Search

Multi-label text classification (MLTC) refers to the problem of dealing with textual data classification based on multiple labels or tags. There are numerous real-world scenarios where the need for assigning labels to a particular object arises, such that the labels are descriptive of the properties of that object. However, in real life, it is not uncommon for one object to hold more than a singular property describing itself, hence it needs multiple labels to be associated with it. In the cases of textual data, one such scenario is assigning labels to a movie describing the genre, which needs more than one genre to specify the plot in a practical scenario. This makes movie genre prediction the desired choice for multi-label classification in many kinds of literature. This paper explores and presents an in-depth analysis of the approach of solving movie-genre prediction problems using the sequential model with universal sentence encoder (USE) for text encoding, alongside the use of label powerset (LP) as the problem transformation approach. Along with that, a comparative analysis of the performance of the model with different optimizers is performed. The best outcome achieved is f1-score 0.69 and accuracy of 0.89 with Adam optimizer, which, upon comparison with other literature of the similar domain, is either an equal or better in performance.

The use of electrical energy is directly proportional to the increase in global population, both concerning growing industrialization and rising residential demand. The need to achieve a balance between electrical energy production and consumption inspires researchers to develop forecasting models for optimal and economical energy use. Mostly, the residential and industrial sectors use metering sensors that only measure the consumed energy but are unable to manage electricity. In this paper, we present a comparative analysis of a variety of deep features with several sequential learning models to select the optimized hybrid architecture for energy consumption prediction. The best results are achieved using convolutional long short-term memory (ConvLSTM) integrated with bidirectional long short-term memory (BiLSTM). The ConvLSTM initially extracts features from the input data to produce encoded sequences that are decoded by BiLSTM and then proceeds with a final dense layer for energy consumption prediction. The overall framework consists of preprocessing raw data, extracting features, training the sequential model, and then evaluating it. The proposed energy consumption prediction model outperforms existing models over publicly available datasets, including Household and Korean commercial building datasets.

This paper presents an unsupervised approach that leverages raw aerial videos to learn to estimate planar homographic transformation between consecutive video frames. Previous learning-based estimators work on pairs of images to estimate their planar homographic transformations but suffer from severe over-fitting issues, especially when applying over aerial videos. To address this concern, we develop a sequential estimator that directly processes a sequence of video frames and estimates their pairwise planar homographic transformations in batches. We also incorporate a set of spatial-temporal knowledge to regularize the learning of such a sequence-to-sequence model. We collect a set of challenging aerial videos and compare the proposed method to the alternative algorithms. Empirical studies suggest that our sequential model achieves significant improvement over alternative image-based methods and the knowledge-rich regularization further boosts our system performance. Our codes and dataset could be found at https://github.com/Paul-LiPu/DeepVideoHomography, Accepted by 2020 25th International Conference on Pattern Recognition (ICPR) 2021

Handwritten recognition has received the greater attention in deep learning research community due to its vast applications and ambiguity in learning methods. CNN in deep learning is now becoming one of the most appealing approaches and has been a crucial factor in the variety of recent success and challenging machine learning applications such as object detection. This research paper is about the extended application of handwritten digit recognition, i.e., realtime detection of the handwritten digits. CNN is used as the model for the classification of the image and more specifically Keras sequential model is used as a classifier. The interface is created by Pygame. The layout of the interface is kept simple and is divided into two frames one for input and other for output. The image pre-processing is the most important step which has done with the help of OpenCV and Scipy. MNIST is the dataset used for training and testing. Handwritten digit recognition has various real-life time uses. It is used in the detection of vehicle number, banks for reading cheques, post offices for arranging letter, and many other tasks.

Graphs and networks are a crucial tool to model various phenomena in almost every branch of science. This book studies several different problems that are related to networks; the common theme among these problems is that they all describe some kind of an entity or effect that propagates though the network. The first part of the book focuses on majority/minority processes, when the actors in the network prefer to switch to the most/least common state in their neighborhood. These processes can model various situations in practice: they can represent the spreading of political opinions in a social network, the adoption of different social media platforms, the production strategies of rival companies, the frequency selection of devices in a wireless network, and many more. The book present several upper and lower bounds on the stabilization time of these processes in a general network topology. The second part studies financial networks: a system of banks that are interconnected by different kinds of debt contracts. If one of the banks goes bankrupt and cannot pay its liabilities, then this might cause other banks to go bankrupt as well, leading to a ripple effect through the whole system. There are countless interesting questions in this setting that are related to the properties of the underlying network, such as how regulators can enforce certain properties in these systems, how banks can execute different actions to improve their situation, or how the system behaves if default announcements are made in a step-by-step fashion. Finally, we conclude with a brief discussion of pebble games, where the propagation effect that we study is computation itself: the network represents the interdependencies of variables in a complex computational task, and a pebble game models our progress of computing more and more values in this graph until we arrive at the final result. The book assumes that readers have a strong background in theoretical computer science, but it requires no prior knowledge of the discussed network problems or the corresponding application areas.

Sequential model predictive control (SMPC) is an effective method to eliminate the weighting factors when two objectives or variables need to be controlled at the same time. However, a problem is that the cost function with multiple objectives or variables cannot operate on the sequential structure. A quadra-layer direct speed SMPC (Q-DS-SMPC) method for permanent magnet synchronous machines (PMSMs) rotor position is proposed in this paper. The rotor position control system with multiple control objectives and a switching vector table with 14 vectors have been applied to select the effective vectors in each model predictive control (MPC) process. Moreover, according to the tracking performances of rotor position and the importance of objectives, a sort of divided MPCs and a group of extracted values of cost function for limitation have been decided. Besides, a multiple layers SMPC principle and a selecting method for the cost functions in each divided MPC have been discussed in this paper. Simulation and experimental results corroborate the benefits of control performances and demonstrate the correctness of the theories and analyses.

IMPORTANCE: The sequential model emerged from the awareness that the persistence of residual symptoms and the frequent occurrence of psychiatric comorbidity were both associated with poor long-term outcome of major depressive disorder (MDD). OBJECTIVE: To conduct an updated meta-analysis to examine the association of the sequential combination of pharmacotherapy and psychotherapy with reduced risk of relapse and recurrence in MDD. DATA SOURCES: Keyword searches were conducted in PubMed, PsycInfo, Web of Science, and the Cochrane Library from inception of each database through November 2019. Reference lists from relevant studies were examined using the following keywords: sequential treatment, drugs and psychotherapy, combined treatment, continuation or maintenance, relapse or recurrence and prevention, and depress* or major depress*, selecting adults and randomized controlled trials as additional limits. Authors of selected articles were contacted if needed. STUDY SELECTION: Randomized clinical trials examining the effectiveness of the sequential use of psychotherapy following response to acute-phase pharmacotherapy in the treatment of adult remitted patients with MDD were selected independently by 2 reviewers. DATA EXTRACTION AND SYNTHESIS: The methods used fulfilled the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline. Data extraction and methodologic quality assessment were conducted independently by the reviewers. Examination of the pooled results was performed based on the random-effects model. Heterogeneity between study results and likelihood of significant publication bias were assessed. Sensitivity analyses and meta-regressions were also run. MAIN OUTCOMES AND MEASURES: The primary outcome measures were relapse or recurrence rates of MDD, as defined by study investigators, at the longest available follow-up. RESULTS: Seventeen randomized clinical trials met criteria for inclusion in the meta-analysis, with 1 study yielding 2 comparisons (2283 patients overall, with 1208 patients in a sequential treatment arm and 1075 in a control arm). The pooled risk ratio for relapse/recurrence of MDD was 0.84 (95% CI, 0.74-0.94), suggesting a relative advantage in preventing relapse/recurrence for the sequential combination of treatments compared with control conditions. CONCLUSIONS AND RELEVANCE: The results of this systematic review and meta-analysis indicate that the sequential integration of psychotherapy following response to acute-phase pharmacotherapy, alone or combined with antidepressant medication, was associated with reduced risk of relapse and recurrence in MDD. The preventive value of the sequential strategy relies on abatement of residual symptoms and/or increase in psychological well-being. The steps for implementing the sequential approach in remitted patients with recurrent MDD are provided.

A model when stated in simple terms is a mathematical equation, which is true when it implies to any model in machine learning including deep neural network. Every model will generate output for a given input, but important is to get output of desired accuracy. Machine learning models are trained on training data, and their best fit is judged on testing data. Before fitting the training data to the model to predict on unknown (test) data, pre-processing of data is essential to ensure model accuracy to acceptable level. This paper presents steps involved in pre-processing raw labelled dataset (Seattle weather) with 25,551 records (from year 1948 to 2017) to make it suitable for input to a deep neural network model. The data is split into 80% of training data and 20% of testing data. Scaling is performed on the data before it is passed to the deep neural network model. Deep neural network model that is multilayer perceptron model using sequential model API with dense layer is built and compiled using Adam optimizer resulting accuracy of 97.33% in predicting rainfall on a particular day.

This paper discusses applying different types of neural networks to classify a dataset of type audio. We used a GTZAN dataset that includes various audio music records representing different conventional categories of music genres. Each shares a set of common traditions; these traditions we call features. We build our proposed Python models using the Anaconda toolkit with TensorFlow (TF) an open-source deep-learning library. In our previous research, we build a multilayer sequential model to classify the dataset and then solve the overfitting issue in that model. In this paper, we build a Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) with Long Short Time Memory (LSTM). Finally, we compared the results to know the capabilities and limitations of Deep Learning (DL). CNN outperformed the other models in terms of training and test accuracy, having 83.74% and 74%, respectively.

This paper is an introduction to how hybridization is performed on a sequential model namely TrellisNet that affects its accuracy. TrellisNet is a temporal convolution Network with a special architecture. An attempt is made to combine this network with the Convolution Layer to improve Trellis Net performance. For this task, here EMNIST, which is an extension of MNIST that contains English handwritten letters and digits is used. The goal is to demonstrate the efficiency and performance of the current combined model. This also gives us an insight of performance this model as compared to other popular models used for Handwritten character recognition. TrellisNet with generalized weight matrices is a truncated Recurrent Network. Thus, a Convolution Layer followed by the defined TrellisNet is designed for achieving the targeted results aimed in this paper. This stacked architecture has delivered encouraging results. The results establish the characteristics of the model and show the problems this model can address.

Learning different subjects to enhance knowledge of students and children, reading habit plays an important role. Students often face problems or reading difficulties are aroused when the students are non-native English learners or suffering from dyslexia. Thus, in the present work, we have built a hybrid sequential model for text simplification (i.e. to translate complex English sentences into simple English sentences). Generally, text simplification is treated as monolingual translation task, in which translation occurs between the same languages. In recent trends, encoder-decoder model plays an important role in text simplification. However, a difference exists between normal machine translation and text simplification. Text simplification can be achieved using other different operations like splitting, merging and removing complex word, etc. It is very difficult to deal with the type of operation (i.e. split, delete or replace) that is required for text simplification if we use encoder decoder model. Our encoder-decoder model is mainly based on two different approaches, one is character-based and another is word-based. We also employed a Named Entity Recognizer (NER) to improve the accuracy of our model because named entities do not change while converting a complex sentence into a simple sentence. For learning process, we also use bi-directional encoder-decoder model to improve the result. In the present attempt, we used a well-known text simplification dataset (i.e. PWKP). Our proposed model achieved 0.82 BLUE score using NER module and word-based bi-directional encoder-decoder model.

An enforcement mechanism monitors a reactive system for undesired behavior at runtime and corrects the system’s output in case it violates the given specification. In this paper, we study the enforcement problem for hyperproperties, i.e., properties that relate multiple computation traces to each other. We elaborate the notion of sound and transparent enforcement mechanisms for hyperproperties in two trace input models: 1) the parallel trace input model, where the number of traces is known a-priori and all traces are produced and processed in parallel and 2) the sequential trace input model, where traces are processed sequentially and no a-priori bound on the number of traces is known. For both models, we study enforcement algorithms for specifications given as formulas in universally quantified HyperLTL, a temporal logic for hyperproperties. For the parallel model, we describe an enforcement mechanism based on parity games. For the sequential model, we show that enforcement is in general undecidable and present algorithms for reasonable simplifications of the problem (partial guarantees or the restriction to safety properties). Furthermore, we report on experimental results of our prototype implementation for the parallel model.

Natural language processing (NLP) aims to analyze a large amount of natural language data. The NLP computes textual data via a set of data processing elements which is sequentially connected to a path data pipeline. Several data pipelines exist for a given set of textual data with various degrees of model accuracy. Instead of trying all the possible paths, such as random search or grid search to find an optimal path, we utilized the Bayesian optimization to search along with the space of hyper-parameters learning. In this study, we proposed a data pipeline selection for NLP using Sequential Model-based Optimization (SMBO). We implemented the SMBO for the NLP data pipeline using Hyperparameter Optimization (Hyperopt) library with Tree of Parzen Estimators (TPE) model and Adaptive Tree of Parzen Estimators (A-TPE) model for a surface model with expected improvement (EI) acquired function.

Background/Aims: There is growing interest in the use of adaptive designs to improve the efficiency of clinical trials. We apply a Bayesian decision-theoretic model of a sequential experiment using cost and outcome data from the ProFHER pragmatic trial. We assess the model’s potential for delivering value-based research. Methods: Using parameter values estimated from the ProFHER pragmatic trial, including the costs of carrying out the trial, we establish when the trial could have stopped, had the model’s value-based stopping rule been used. We use a bootstrap analysis and simulation study to assess a range of operating characteristics, which we compare with a fixed sample size design which does not allow for early stopping. Results: We estimate that application of the model could have stopped the ProFHER trial early, reducing the sample size by about 14%, saving about 5% of the research budget and resulting in a technology recommendation which was the same as that of the trial. The bootstrap analysis suggests that the expected sample size would have been 38% lower, saving around 13% of the research budget, with a probability of 0.92 of making the same technology recommendation decision. It also shows a large degree of variability in the trial’s sample size. Conclusions: Benefits to trial cost stewardship may be achieved by monitoring trial data as they accumulate and using a stopping rule which balances the benefit of obtaining more information through continued recruitment with the cost of obtaining that information. We present recommendations for further research investigating the application of value-based sequential designs.

In this paper, we discuss applying a neural network model to classify a dataset of type audio. We used a GTZAN dataset that includes different audio music records representing different conventional categories of music genres. Each shares a set of common traditions; these traditions we call it features. We build our proposed Python models using the Anaconda toolkit that has TensorFlow (TF), an open-source deep-learning library. We build a multilayer sequential model to classify the dataset and then try to solve the model’s overfitting issue. Results show that our proposed overfitting solution tends to increase Testing Accuracy equals 60.55% and reduce the Testing Error by 1.1726.

Neural Architecture Search (NAS) is the process of automating architecture engineering, searching for the best deep learning configuration. One of the main NAS approaches proposed in the literature, Progressive Neural Architecture Search (PNAS), seeks for the architectures with a sequential model-based optimization strategy: it defines a common recursive structure to generate the networks, whose number of building blocks rises through iterations. However, NAS algorithms are generally designed for an ideal setting without considering the needs and the technical constraints imposed by practical applications. In this paper, we propose a new architecture search named Pareto-Optimal Progressive Neural Architecture Search (POPNAS) that combines the benefits of PNAS to a time-accuracy Pareto optimization problem. POPNAS adds a new time predictor to the existing approach to carry out a joint prediction of time and accuracy for each candidate neural network, searching through the Pareto front. This allows us to reach a trade-off between accuracy and training time, identifying neural network architectures with competitive accuracy in the face of a drastically reduced training time.

The evaluation of descriptive answer scripts is a monotonous, time-intensive process and is prone to human errors and sometimes bias. As a result, a student’s career can be at stake. Hence, in order to mitigate the errors and efforts during answer script evaluation, an automated evaluation system is proposed to assign marks to the answer scripts correctly. Our proposed system, ‘Descriptive Indic Answer Script Evaluation using Deep Learning’ will provide the marks to an answer based on its content and correctness depending upon the model trained using natural language processing and deep learning techniques on the dataset containing some of the manually evaluated sample answers. In the proposed sequential model, the learning is done by the convolutional and LSTM cells. The input to the system is the extracted text from the answer script which after preprocessing was fed to the model for training. As there was no significant system available for Indic languages, the system will be very helpful for the education boards conducting examinations in Indic languages. In this paper, among the different Indic languages, we have focussed on Odia language as very minuscule amount of work has been done in this language with regard to any NLP task.

A recommendation system is a set of programs that utilize different methodologies for relevant item selection for the user. In recent years deep neural networks have been used heavily for improving recommendation quality in every domain. We describe a model for music recommendation system that uses the BERT (Bidirectional Encoder Representations from Transformers) model. In the past, other deep neural networks have been used for music recommendation, which capture the the unidirectional sequential nature of a user’s data. Unlike other sequential techniques of recommendation, BERT uses bidirectional training of a user’s sequence for better recommendation. BERT uses the encoder part of the Transformer model, which uses an attention mechanism to learn contextual relations between a user’s past interactions. The proposed model relies on a user’s previous interaction to determine the bidirectional encoding for the model, which considers both the left and the right contexts. We evaluated our model with a baseline deep sequential model using two different datasets, and comparative results show that the model outperforms other sequential models.

In this paper, we presents a new mathematical formula for COVID-19 spread virus based on polynomial regression. Although this statistical estimation aims to fits a nonlinear model to the observed data vector (from Covid-19 data series provided from the World Health Organization (WHO)). Moreover, a survey based on the WHO's covid 19 symptoms and recommandations with ten(10) chosen questions is developed, in addition, the person's age is also taken into consideration. It aimed at constituting COVID-19 database from volunteers responses using the OCSS (Open COVID-19 Survey System), that is designed with Qt5 and coded with Python on Ubuntu system. Chosen individuals displacements parameter in the developed formula is based on the best fit correlation from the model and lifestyle state [number of walking per day (confined / unconfined)]. considering the reduced number of volunteers dataset (100), in order to enlarge the Learning database, we generate a 1000 synthetic random response score from OCSS system. The 100 real responses constitute the test database. Finally, using python Keras Sequential model Tensorflow backends, an opensource deep learning frameworks, to classify OCSS data and give better accuracy with the obtained model of 99.10 %, Loss:0.03. Test phase gives a high accuracy of 98.00 %, Loss:0.08. We propose a nearly four months (until Oct 14) forecasting based on the obtained model and proposed formula.

In Collaborative Filtering methods, tailored recommendations cannot be obtained when the user-item matrix is sparse (i.e., has low user-item interactions such as item ratings or purchases). Conventional recommendation systems (ChoiRec12, HPCRec18, HSPRec19) utilizing mining techniques such as clustering, frequent and sequential pattern mining along with click and purchase similarity measures for item recommendation cannot perform well when the user-item interactions are less, as the number of items keep increasing rapidly. Additionally, they have not explored the integration of semantic information of products extracted from customers' purchase histories into the item matrix and the pattern mining process. To address this problem, this paper proposes (SEMSRec) which integrates semantic information of E-commerce products extracted from purchase histories into all phases of recommendation process (pre-processing, pattern mining and recommendation). This is achieved by i) learning semantic similarities between items from customers' purchase histories using prod2vec model, ii) leveraging this information to mine semantically rich sequential purchase patterns and, iii) enriching the item matrix with semantic and sequential product purchase information before applying item based collaborative filtering. Thus, SEMSRec can provide Top-K personalized recommendations based on semantic similarities between items without the need for users' ratings on items. Experimental results on publically available E-commerce data set show that SEMSRec provides more relevant recommendations over other existing methods.

Decision Making forms the basis of any successful athlete's career. The importance of making the right decision at crucial junctures during a game is eventually the difference between winning or losing a match. Team sports have an important characteristic in that they are usually a set of non-discrete events and a good decision by one athlete, doesn't necessarily lead to a reward. It is a sequence of good decisions that finally lead to the reward, which explains why goals are usually so rare in a game of Soccer. Through this paper, we explore Soccer and look at the number of possibilities that are available at any given time to a player and what is their best possible action. We propose a Model-driven Domain specific Sequential model that aims to predict the best possible action that can be taken by a Soccer Athlete (player), to maximize the probability of scoring a goal.

Despite considerable progress in the development of rapid evaluation methods for physics-based reservoir model simulators there still exists a significant gap in acceleration and accuracy needed to enable complex optimization methods, including Monte Carlo and Reinforcement Learning. The latter techniques bear a great potential to improve existing workflows and create new ones for a variety of applications, including field development planning. Building on latest developments in modern deep learning technology, this paper describes an end-to-end deep surrogate model capable of modeling field and individual-well production rates given arbitrary sequences of actions (schedules) including varying well lo-cations, controls and completions. We focus on generalization properties of the surrogate model which is trained given a certain number of simulations. We study its spatial and time interpolation and extrapolation properties using the SPE9 case, followed by a validation on a large-scale real field. Our results indicate that the surrogate model achieves acceleration rates of about 15000x and 40000x for the SPE9 and the real field, respectively, incurring relative error ranging between 2% and 4% in the interpolation case, and between 5% and 12% in the various spacial and time extrapolation cases. These results provide concrete measures of the efficacy of the deep surrogate model as an enabling technology for the development of optimization techniques previously out of reach due to computational complexity.

Movement recognition is a hot issue in machine learning. The gesture recognition is related to video processing, which gives problems in various aspects. Some of them are separating the image against the background firmly. This problem has consequences when there are incredibly different settings from the training data. The next challenge is the number of images processed at a time that forms motion. Previous studies have conducted experiments on the Deep Convolutional Neural Network architecture to detect actions on sequential model balancing each other on frames and motion between frames. The challenge of identifying objects in a temporal video image is the number of parameters needed to do a simple video classification so that the estimated motion of the object in each picture frame is needed. This paper proposed the classification of hand movement patterns with the Single Stream Temporal Convolutional Neural Networks approach. This model was robust against extreme non-training data, giving an accuracy of up to 81,7%. The model used a 50 layers ResNet architecture with recorded video training.

Existing methods of level generation using latent variable models such as VAEs and GANs do so in segments and produce the final level by stitching these separately generated segments together. In this paper, we build on these methods by training VAEs to learn a sequential model of segment generation such that generated segments logically follow from prior segments. By further combining the VAE with a classifier that determines whether to place the generated segment to the top, bottom, left or right of the previous segment, we obtain a pipeline that enables the generation of arbitrarily long levels that progress in any of these four directions and are composed of segments that logically follow one another. In addition to generating more coherent levels of non-fixed length, this method also enables implicit blending of levels from separate games that do not have similar orientation. We demonstrate our approach using levels from Super Mario Bros., Kid Icarus and Mega Man, showing that our method produces levels that are more coherent than previous latent variable-based approaches and are capable of blending levels across games., Comment: 8 pages, 9 figures, 11th Workshop on Procedural Content Generation (PCG 2020)

To avoid the time-consuming weighting-factor tuning work in conventional finite control set model predictive control (FCS-MPC), a sequential model predictive control (SMPC) scheme for stand-alone voltage source inverters of an islanded ac microgrid is proposed in this paper. The main idea is that two cost functions for separately minimizing the capacitor voltage and the inductor current tracking errors of the $LC$ filter are deployed in a sequential structure instead of being integrated into a single cost function. As a result, the weighting factor for balancing the two control objectives is eliminated. Moreover, the proposed SMPC does not degrade the output-voltage control performance too much. Simulation results are provided, verifying the effectiveness of the presented approach.

In recent years, Photovoltaic System (PV) have been installed in parking lots in order to provide the green energy to Electric vehicles (EVs). Energy Synchronizing between PV generations and EVs demand is a function of different variables, and it is very challenging. Having an accurate prediction of PV generation helps to ease the complexity of this problem. Although various Machine Learning (ML) techniques have been applied and resulted well, traditional ML approaches need years of history of PV generations to make an accurate prediction. In many cases, parking lots or the houses recently equipped by PV panels, and this information is not available. Therefore, the primary motivation of this work is to build a reliable deep learning forecasting model based on Long Short Term Memory (LSTM) architecture in order to make a short-term prediction based on the limited previous observation. The proposed model is applied to a month of the PV power generation data and resulted in the promising accuracy with the Mean Absolute Percentage Error (MAPE) value of 0.028.

Electronic health records (EHR) are characterized as non-stationary, heterogeneous, noisy, and sparse data; therefore, it is challenging to learn the regularities or patterns inherent within them. In particular, sparseness caused mostly by many missing values has attracted the attention of researchers, who have attempted to find a better use of all available samples for determining the solution of a primary target task through the defining a secondary imputation problem. Methodologically, existing methods, either deterministic or stochastic, have applied different assumptions to impute missing values. However, once the missing values are imputed, most existing methods do not consider the fidelity or confidence of the imputed values in the modeling of downstream tasks. Undoubtedly, an erroneous or improper imputation of missing variables can cause difficulties in modeling as well as a degraded performance. In this study, we present a novel variational recurrent network that (i) estimates the distribution of missing variables allowing to represent uncertainty in the imputed values, (ii) updates hidden states by explicitly applying fidelity based on a variance of the imputed values during a recurrence (i.e., uncertainty propagation over time), and (iii) predicts the possibility of in-hospital mortality. It is noteworthy that our model can conduct these procedures in a single stream and learn all network parameters jointly in an end-to-end manner. We validated the effectiveness of our method using the public datasets of MIMIC-III and PhysioNet challenge 2012 by comparing with and outperforming other state-of-the-art methods for mortality prediction considered in our experiments. In addition, we identified the behavior of the model that well represented the uncertainties for the imputed estimates, which indicated a high correlation between the calculated MAE and the uncertainty., 11 pages, 4 figures

Sequential model-based design of experiments (MBDoE) uses information from previous experiments to select run conditions for new experiments. Computation of the objective functions for popular MBDoE can be impossible due to a non-invertible Fisher Information Matrix (FIM). Previously, we evaluated a leave-out (LO) approach that design experiments by removing problematic model parameters from the design process. However, the LO approach can be computationally expensive due to its iterative nature and some model parameters are ignored. In this study, we propose a simple Bayesian approach that makes the FIM invertible by accounting for prior parameter information. We compare the proposed Bayesian approach to the LO approach for designing sequential A-optimal experiments. Results from a pharmaceutical case study show that the Bayesian approach is superior, on average, to the LO approach for design of experiments. However, for subsequent parameter estimation, a subset-selection-based LO approach gives better parameter values than the Bayesian approach.

Spring phenology influences terrestrial ecosystem carbon, water and energy exchanges between the biosphere and atmosphere. Accurate prediction of spring phenology is therefore a prerequisite to foresee the impacts of climate warming on terrestrial ecosystems. In the present study, we studied the model performance of four widely used process-based models of spring leaf unfolding, including both a one-phase model (not considering a chilling phase: the Thermal Time model) and three two-phase models (all accounting for a required chilling period: the Parallel model, the Sequential model, the Unified model). Models were tested on five deciduous tree species occurring across Europe. We specifically investigated the divergence of their phenology predictions under future climate warming scenarios and studied the differences in the chilling periods. We found that, in general, the Iwo-phase models performed slightly better than the one-phase model when fitting to the observed data, with all two-phase models performing similarly. However, leaf unfolding projections diverged substantially among the two-phase models over the period 2070-2100. furthermore, we found that the modeled end dates of the chilling periods in these models also diverged, with advances for both the Sequential and Parallel models during the period 2070-2100 (compared to the period 1980-2010), and delays in the Unified model. These findings thus highlight large uncertainty in the two-phase phenology models and confirm that the mechanism underlying the leaf unfolding process is not yet understood. We therefore urgently need an improved understanding of the leaf unfolding process in order to improve the representation of phenology in terrestrial ecosystem models. (C) 2020 Elsevier B.V. All rights reserved.

A large amount of work has been done on the KDD 99 dataset, most of which include the use of a hybrid anomaly and misuse detection model done in parallel with each other. In order to further classify the intrusions, our approach to network intrusion detection includes the use of two different anomaly detection models followed by misuse detection applied to the combined output obtained from the previous step. The end goal of this is to verify the anomalies detected by the anomaly detection algorithm and clarify whether they are actually intrusions or random outliers from the trained normal (and thus to try and reduce the number of false positives). We aim to detect a pattern in this novel intrusion technique itself, and not the handling of such intrusions. The intrusions were detected to a very high degree of accuracy.

International audience; Sequential model-based optimization (SMBO) approaches are algorithms for solving problems that require computationally or otherwise expensive function evaluations. The key design principle of SMBO is a substitution of the true objective function by a surrogate, which is used to propose the point(s) to be evaluated next. SMBO algorithms are intrinsically modular, leaving the user with many important design choices. Significant research efforts go into understanding which settings perform best for which type of problems. Most works, however, focus on the choice of the model, the acquisition function, and the strategy used to optimize the latter. The choice of the initial sampling strategy, however, receives much less attention. Not surprisingly, quite diverging recommendations can be found in the literature. We analyze in this work how the size and the distribution of the initial sample influences the overall quality of the efficient global optimization (EGO) algorithm, a well-known SMBO approach. While, overall, small initial budgets using Halton sampling seem preferable, we also observe that the performance landscape is rather unstruc-tured. We furthermore identify several situations in which EGO performs unfavorably against random sampling. Both observations indicate that an adaptive SMBO design could be beneficial, making SMBO an interesting test-bed for automated algorithm design. CCS CONCEPTS • Computing methodologies → Continuous space search.

General-purpose automated algorithm configuration procedures have enabled impressive improvements in the state of the art in solving challenging problems from AI, operations research and other areas. The most successful configurators combine multiple techniques to search vast combinatorial spaces of parameter settings for a given algorithm as efficiently as possible. Specifically, two of the most prominent general-purpose algorithm configurators, SMAC and irace, can be seen as combinations of Bayesian optimisation and racing, and of racing and an estimation of distribution algorithm, respectively. Here, we investigate an approach that combines all three of these techniques into one single configurator, while exploiting prior knowledge contained in expert-chosen default parameter values. We demonstrate significant performance improvements over irace and SMAC on a broad range of running time optimisation scenarios from AClib.

Coarse-grained modeling, machine learning, and optimization enable efficient design of polymers with target properties., The chemical design of polymers with target structural and/or functional properties represents a grand challenge in materials science. While data-driven design approaches are promising, success with polymers has been limited, largely due to limitations in data availability. Here, we demonstrate the targeted sequence design of single-chain structure in polymers by combining coarse-grained modeling, machine learning, and model optimization. Nearly 2000 unique coarse-grained polymers are simulated to construct and analyze machine learning models. We find that deep neural networks inexpensively and reliably predict structural properties with limited sequence information as input. By coupling trained ML models with sequential model-based optimization, polymer sequences are proposed to exhibit globular, swollen, or rod-like behaviors, which are verified by explicit simulations. This work highlights the promising integration of coarse-grained modeling with data-driven design and represents a necessary and crucial step toward more complex polymer design efforts.

Moments of uncertainty are common for learners when practicing a second language. The appropriate management of these events could help avoid the development of frustration and benefit the learner's experience. Therefore, its detection is crucial in language practice conversations. In this study, an experimental conversation between an adult second language learner and a social robot is employed to visually characterize the learners' uncertainty. The robot's output is manipulated in prosody and lexical levels to provoke uncertainty during the conversation. These reactions are then processed to obtain Facial Action Units (AUs) and Gaze features. Preliminary results show distinctive behavioral patterns of uncertainty among the participants. Based on these results, a new annotation scheme is proposed, which will expand the data used to train sequential models to detect uncertainty. As future steps, the robotic conversational partner will use this information to adapt its behavior in dialogue generation and language complexity. QC 20230208

This work presents a Sequential Model Predictive Voltage Control of packed u cell (PUC) for grid connected applications. PUC inverter has become a research of interest due to the simultaneous advantages of flying capacitor and cascaded H bridge with less component count. Grid connected PUC provides the challenge of simultaneous control of link voltage and load current. Model Predictive Control provides superior multi objective, multi constraint handling and simplicity in implementation. The principle of model predictive control is about minimizing the cost function to get an optimized switching vector for every sample. Sequential model predictive control is a weight factor less algorithm to simultaneously control any number of control variables. However, the performance highly depends on the priority and the switch redundancy of the power converter. Performance analysis of the sequential MPC under various operating conditions has been presented for a PUC inverter. MATLAB®/ SimPowerSystems has been used to perform the simulations and performance has been compared with the existing weight factors based model predictive control algorithm.

Visually impaired people face several problems in their daily life. One of the biggest problems is to visiting unfamiliar places and identifying public places like pharmacy store, restrooms, pedestrian signs on roads, etc. Although there are some conventional methods that are available to aid visually impaired people but these are inefficient to use without assistance. The proposed method presents a framework which will help visually impaired people to identify the common public amenities while visiting any unfamiliar places. This method uses deep learning for recognizing some daily used places. For this purpose, VGG16 model is used to extract features from the images and train the sequential model. The model has been tested on varying images of different class that are present in the database. The developed algorithm achieves an accuracy of 95.88%. The obtained result of the developed model shows that it is an efficient method for assisting visually impaired people in real time application.

Mobility prediction plays an important role in a wide range of location-based applications and services. However, two important challenges are not well addressed in existing literature: 1) explicit high-order interactions of spatio-temporal features are not systemically modeled; 2) most existing algorithms place attention mechanisms on top of recurrent network, so they can not allow for full parallelism and are inferior to self-attention for capturing long-range dependence. To this end, we propose MoveNet, a self-attention based sequential model, to predict each user’s next destination based on her most recent visits and historical trajectory. MoveNet first introduces a cross based learning framework for modeling feature interactions. With self-attention on both the most recent visits and historical trajectory, MoveNet can use an attention mechanism to capture user’s long-term regularity in a more efficient and effective way. We evaluate MoveNet with three real-world mobility datasets, and show that MoveNet outperforms the state-of-the-art mobility predictor by around 10% in terms of accuracy, and simultaneously achieves faster convergence and over 4x training speedup.

A dataset of cherry full bloom dates across the full diversity of Australian growing regions was compiled and utilised for the first time. The primary data source was from growers located across Australia's major cherry growing districts. Records were of varying length and were used to investigate the potential of using data that substitute 'space-for-time' in phenology assessments. Full bloom timing for three cultivars, Bing, Lapins and Van, were collated. The data showed variation in full bloom time across the sites as well as inter-annual variability within sites. This highlighted the potential benefit of a predictive model for growers to better manage the flowering period. The performance of the sequential and the chill overlap models were evaluated using this data. Both models resulted in good statistical representation of the data with RMSE of 3.2 to 5.3 days and 3.3 to 5.4 days for the training and validating data, respectively. The parameterisations of the models differed with the sequential model estimating higher chilling requirement (65-68 CP) than the chill overlap model (46-48 CP). The sequential model estimated similar heat requirements between the cultivars, with Van marginally higher (7250-8000 GDH). The chill overlap model estimated a larger difference in the minimum heat requirement with Van higher (8223 GDH) than Bing and Lapins (6262 and 6486 GDH, respectively). Collation of grower records proved to be a valuable data source which, in the absence of funded programs, could be better utilised for further research that aligns outcomes with industry needs, such as decision support around polliniser varieties, design of growing systems and use of dormancy breakers to synchronise flowering for optimal fruit set and quality.