Ensemble of Support Vector Machine and Ontological Structures to Generate Abstractive Text Summarization

Achieving at least some level of explainability requires complex analyses for many machine learning systems, such as common black-box models. We recently proposed a new rule-based learning system, SupRB, to construct compact, interpretable and transparent models by utilizing separate optimizers for the model selection tasks concerning rule discovery and rule set composition. This allows users to specifically tailor their model structure to fulfil use-case specific explainability requirements. From an optimization perspective, this allows us to define clearer goals and we find that—in contrast to many state of the art systems—this allows us to keep rule fitnesses independent. In this paper we investigate this system’s performance thoroughly on a set of regression problems and compare it against XCSF, a prominent rule-based learning system. We find the overall results of SupRB’s evaluation comparable to XCSF’s while allowing easier control of model structure and showing a substantially smaller sensitivity to random seeds and data splits. This increased control can aid in subsequently providing explanations for both training and final structure of the model.KeywordsRule-based learningLearning classifier systemsEvolutionary machine learningInterpretable modelsExplainable AI

A system for extracting and comparing memes in online forums

This paper proposes an automatic fire detection system that is able to work with both stationary and non-stationary video achieving a 91.90% accuracy using color, texture and shape information. As an extension to prior approaches in the video-based fire detection field, the proposed system replaces the use of threshold-driven color spectroscopy with a trained classifier, resulting in a better color segmentation. From the color-classified regions, non-temporal texture features are extracted and used as an input to a trained Random Forests classifier. Using standardized metrics and a large dataset we compare our solution to other state of the art systems, showing that it outperforms the prior approaches in both frame by frame classification and fire region location. The method has been validated over more than 24 thousand frames proving to offer high precision and recall.

In this paper, we present a hybrid approach for performing token and sentence levels Dialect Identification in Arabic. Specifically we try to identify whether each token in a given sentence belongs to Modern Standard Arabic (MSA), Egyptian Dialectal Arabic (EDA) or some other class and whether the whole sentence is mostly EDA or MSA. The token level component relies on a Conditional Random Field (CRF) classifier that uses decisions from several underlying components such as language models, a named entity recognizer and and a morphological analyzer to label each word in the sentence. The sentence level component uses a classifier ensemble system that relies on two independent underlying classifiers that model different aspects of the language. Using a featureselection heuristic, we select the best set of features for each of these two classifiers. We then train another classifier that uses the class labels and the confidence scores generated by each of the two underlying classifiers to decide upon the final class for each sentence. The token level component yields a new state of the art F-score of 90.6% (compared to previous state of the art of 86.8%) and the sentence level component yields an accuracy of 90.8% (compared to 86.6% obtained by the best state of the art system).

Due to rising DC-Voltages and voltage slew rates (du/dt) partial discharge (PD) phenomena move into the focus for low voltage traction drives. State of the art systems are designed to be PD-free at any point of operation. Elevated temperatures yield decreased partial discharge inception voltages (PDIV). In the design process, this is considered by measurements at rated operation employing safety factors which predict the PDIV at higher temperatures. Traction drives in electric vehicles have an increasing electrical and thermal utilization to increase power density. To allow for a more precise design process for these machines, studies on the temperature dependence of PDIV are performed and discussed in this paper.

Ontology alignment plays a critical role in knowledge integration and has been widely investigated in the past decades. State of the art systems, however, still have considerable room for performance improvement especially in dealing with new (industrial) alignment tasks. In this paper we present a machine learning based extension to traditional ontology alignment systems, using distant supervision for training, ontology embedding and Siamese Neural Networks for incorporating richer semantics. We have used the extension together with traditional systems such as LogMap and AML to align two food ontologies, HeLiS and FoodOn, and we found that the extension recalls many additional valid mappings and also avoids some false positive mappings. This is also verified by an evaluation on alignment tasks from the OAEI conference track.

Modern CPUs have evolved from bulky die size and low power into high powered machines that requires a need for significantly improved thermal interface material to conduct heat from the die surfaces to the Cu-heat spreader. Automotive applications of these CPUs requires a high performance characteristics of TIMs since environment is extremely harsh and reliability of the materials becomes a primacy for the operation of the vehicle. TIM to Copper interfaces becomes one of the primary locations of failure and act as forerunner to the breakdown of the whole state of the art system. Assessment of reliability of end applicants requires an understanding of damage progression of the TIM-Copper interfaces as a function of operating temperature and times. Our study exposes the TIM-Copper interfaces to high temperature and then the interface fracture toughness is quantified. A four-point bend composite bend was fabricated and then the sample was kept at high temperature of 100°C, 150°C for aging time 10 days, 30 days, 60 days and 120 days. Quasi-static bending loads was used to observe and determine interfacial delamination of the sample specimen. A 2D- digital image correlation method was also employed to understand the progression of crack growth and the crack tip opening displacement (CTOD) to equate the degradation of various TIM interfaces at different aging schedules.

With the advent of various smart devices for IoT smart home environment, there has been an immense interest in services related to predicting the recurrent patterns in the form of user behavior. The connected environment of various digital devices inside a smart home provides an opportunity to figure out the interaction in the form of patterns and present it to the end user with some semantics. Those patterns are then recommended with an aim to have high rate of acceptance. But with change in the user's routine and preferences, the system needs to be dynamic and adaptive accordingly. The proposed system aims to solve this problem by implementing a highly scalable custom Feedback Learner Framework through a Feedback Learner Service which analyses user's interaction with the suggested automations and related devices. As state of the art systems have a limitation of adapting to the user's behavior in such short time period, our system comes to the rescue and increase the chances of user adaptation for various IoT recommendation services. This framework is scenario agnostic and can be utilized by multiple services based on the pub-sub model. This also helps in prioritizing automations and rectifying frequency (increase/decrease) of automations delivered to them. We validate our system through various traditional methods to make it more robust for quick and unexpected changes in the smart home environment.

In state of the art systems, workload scheduling and server fan speed operate independently leading to cooling inefficiencies. In this work we propose GentleCool, a proactive multi-tier approach for significantly lowering the fan cooling costs without compromising the performance. Our technique manages the fan speed through intelligently allocating the workload across different machines. The experimental results show our approach delivers average cooling energy savings of 72% and improves the mean time between failures (MTBF) of the fans by 2.3X compared to the state of the art.

Abstract

Many problems in natural language processing (NLP) can be formulated as classification problems. The complexity of the natural language make it difficult to select the discriminating attributes for a given task, the reliability of attribute values is often dubious and vary with the domain of the corpus. This article argues that the bayesian network formalism is well adapted for the classification of this type of data. We have estimated the benefit brought by the bayesian classifior on a real NLP application: the distinction between the impersonal and anaphoric occurrences of the english pronoun 'it'. After having implemented various state of the art systems able to perform this distinction, we have integrated them within a single classifior. Each system has been tested on two different domain corpora. The results obtained by our classifior are significantly better than those of the state of the art systems.

Melanoma lesion detection and segmentation using deep region based convolutional neural network and fuzzy C-means clustering

It is widely accepted that in a text, sentences and clauses cannot be understood in isolation but in relation with each other through discourse relations that may or may not be explicitly marked. Discourse relations have been found useful in many applications such as machine translation, text summarization, and question answering; however, they are often not considered in computational language applications because domain and genre independent robust discourse parsers are very few. In this paper, we analyze existing approaches to identify five discourse relations automatically (namely, comparison, contingency, illustration, attribution, and topic-opinion), and propose a new approach to identify attributive relations. We evaluate the accuracy of each approach with respect to the discourse relations it can identify and compare it to a human gold standard. The evaluation results show that the state of the art systems are rather effective at identifying most of the relations considered, but other relations such as attribution are still not identified with high accuracy.KeywordsMachine TranslationSentiment AnalysisDependency RelationParse TreeDiscourse RelationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

In state of the art systems, workload scheduling and server fan speed operate independently leading to cooling inefficiencies. In this work we propose GentleCool, a proactive multi-tier approach for significantly lowering the fan cooling costs without compromising the performance. Our technique manages the fan speed through intelligently allocating the workload across different machines. The experimental results show our approach delivers average cooling energy savings of 72% and improves the mean time between failures (MTBF) of the fans by 2.3X compared to the state of the art.

Centrifugally Tensioned Metastable Fluid Detector systems (CTMFDs) have a number of valuable advantages over other conventional, state of the art systems (e.g., <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> He tubes). CTMFDs can be configured to attain intrinsic efficiencies over 90% for neutron energies from the thermal to the fast region (ev to MeV). TMFDs can detect alpha and fission recoil interactions exceeding 10 times lower activity than that of conventional spectrometers (i.e., <; ~0.05 Bq/g). The tension pressures used for detection (-10 bars) are over 1000 times greater than the fluctuations resulting from even extreme external events such as during earthquakes. TMFDs are also inherently gamma blind making them ideal for use in high gamma fields where traditional detectors fail such as in active interrogation applications. Unlike current state of the art active and passive interrogation systems, the CTMFD relies on simple to use, straightforward, and inexpensive electronics with lab-prototypes costing around ~$100-1000. Finally, the CTMFD system can switch between detection of neutrons, alpha recoils, and fission products with simplicity within the same system. As a novel transformational advancement, the CTMFD has been re-configured to now allow for multiple detectors enclosed within the envelope of a single system, resulting in the Multi-Arm Centrifugally Tensioned Metastable Fluid Detectors (MAC-TMFDs). This system embodiment now allows for the creation of several independently operating sensing regions within a single TMFD. In this way, a single detector can effectively be converted into multiple situation-specific sensors within a simple package. This advancement allows for rapid neutron/alpha spectroscopy with a single system. As part of capability validation, alpha spectroscopy has been performed with a two sensitive volume region apparatus. This system's sensitivity shows significant improvement over state-of-the-art liquid scintillation counters ~ 1 to 10 times more sensitive than a Beckman LS6500 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TM</sup> spectrometer. Also, gamma blind neutron detection using (n, alpha) Pu-Be and fission <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">252</sup> Cf neutron sources have been possible to attain along with discrimination of each source. The resulting detection data has been shown to remain compatible with the underlying science of a traditional CTMFD system. Further analysis shows that the leap-ahead MAC-TMFD is amenable for on-demand scalability.