existing data to derive novel biological hypotheses, and is motivated by the public availability of a large number of independent studies. Current methods are based on breaking down studies into multiple comparisons between phenotypes (e.g. disease vs. healthy), based on the studies' experimental designs, followed by computing the overlap between the resulting differential expression signatures. While useful, in this methodology each study yields multiple independent phenotype comparisons, and connections are established not between studies, but rather between subsets of the studies corresponding to phenotype comparisons. We propose a rank-based statistical meta-analysis framework that establishes global connections between transcriptomics studies without breaking down studies into sets of phenotype comparisons. By using a rank product method, our framework extracts global features from each study, corresponding to genes that are consistently among the most expressed or differentially expressed genes in that study. Those features are then statistically modelled via a term-frequency inverse-document frequency (TF-IDF) model, which is then used for connecting studies. Our framework is fast and parameter-free; when applied to large collections of Homo sapiens and Streptococcus pneumoniae transcriptomics studies, it performs better than similarity-based approaches in retrieving related studies, using a Medical Subject Headings gold standard. Finally, we highlight via case studies how the framework can be used to derive novel biological hypotheses regarding related studies and the genes that drive those connections. Our proposed statistical framework shows that it is possible to perform a meta-analysis of transcriptomics studies with arbitrary experimental designs by deriving global expression features rather than decomposing studies into multiple phenotype comparisons.