Suboptimal Model Research Articles

Model primitives for hierarchical lifelong reinforcement learning

Learning interpretable and transferable subpolicies and performing task decomposition from a single, complex task is difficult. Some traditional hierarchical reinforcement learning techniques enforce this decomposition in a top-down manner, while meta-learning techniques require a task distribution at hand to learn such decompositions. This paper presents a framework for using diverse suboptimal world models to decompose complex task solutions into simpler modular subpolicies. This framework performs automatic decomposition of a single source task in a bottom up manner, concurrently learning the required modular subpolicies as well as a controller to coordinate them. We perform a series of experiments on high dimensional continuous action control tasks to demonstrate the effectiveness of this approach at both complex single task learning and lifelong learning. Finally, we perform ablation studies to understand the importance and robustness of different elements in the framework and limitations to this approach.

DATA IMBALANCE IN LANDSLIDE SUSCEPTIBILITY ZONATION: UNDER-SAMPLING FOR CLASS-IMBALANCE LEARNING

Abstract. Machine learning methods such as artificial neural network, support vector machine etc. require a large amount of training data, however, the number of landslide occurrences are limited in a study area. The limited number of landslides leads to a small number of positive class pixels in the training data. On contrary, the number of non-landslide pixels (negative class pixels) are enormous in numbers. This under-represented data and severe class distribution skew create a data imbalance for learning algorithms and suboptimal models, which are biased towards the majority class (non-landslide pixels) and have low performance on the minority class (landslide pixels).In this work, we have used two algorithms namely EasyEnsemble and BalanceCascade for balancing the data. This balanced data is used with feature selection methods such as fisher discriminant analysis (FDA), logistic regression (LR) and artificial neural network (ANN) to generate LSZ maps The results of the study show that ANN with balanced data has major improvements in preparation of susceptibility maps over imbalanced data, where as the LR method is ill-effected by data balancing algorithms. The FDA does not show significant changes between balanced and imbalanced data.

Fuzzy Hysteresis Smoothing: A New Approach for Image Denoising

Hysteresis smoothing (HS) is one of the recently proposed image denoising schemes, which employs hard threshold levels to model the hysteresis process. Nevertheless, hard thresholding is in contrast with what is observed from the behavior of the hysteresis phenomenon in nature, and this implies that the HS may lead to a suboptimal model. On the other hand, it is proved that fuzzy logic is a promising research field in solving various problems by applying the uncertainty characteristics. Hence, in this article, we develop a novel HS methodology based on the fuzzy norms, which, in addition to incorporating the advantages of the HS, also manages the undesired effects of hard thresholding. In this method, which is called fuzzy HS (FHS), pointwise hard thresholding is replaced by an interval soft manner, which allows the threshold levels to be determined commensurate with the fuzzy norm's free parameter. It is shown that among the classical fuzzy norms, the Yager one can carry out the FHS relatively well. However, this norm causes oversaturation in the hysteresis loop, which leads to an oversmoothing problem in the image denoising process. To alleviate this, a new fuzzy norm with logarithmic nature, named expansion norm, has been proposed, which improves the accuracy of the FHS. The comparative simulations demonstrate the significant superiority of the FHS in terms of both objective and subjective criteria over the classical HS. Besides, the results indicate that the proposed scheme has competitive denoising performance in comparison with some well-known algorithms.

Attention-Based Pose Sequence Machine for 3D Hand Pose Estimation

Most of the existing methods for 3D hand pose estimation are performed from a single depth map. In that case, the depth missing challenges from input frames caused by hand self-occlusions and imaging quality lead to multi-valued mapping phenomenon and sub-optimal model. In this paper, we proposed a novel recurrent architecture named Attention-based Pose Sequence Machine (APSM) to alleviate challenges by introducing temporal consistency. As for recurrent unit (RU), we extend traditional Gated Recurrent Unit (GRU) with 3D convolutional neural networks (CNNs) to handle voxelized inputs and features, and a novel RU named Deep Gated Recurrent Unit (DGRU) was proposed by rebuilding deeper gates based on GRU. To improve the model performance, a novel spatial attention mechanism denoted as Attention Model (AM) was proposed. Ablation experiments are designed to validate each contribution of our work, and experiments on two publicly available dataset show that our work outperforms state-of-the-art on hand pose estimation.

Methods for preprocessing time and distance series data from personal monitoring devices

There is a need to develop more advanced tools to improve guidance on physical exercise to reduce risk of adverse events and improve benefits of exercise. Vast amounts of data are generated continuously by Personal Monitoring Devices (PMDs) from sports events, biomedical experiments, and fitness self-monitoring that may be used to guide physical exercise. Most of these data are sampled as time- or distance-series. However, the inherent high-dimensionality of exercise data is a challenge during processing. As a result, current data analysis from PMDs seldomly extends beyond aggregates.Common challanges are:•alterations in data density comparing the time- and the distance domain;•large intra and interindividual variations in the relationship between numerical data and physiological properties;•alterations in temporal statistical properties of data derived from exercise of different exercise durations.These challenges are currently unresolved leading to suboptimal analytic models. In this paper, we present algorithms and approaches to address these problems, allowing the analysis of complete PMD datasets, rather than having to rely on cumulative statistics. Our suggested approaches permit effective application of established Symbolic Aggregate Approximation modeling and newer deep learning models, such as LSTM.

Generation of a PARK2 homozygous knockout induced pluripotent stem cell line (GIBHi002-A-1) with two common isoforms abolished.

Loss of function mutations in PARK2 (encoding PARKIN) cause autosomal recessive Parkinson's disease (PD), which often manifests at a juvenile age. Molecular and biochemical studies show that PARKIN functions as an E3 ubiquitin ligase controlling mitochondrial homeostasis. Yet, the exact mechanisms are unclear due to the use of sub-optimal models including cancer cells and fibroblasts. We have generated a PARK2 knockout (KO) isogenic cell line using a well-characterized induced pluripotent stem cell (iPSC) clone with good differentiation potential. This cell line lacks the expression of all PARKIN isoforms and is valuable for elucidating the role of PARK2 mutations in PD.

Economic Implications of Hepatocellular Carcinoma Surveillance and Treatment: A Guide for Clinicians.

The incidence of hepatocellular carcinoma (HCC) is increasing worldwide, with significant morbidity and associated costs. Treatment allocation depends on the stage of diagnosis; however, resource utilization can be significant across all stages. We aimed to summarize the available data on the cost effectiveness of surveillance of and treatments for HCC in the context of current treatment guidelines. We performed a focused review of studies investigating the economic burden and cost effectiveness of HCC surveillance treatment modalities published between January 2000 and January 2019. The overall economic burden of HCC is increasing in the USA and in several countries worldwide due to its rising incidence and the proliferation of therapies. Liver transplantation is a cost-effective strategy for early-stage HCC treatment in selected patients. In settings where liver transplantation is not available or in patients awaiting transplant, ablative or locoregional therapies are cost effective with increases in quality-adjusted life-years. First-line therapy with sorafenib for advanced stage HCC is cost effective in the treatment of compensated cirrhosis. The cost effectiveness of recently approved systemic therapies for advanced HCC require further investigation. Existing studies have shown that guideline-recommended surveillance techniques and several available therapies for the treatment of HCC are cost effective; however, there are limitations in the literature, including reliance on suboptimal modeling with incomplete/simplified model structure or inadequate inputs. With increasing therapeutic options in patients with HCC, understanding their relative value is critical in designing HCC treatment algorithms.

Conclusions from a behavioral aging study on male and female F2 hybrid mice on age-related behavior, buoyancy in water-based tests, and an ethical method to assess lifespan

Due to strain-specific behavioral idiosyncrasies, inbred mouse strains are suboptimal research models for behavioral aging studies. The aim of this study is to determine age-related behavioral changes of F2 hybrid C57BL/6NxBALB/c male and female mice. Lifespan was followed (nmales=48, nfemales=51) and cohorts of mature adult (7 months), middle-aged (15 months), and old mice (22 months of age; n=7-12 per group) were assessed regarding open-field activity, exploration, passive avoidance learning/memory, and depressive-like behavior. We found that both males and females demonstrated decreased exploratory behavior with age, while memory and depressive-like behavior were maintained. Females exhibited enhanced depressive-like behavior compared to males; however, a correlation between fat mass and swimming activity in the test directly accounted for 30-46% of this behavioral sex difference. In addition, we suggest a method to qualitatively estimate natural lifespan from survival analyses in which animals with signs of pain or severe disease are euthanized. This is, to our knowledge, the first behavioral study to consider both sex and aging in hybrid mice. We here define decreased exploratory behavior as a conserved hallmark of aging independent of sex, highlight the effect of buoyancy in water tests, and provide a method to assay lifespan with reduced animal suffering.

The Economic Development of Urban Planning in China

To study the enlightenment of development economics on the economic development of urban planning, firstly, the research background and significance of the enlightenment of development economics on the economic development of urban planning are introduced. Secondly, through the main method of literature review, social investigation method, theoretical model analysis method and so on, combined with relevant theory of economics and related data analysis, it is found that economic theory can effectively explain the problems existing in the urban planning, avoid possible problems in urban planning, and provide effective theory basis for the development of the city. In this study, the theoretical value of the sub-optimal theoretical model, the supply and demand analysis model and the marginal analysis model under the pareto optimal state in the economic development of urban planning is mainly studied. A large number of theories and research results confirm that to do a good job in urban planning must be guided by relevant economic theories, which will make the progress of urban planning more rapid.

Parallel Traversal of Large Ensembles of Decision Trees

Machine-learnt models based on additive ensembles of regression trees are currently deemed the best solution to address complex classification, regression, and ranking tasks. The deployment of such models is computationally demanding: to compute the final prediction, the whole ensemble must be traversed by accumulating the contributions of all its trees. In particular, traversal cost impacts applications where the number of candidate items is large, the time budget available to apply the learnt model to them is limited, and the users’ expectations in terms of quality-of-service is high. Document ranking in web search, where sub-optimal ranking models are deployed to find a proper trade-off between efficiency and effectiveness of query answering, is probably the most typical example of this challenging issue. This paper investigates multi/many-core parallelization strategies for speeding up the traversal of large ensembles of regression trees thus obtaining machine-learnt models that are, at the same time, effective, fast, and scalable. Our best results are obtained by the GPU-based parallelization of the state-of-the-art algorithm, with speedups of up to 102.6x.

Machine learning for downscaling: the use of parallel multiple populations in genetic programming

In the implementation of traditional GP algorithm as models are evolved in a single deme (an environment in which a population of models is evolved) it may tend to produce sub-optimal models with poor generalisation skills due to lack of model diversity. As a solution to above issue, in this study the potential of evolving models in parallel multiple demes with different genetic attributes (parallel heterogeneous environments) and subsequent further evolution of some of the fittest models selected from each deme in another deme called the master deme was investigated, in relation to downscaling of large-scale climate data to daily minimum temperature (Tmin) and daily maximum temperature (Tmax). It was discovered that independent of the climate regime (i.e. warm or cold) and the geographic location of the observation station, a fraction of the fittest models (e.g. 25%) obtained from the last generation of each deme alone are sufficient for the formulation of a diverse initial population of models for the master deme. Also, independent of the climate regime and the geographic location of the observation station, both daily Tmin and Tmax downscaling models developed with the parallel multi-population genetic programming (PMPGP) algorithm showed better generalisation skills compared to that of models developed with the traditional single deme GP, even when the amount of redundant information in the data of predictors was high. The models developed for daily Tmin and Tmax with the PMPGP algorithm simulated fewer unphysically large outliers compared to that of models developed with the GP algorithm.

Mapping dead forest cover using a deep convolutional neural network and digital aerial photography

Tree mortality is an important forest ecosystem variable having uses in many applications such as forest health assessment, modelling stand dynamics and productivity, or planning wood harvesting operations. Because tree mortality is a spatially and temporally erratic process, rates and spatial patterns of tree mortality are difficult to estimate with traditional inventory methods. Remote sensing imagery has the potential to detect tree mortality at spatial scales required for accurately characterizing this process (e.g., landscape, region). Many efforts have been made in this sense, mostly using pixel- or object-based methods. In this study, we explored the potential of deep Convolutional Neural Networks (CNNs) to detect and map tree health status and functional type over entire regions. To do this, we built a database of around 290,000 photo-interpreted trees that served to extract and label image windows from 20 cm-resolution digital aerial images, for use in CNN training and evaluation. In this process, we also evaluated the effect of window size and spectral channel selection on classification accuracy, and we assessed if multiple realizations of a CNN, generated using different weight initializations, can be aggregated to provide more robust predictions. Finally, we extended our model with 5 additional classes to account for the diversity of landcovers found in our study area. When predicting tree health status only (live or dead), we obtained test accuracies of up to 94%, and up to 86% when predicting functional type only (broadleaf or needleleaf). Channel selection had a limited impact on overall classification accuracy, while window size increased the ability of the CNNs to predict plant functional type. The aggregation of multiple realizations of a CNN allowed us to avoid the selection of suboptimal models and help to remove much of the speckle effect when predicting on new aerial images. Test accuracies of plant functional type and health status were not affected in the extended model and were all above 95% for the 5 extra classes. Our results demonstrate the robustness of the CNN for between-scene variations in aerial photography and also suggest that this approach can be applied at operational level to map tree mortality across extensive territories.

Measuring the Pancreatic β Cell Mass in Vivo with Exendin SPECT during Hyperglycemia and Severe Insulitis.

Targeting the glucagon-like peptide-1 receptor with radiolabeled exendin is a very promising method to noninvasively determine the β cell mass in the pancreas, which is needed to unravel the pathophysiology of type 1 and type 2 diabetes. The present study aimed to explore the effects of both hyperglycemia and insulitis on the uptake of exendin in a spontaneous type 1 diabetes mouse model, nonobese diabetic (NOD) mice. NOD mice (n = 75, 7-21 weeks old) were injected intravenously with [111In]In-DTPA-exendin-3, and single-photon emission computed tomography (SPECT) images were acquired 1 h pi. The pancreatic accumulation of [111In]In-DTPA-exendin-3 was quantified in vivo using SPECT and by ex vivo counting and correlated to the β cell mass (BCM). The influence of insulitis and hyperglycemia on the exendin uptake was assessed. The pancreas could be visualized longitudinally using SPECT. A linear correlation was found between the BCM (%) and pancreatic uptake (%ID/g) as measured by ex vivo counting (Pearson r = 0.64, p < 0.001), which was not affected by either insulitis (Pearson r = 0.66, p = 0.83) or hyperglycemia (Pearson r = 0.57, p = 0.51). Biodistribution and ex vivo autoradiography revealed remaining [111In]In-DTPA-exendin-3 uptake in the pancreas despite total ablation of BCM. Despite hyperglycemia and severe insulitis, we have found a good correlation between BCM and pancreatic exendin uptake, even in a suboptimal model with relatively high background activity.

Improvable models as scaffolds for promoting productive disciplinary engagement in an engineering design activity

AbstractBackgroundThere is a need to find ways of productively engaging K‐12 students with engineering design. I investigate a new class of physical models, referred to as Improvable Models, as scaffolds for helping middle school students productively engage in engineering design practices.Purpose/HypothesisThe purpose of this study is to answer two questions: (a) How do students use Improvable Models? and (b) How does the use of Improvable Models as scaffolds support students' design of engineering solutions?Design/MethodDesign‐based research was conducted in a sixth‐grade classroom. Students used a type of Improvable Model, specifically the Suboptimal System (SS) seed model. Improvable Models make the entire design space accessible and enable iterative design optimization. An SS seed model represents inefficient design decisions and is used as an initiator for iterative design optimization. The study spanned 10 lessons, each 45 min long, totaling 8 days of instructional activity spread across 1 month. Data sources included student conversations, design work, and presentations. The productive disciplinary engagement framework was used to analyze the data.ResultsStudents used the seed model for productively engaging with five disciplinary practices: decomposing the design challenge into individual design parameters, considering the relationship between design parameters, reasoning through multiple design parameters and making trade‐offs, weighing multiple solutions, and using design heuristics for ideation and design optimization. The seed model scaffolded students' productive engagement in various disciplinary practices such as informed decision‐making and systematic exploration.ConclusionThis study provides an approach for designing learning environments for productively engaging K‐12 students in engineering design practices using a suboptimal model.

Abstract 90: The “reproducibility crisis” of animal studies in oncology - How did we get here, and how can we resolve it

Abstract Historically, a large majority of clinical phase III trials fail to gain regulatory approval due primarily to a lack of demonstrated efficacy, demonstrating the challenge of conducting clinically relevant efficacy studies in laboratory animals. In 2012, Amgen researchers published their finding that an astoundingly-high proportion (90%) of more than 50 republished in vivo studies in oncology was not reproducible. This sparked an intense debate about data validity in preclinical sciences, and several reviews have been published in the last two years analysing this issue. A number of best-practice initiatives and resources have since been created along with the development of animal study workflow software applications and databases to address the causes of this significant, global problem in cancer research. Major sources of irreproducibility in animal research such as bias, suboptimal disease models, lack of controls, and most importantly bad documentation practices will be discussed. Several initiatives have resulted in practically-applicable concepts and publicly-available resources to address this issue, such as ARRIVE guidelines, PHISPS protocols and the N3R Design Assistant, will be described. Electronic spreadsheets have been the most prevalently-used tool to plan and conduct preclinical efficacy studies in oncology. The advent of animal study workflow software and database technology, designed specifically for in vivo oncology research, has seen increasingly widespread adoption in recent years in both academia and industry. By employing standardized and more detailed approaches to the processes documenting highly detailed study design, data collection, method and procedure descriptions and adherence, real-time monitoring of animal welfare parameters, standardized graphing, data analyses and report generation, measurement and task scheduling, as well as data preservation approaches, will be explored. Using some real-life examples, the advantages and disadvantages of preclinical study automation and the use of electronic spreadsheets will be discussed in the context of helping to resolve the animal study “Reproducibility Crisis”. Citation Format: Hannes Hentze, Eric Ibsen. The “reproducibility crisis” of animal studies in oncology - How did we get here, and how can we resolve it [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 90.

Genetic Algorithm‐Based Variable Selection in Prediction of Hot Metal Desulfurization Kinetics

Sulfur is considered as one of the main impurities in hot metal and hot metal desulfurization is often carried out using injection of fine‐grade desulfurization reagent. The selection of variables used for predicting the course of hot metal desulphurization requires expert knowledge. However, it is difficult to model the complex interactions in the process and to evaluate a high number of possible variable subsets with manual variable selection techniques. As the amount of data gathered from the process increases, manual variable selection becomes too time‐consuming and might lead to a suboptimal prediction model. The objective of this work is to execute an automatic variable selection procedure for prediction of hot metal desulfurization based on an industrial scale data set. The variable selection problem is formulated as a constrained optimization problem, in which the objective function is formulated based on repeated leave‐multiple‐out cross‐validation. The implemented solution strategy is a binary‐coded genetic algorithm (GA). By making use of the developed model, the effect of the main production variables on the rate and efficiency of primary hot metal desulfurization is quantified. The variables related to properties of the reagent and the injection parameters were found to be of great importance.

The “Crisis of Reproducibility” of Animal Studies: Causes and Solutions

Abstract A majority of post-phase III clinical trial therapies have historically failed to receive regulatory approval due to a lack of demonstrated efficacy. According to a 2012 study published by Amgen researchers, 90% of the 50+ in vivo oncology studies were irreproducible. Over the last two years, questions and debates concerning the validity of the preclinical science data have resulted in multiple published reviews. Suboptimal disease models, inadequate or inappropriate controls, and poor documentation practices (including electronic spreadsheets) will be discussed as primary causes of irreproducibility in animal research. To address these concerns, several best-practice resources and initiatives have been made available to the research community. The ARRIVE guidelines, PHISPS protocols, and the N3R Design Assistant will be described as practically-applicable concepts and publicly-available resources to address issues contributing to the “Crisis of Reproducibility”. Researchers have increasingly adopted animal study workflow software applications and databases, which were developed to address this crisis. By applying such automated and time-saving technologies, researchers can now more easily standardize the highly variable but detailed processes of study design, procedure adherence and descriptions, animal welfare parameter monitoring, graphing, data analyses, report creation, task and measurement scheduling, and data preservation. Using some real-life examples, the advantages and disadvantages of preclinical study software applications and the use of electronic spreadsheets will be discussed in the context of helping to resolve the “Crisis of Reproducibility” in animal studies.

Latent Representation Learning for Alzheimer's Disease Diagnosis With Incomplete Multi-Modality Neuroimaging and Genetic Data.

The fusion of complementary information contained in multi-modality data [e.g., magnetic resonance imaging (MRI), positron emission tomography (PET), and genetic data] has advanced the progress of automated Alzheimer's disease (AD) diagnosis. However, multi-modality based AD diagnostic models are often hindered by the missing data, i.e., not all the subjects have complete multi-modality data. One simple solution used by many previous studies is to discard samples with missing modalities. However, this significantly reduces the number of training samples, thus leading to a sub-optimal classification model. Furthermore, when building the classification model, most existing methods simply concatenate features from different modalities into a single feature vector without considering their underlying associations. As features from different modalities are often closely related (e.g., MRI and PET features are extracted from the same brain region), utilizing their inter-modality associations may improve the robustness of the diagnostic model. To this end, we propose a novel latent representation learning method for multi-modality based AD diagnosis. Specifically, we use all the available samples (including samples with incomplete modality data) to learn a latent representation space. Within this space, we not only use samples with complete multi-modality data to learn a common latent representation, but also use samples with incomplete multi-modality data to learn independent modality-specific latent representations. We then project the latent representations to the label space for AD diagnosis. We perform experiments using 737 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, and the experimental results verify the effectiveness of our proposed method.

Perceived Target Range Shapes Human Sound-Localization Behavior.

The auditory system relies on binaural differences and spectral pinna cues to localize sounds in azimuth and elevation. However, the acoustic input can be unreliable, due to uncertainty about the environment, and neural noise. A possible strategy to reduce sound-location uncertainty is to integrate the sensory observations with sensorimotor information from previous experience, to infer where sounds are more likely to occur. We investigated whether and how human sound localization performance is affected by the spatial distribution of target sounds, and changes thereof. We tested three different open-loop paradigms, in which we varied the spatial range of sounds in different ways. For the narrowest ranges, target-response gains were highly idiosyncratic and deviated from an optimal gain predicted by error-minimization; in the horizontal plane the deviation typically consisted of a response overshoot. Moreover, participants adjusted their behavior by rapidly adapting their gain to the target range, both in elevation and in azimuth, yielding behavior closer to optimal for larger target ranges. Notably, gain changes occurred without any exogenous feedback about performance. We discuss how the findings can be explained by a sub-optimal model in which the motor-control system reduces its response error across trials to within an acceptable range, rather than strictly minimizing the error.

An optimized, clinically relevant mouse model of cisplatin-induced ototoxicity

Cisplatin-induced ototoxicity results in significant, permanent hearing loss in pediatric and adult cancer survivors. Elucidating the mechanisms underlying cisplatin-induced hearing loss as well as the development of therapies to reduce and/or reverse cisplatin ototoxicity have been impeded by suboptimal animal models. Clinically, cisplatin is most commonly administered in multi-dose, multi-cycle protocols. However, many animal studies are conducted using single injections of high-dose cisplatin, which is not reflective of clinical cisplatin administration protocols. Significant limitations of both high-dose, single-injection protocols and previous multi-dose protocols in rodent models include high mortality rates and relatively small changes in hearing sensitivity. These limitations restrict assessment of both long-term changes in hearing sensitivity and effects of potential protective therapies. Here, we present a detailed method for an optimized mouse model of cisplatin ototoxicity that utilizes a multi-cycle administration protocol that better approximates the type and degree of hearing loss observed clinically. This protocol results in significant hearing loss with very low mortality. This mouse model of cisplatin ototoxicity provides a platform for examining mechanisms of cisplatin-induced hearing loss as well as developing therapies to protect the hearing of cancer patients receiving cisplatin therapy.