Abstract The limitations on the scalability of computer systems imposed by the dark‐silicon effects are so severe that they support the extensive use of heterogeneity such as the GP‐GPU for general purpose processing. Performance simulators of GP‐GPU heterogeneous systems aim to provide performance accuracy at the cost of execution time. In this work, we handle time‐consuming simulations of design space exploration systems based on GPUs. We have developed performance predictors based on machine learning (ML) algorithms and evaluated them in accuracy and throughput (number of predictions per second). We measure model accuracy through the mean absolute percentage error (MAPE) and the model efficiency through a throughput metric (millions of predictions per second). Our experiments revealed that decision trees predictors are the most promising regarding accuracy and efficiency. We applied the best predictors into the MultiExplorer, a dark silicon‐aware design space exploration tool that allows designers to explore the architecture and microarchitecture of multicore/manycore system design.

In software projects, applications are often monitored by systems that automatically identify crashes, collect their information into reports, and submit them to developers. Especially in popular applications, such systems tend to generate a large number of crash reports in which a significant portion of them are duplicate. Due to this high submission volume, in practice, the crash report deduplication is supported by devising automatic systems whose efficiency is a critical constraint. In this paper, we focus on improving deduplication system throughput by speeding up the stack trace comparison. In contrast to the state-of-the-art techniques, we propose FaST, a novel sequence alignment method that computes the similarity score between two stack traces in linear time. Our method independently aligns identical frames in two stack traces by means of a simple alignment heuristic. We evaluate FaST and five competing methods on four datasets from open-source projects using ranking and binary metrics. Despite its simplicity, FaST consistently achieves state-of-the-art performance regarding all metrics considered. Moreover, our experiments confirm that FaST is substantially more efficient than methods based on optimal sequence alignment.

Bug tracking systems (BTS) are widely used in software projects. An important task in such systems consists of identifying duplicate bug reports, i.e., distinct reports related to the same software issue. For several reasons, reporting bugs that have already been reported is quite frequent, making their manual triage impractical in large BTSs. In this paper, we present a novel deep learning network based on soft-attention alignment to improve duplicate bug report detection. For a given pair of possibly duplicate reports, the attention mechanism computes interdependent representations for each report, which is more powerful than previous approaches. We evaluate our model on four well-known datasets derived from BTSs of four popular open-source projects. Our evaluation is based on a ranking-based metric, which is more realistic than decision-making metrics used in many previous works. Achieved results demonstrate that our model outperforms state-of-the-art systems and strong baselines in different scenarios. Finally, an ablation study is performed to confirm that the proposed architecture improves the duplicate bug reports detection.

When breaking news occurs, stock quotes can change abruptly in a matter of seconds. The human analysis of breaking news can take several minutes, and investors in the financial markets need to make quick decisions. Such challenging scenarios require faster ways to support investors. In this work, we propose the use of bidirectional encoder representations from transformers BERT to perform sentiment analysis of news articles and provide relevant information for decision making in the stock market. This model is pre-trained on a large amount of general-domain documents by means of a self-learning task. To fine-tune this powerful model on sentiment analysis for the stock market, we manually labeled stock news articles as positive, neutral or negative. This dataset is freely available and amounts to 582 documents from several financial news sources. We fine-tune a BERT model on this dataset and achieve 72.5% of F-score. Then, we perform some experiments highlighting how the output of the obtained model can provide valuable information to predict the subsequent movements of the Dow Jones Industrial (DJI) Index.

Unsupervised domain adaptation is an attractive option when labeled data is lacking for some domain of interest but is available for other domain. Part-of-speech (POS) tagging is often considered a solved task when enough labeled data is available in the domain of interest. However, when considering a domain adaptation scenario, this is far from true. Several approaches have been proposed for domain adaptation of POS taggers, however as far as we know, all of them are based on handcrafted features. In this work, we employ a machine learning method whose input is exclusively composed of the raw text. This method learns word- and character-level representations (embeddings), and has been successfully applied to intra-domain tasks. We show that this method achieves strong performances on the domain adaptation of English and Portuguese POS taggers.

We describe a structure learning system for unrestricted coreference resolution that explores two key modeling techniques: latent coreference trees and automatic entropy-guided feature induction. The latent tree modeling makes the learning problem computationally feasible because it incorporates a meaningful hidden structure. Additionally, using an automatic feature induction method, we can efficiently build enhanced nonlinear models using linear model learning algorithms. We present empirical results that highlight the contribution of each modeling technique used in the proposed system. Empirical evaluation is performed on the multilingual unrestricted coreference CoNLL-2012 Shared Task datasets, which comprise three languages: Arabic, Chinese and English. We apply the same system to all languages, except for minor adaptations to some language-dependent features such as nested mentions and specific static pronoun lists. A previous version of this system was submitted to the CoNLL-2012 Shared Task closed track, achieving an official score of 58.69, the best among the competitors. The unique enhancement added to the current system version is the inclusion of candidate arcs linking nested mentions for the Chinese language. By including such arcs, the score increases by almost 4.5 points for that language. The current system shows a score of 60.15, which corresponds to a 3.5% error reduction, and is the best performing system for each of the three languages.