Interpretable Representation Research Articles

Conceptual graph operations for formal visual reasoning in the medical domain

ObjectiveConceptual graphs (CGs) are used to represent clinical guidelines because they support visual reasoning with a logical background, making them a potentially valuable representation for guidelines. Materials and methodsConceptual graph formalism has an essential and basic component: a formal vocabulary that drives all of the other mechanisms, notably specialization and projection. The graph's theoretical operations, such as projection, rules, derivation, constraints, probabilities and uncertainty, support diagrammatic reasoning. ResultsA conceptual graph's graphical user interface includes a multilingual vocabulary management, some query and decision-making facilities and visual graph representations that are simple and interesting for user interactions. The described proposition using the Conceptual Graph user interface (CoGui) improves the performance of the actors in the diagnostic context of heart failure with preserved ejection fraction. DiscussionCGs capture the essential features of the medical processes underlying clinical reasoning. CGs are indeed useful as a way for the physician to represent guidelines, and well-defined semantic representations allow users to have a maximal understanding of the knowledge reasoning process. ConclusionCG operations of visual representations that uncover some of the actual complexities of clinicians’ reasoning have been tested in clinical guideline comprehension and used to translate text and diagrammatic guidelines into computer interpretable representations.

Hierarchical Adaptive Regression Kernels for Regression With Functional Predictors

We propose a new method for regression using a parsimonious and scientifically interpretable representation of functional predictors. Our approach is designed for data that exhibit features such as spikes, dips, and plateaus whose frequency, location, size, and shape varies stochastically across subjects. We propose Bayesian inference of the joint functional and exposure models, and give a method for efficient computation. We contrast our approach with existing state-of-the-art methods for regression with functional predictors, and show that our method is more effective and efficient for data that include features occurring at varying locations. We apply our methodology to a large and complex dataset from the Sleep Heart Health Study, to quantify the association between sleep characteristics and health outcomes. Software and technical appendices are provided in the online supplementary materials.

Fuzzy rules for describing subgroups from Influenza A virus using a multi-objective evolutionary algorithm

Extraction of biologically-meaningful knowledge is one of the important and challenging tasks in bioinformatics, in particular computational analysis of DNA and protein sequences, in order to identify biological function(s) and behaviour(s) of newly-extracted sequences. Computational intelligence techniques in corporation with sequence-driven features have been applied to tackle the problem and help classify different functional classes of the sequences. In order to study this problem, subgroup discovery algorithms together with a signal processing-based feature extraction method are applied, where the sequences are represented as a signal. The applicability of this method has been studied through four different Neuraminidase genes of Influenza A subtypes, H1N1, H2N2, H3N2 and H5N1. The results yielded not only higher predictive accuracy over these four classes of the proteins but also interpretable rule-based representation of the descriptive model with a significantly reduced feature set driven by means of the signal processing method. Subgroup discovery technique based on evolutionary fuzzy systems is expected to open new areas of research in bioinformatics and further help identify and understand more focused therapeutic protein targets.

Geometric steerable medial maps

To provide more intuitive and easily interpretable representations of complex shapes/organs, medial manifolds should reach a compromise between simplicity in geometry and capability of restoring the anatomy/shape of the organ/volume. Existing morphological methods show excellent results when applied to 2D objects, but their quality drops across dimensions. This paper contributes to the computation of medial manifolds from a theoretical and a practical point of view. First, we introduce a continuous operator for accurate and efficient computation of medial structures of arbitrary dimension. Second, we present a validation protocol for assessing the suitability of medial surfaces for anatomical representation in medical applications. We evaluate quantitatively the performance of our method with respect to existing approaches and show its higher performance for medical imaging applications in terms of medial simplicity and capability of reconstructing the anatomical volume.

Stylisation and representation in subtitles: can less be more?

This article considers film dialogues and interlingual subtitles from the point of view of linguistic and cultural representation, and revisits from that perspective the question of loss, as a platform for considering alternative views on the topic and broader theoretical issues. The cross-cultural pragmatics perspective and focus on viewers’ reactions that dealing with representation entails cast the question of loss in a different light and opens up avenues for alternative modes of analysis. They make room for subtitles to be construed as producing their own systems of multimodal textual representation and modes of interpretation, and for their text to be recognised as having a greater expressive and representational potential than face values might suggest. This is the argument, informed by Fowler's Theory of Mode (1991, 2000), that is taken up in the paper, and harnessed to the review of examples or observations from recent studies on subtitles, and complementary evidence from dubbing. The capacity of subtitles to produce insights into the cultures and languages represented is of particular interest, and has wider implications for the culturally instrumental functions of subtitles and translation strategies.

Where are possible worlds? (arguments for ReALIS)

We follow Pollard (2007) in assuming that the mainstream Kripke/Montagueinspired possible-worlds semantics is “a framework known to have dubious foundations” (primarily because of the granularity problem), and “worlds are constructed from propositions […], and not the other way around”. We intend to work out this approach in a DRT-based framework, called ReALIS, in order to account for phenomena concerning referent accessibility, at the same time. We claim that our system offers a general solution to problems of intensional identity, and it is devoid of DRT’s “extra level” problem—by embedding discourse representations in the world model, not directly but as parts of the representations of interpreters’ minds, i.e., their (permanently changing) information states/“internal worlds”. Hence, there is simply no intensionality in ReALIS as interpreters’ “worldlets” (in description of their brains within the entire model of the universe) carry all kinds of information (BDI, guesswork, dream) typically “entrusted to” possible worlds.

An integrated data collection and analysis framework for remote monitoring and planning of construction operations

Recent advances in data collection and operations analysis techniques have facilitated the process of designing, analyzing, planning, and controlling of engineering processes. Mathematical tools such as graphical models, scheduling techniques, operations research, and simulation have enabled engineers to create models that represent activities, resources, and the environment under which a project is taking place. Traditionally, most simulation paradigms use static or historical data to create computer interpretable representations of real engineering systems. The suitability of this approach for modeling construction operations, however, has always been a challenge since most construction projects are unique in nature as every project is different in design, specifications, methods, and standards. Due to the dynamic nature and complexity of most construction operations, there is a significant need for a methodology that combines the capabilities of traditional modeling of engineering systems and real time field data collection. This paper presents the requirements and applicability of a data-driven modeling framework capable of collecting and manipulating real time field data from construction equipment, creating dynamic 3D visualizations of ongoing engineering activities, and updating the contents of a discrete event simulation model representing the real engineering system. The developed framework can be adopted for use by project decision-makers for short-term project planning and control since the resulting simulation and visualization are completely based on the latest status of project entities.

Fast Inference for the Latent Space Network Model Using a Case-Control Approximate Likelihood

Network models are widely used in social sciences and genome sciences. The latent space model proposed by Hoff et al. (2002), and extended by Handcock et al. (2007) to incorporate clustering, provides a visually interpretable model-based spatial representation of relational data and takes account of several intrinsic network properties. Due to the structure of the likelihood function of the latent space model, the computational cost is of order O(N 2), where N is the number of nodes. This makes it infeasible for large networks. In this article, we propose an approximation of the log-likelihood function. We adapt the case-control idea from epidemiology and construct a case-control log-likelihood, which is an unbiased estimator of the log-full likelihood. Replacing the full likelihood by the case-control likelihood in the Markov chain Monte Carlo estimation of the latent space model reduces the computational time from O(N 2) to O(N), making it feasible for large networks. We evaluate its performance using simulated and real data. We fit the model to a large protein–protein interaction data using the case-control likelihood and use the model fitted link probabilities to identify false positive links. Supplemental materials are available online.

How far can one compress digital fault records? Analysis of a matching pursuit-based algorithm

In this paper, we verify how far electric disturbance signals can be compressed without compromising the analysis of encoded fault records. A recently proposed compression algorithm, referred to as Damped Sinusoidal Matching Pursuit (DSMP) has the remarkable feature of obtaining both compact and physically interpretable representations. However, for fault analysis applications, one is primarily interested in how accurate can be the analysis performed on compressed signals, instead of evaluating mean-squared error figures. Unlike previous works in digital fault records compression, the performance of the DSMP compression method is evaluated using a protocol based on fault analysis procedures commonly performed by expert engineers. This protocol is applied for comparing the results obtained in the analysis of both uncompressed records and their compressed versions at different compression ratios. The results show that the DSMP is a reliable compression system since it achieves high compression ratios (6.4:1) without causing fault analysis misinterpretation.

Archetypal analysis for machine learning and data mining

Archetypal analysis (aa) proposed by Cutler and Breiman (1994) [7] estimates the principal convex hull (pch) of a data set. As such aa favors features that constitute representative ‘corners’ of the data, i.e., distinct aspects or archetypes. We currently show that aa enjoys the interpretability of clustering – without being limited to hard assignment and the uniqueness of svd – without being limited to orthogonal representations. In order to do large scale aa, we derive an efficient algorithm based on projected gradient as well as an initialization procedure we denote FurthestSum that is inspired by the FurthestFirst approach widely used for k-means (Hochbaum and Shmoys, 1985 [14]). We generalize the aa procedure to kernel-aa in order to extract the principal convex hull in potential infinite Hilbert spaces and derive a relaxation of aa when the archetypes cannot be represented as convex combinations of the observed data. We further demonstrate that the aa model is relevant for feature extraction and dimensionality reduction for a large variety of machine learning problems taken from computer vision, neuroimaging, chemistry, text mining and collaborative filtering leading to highly interpretable representations of the dynamics in the data. Matlab code for the derived algorithms is available for download from www.mortenmorup.dk.

SPATIAL ANALYSIS OF THE APTITUDE TO LATE MATERNITY ON THE ISLAND OF SARDINIA

This study examines local heterogeneity in the aptitude of Sardinian mothers towards late reproduction, and explores its temporal persistence and association with both post-reproductive longevity and propensity to consanguineous marriage. Data on women's fertility from 1961 and birth records for 1980-1996 from Vital Statistics were analysed by means of the following indicators: the incidence of old mothers at last childbirth, female mortality (1980-2001) at 80 years of age and over and the proportion of consanguineous marriages (1930-1969). A variable kernel-smoothing method was used to create interpretable representations of the true spatial structure of the indicators, and to highlight areas of higher than expected intensity. In particular, an area of reproductive and post-reproductive longevity was identified where the traits combine with a higher tendency to relatedness. Intriguingly, this area corresponds approximately to the geographically and historically well defined central-eastern zone, which was the refuge of Sardinians during past invasions, and overlaps the Ogliastra region, which has been widely studied for its genetic homogeneity.

Swarm Intelligence for Non-Negative Matrix Factorization

The Non-negative Matrix Factorization (NMF) is a special low-rank approximation which allows for an additive parts-based and interpretable representation of the data. This article presents efforts to improve the convergence, approximation quality, and classification accuracy of NMF using five different meta-heuristics based on swarm intelligence. Several properties of the NMF objective function motivate the utilization of meta-heuristics: this function is non-convex, discontinuous, and may possess many local minima. The proposed optimization strategies are two-fold: On the one hand, a new initialization strategy for NMF is presented in order to initialize the NMF factors prior to the factorization; on the other hand, an iterative update strategy is proposed, which improves the accuracy per runtime for the multiplicative update NMF algorithm. The success of the proposed optimization strategies are shown by applying them on synthetic data and data sets coming from the areas of spam filtering/email classification, and evaluate them also in their application context. Experimental results show that both optimization strategies are able to improve NMF in terms of faster convergence, lower approximation error, and better classification accuracy. Especially the initialization strategy leads to significant reductions of the runtime per accuracy ratio for both, the NMF approximation as well as the classification results achieved with NMF.

Characterizing sensitivity and uncertainty in a multiscale model of a complex coral reef system

Abstract Sensitivity and uncertainty are intrinsic properties of ecological models, and their characterization is an important step in the modelling process. We use a spatially explicit multi-scale model of a coral reef system to explore four aspects of model sensitivity and uncertainty: (i) sensitivity to initial conditions; (ii) sensitivity to parameter values; (iii) sensitivity to spatio-temporal resolution; and (iv) the effects of uncertainty about spatio-temporal variability of ecological processes on the shape of distributions of model predictions. We use reef community composition, visualized in multivariate space, as a response variable. This approach provides an easily interpretable representation of changes in reef community composition under different parameter conditions and spatio-temporal resolutions. It is also a useful means to visualize distributions of model outcomes under differing assumptions about the nature of variability in ecological processes in the real world. Our results indicate that reef state and recovery trajectories are particularly sensitive to parameters determining coral growth and mortality rates. Variability in model outcomes depends on assumptions about the way parameters vary in space and time, and is greater at local scales than at subregional and regional scales. We highlight the fact that predictions based on the kind of model presented here are likely to be more robust for subregions and regions than for particular reef localities.

A method for modeling and visualizing the three-dimensional organization of neuron populations from replicated data: Properties, implementation and illustration

Understanding how the architecture of neuronal populations contributes to brain function requires three-dimensional representations and analyses. Neuroanatomical techniques are available to locate neurons in animal brains. Repeating an experiment in different individuals yields a collection of point patterns from which common organization principles are generally difficult to extract. We recently addressed the problem of generating statistical density maps to integrate replicated point pattern data into meaningful, interpretable representations. Applications to different neuroanatomical systems illustrated the ability of our method to reveal organization rules that cannot be perceived directly on raw data. To make the method practicable for further applications, the aim of the present paper is to establish general guidelines for appropriate parameter tuning, valid result interpretation as well as efficient implementation. Accordingly, we characterize the method by analyzing the role of its main parameter, by reporting results on its statistical properties and by demonstrating its robustness, using both simulated and real neuroanatomical data.

The first aboriginal intermediaries: Interpreting and interpreted

The challenges of interlinguistic and crosscultural communication at their extreme are epitomized in the initial contacts between Europeans and Aboriginal Australians. After sketching the linguistic encounters of the Australian continent’s original inhabitants with Asians and pre-settlement European explorers, this paper focuses on the circumstances, role and representations of the first identifiable Aboriginal interpreters — man, woman, child — in the early years of the colony. How do contact narratives represent interpreting and how might these fragmentary records — and their gaps and distortions — be interpreted through a close reading? Were the Aboriginal interpreters able to retake some of their lost agency by appropriating the colonizer’s language to their own purposes or manipulating the interpreting situations? Were they successful in bridging worlds, and how were they viewed by the two sides between which they mediated? This paper foregrounds these marginal figures who were nevertheless key linguistic and cultural mediators when the Indigenous oral societies grounded in a cosmology known as the Dreaming first encountered a society dominated by the English language and written forms and rooted in an alien worldview.

Modeling Dependencies in Brand Choice Outcomes Across Complementary Categories

We build an econometric model of a household's contemporaneous brand choice outcomes in complementary product categories. This model explicitly captures cross-category dependencies in brand choice outcomes of a household. Such dependencies have not been modeled in existing multi-category demand models.Our model accommodates cross-category dependencies that arise on account of three component effects: (1) complementarity due to the additional utility that a household derives from the joint purchase of brands in complementary categories, (2) marketing spillovers due to the effects of brands’ prices in one category affecting the households’ latent utilities for brands in the complementary category, (3) unobserved dependencies due to correlations in households’ latent utilities for brands across categories.We estimate our proposed multi-category brand choice model using scanner panel data on cake mix and frosting categories. We find that complementarity accounts for the vast majority of the estimated cross-category effects in demand. We also find that as much as 55 percent of the total retail profit impact of price promotions arise on account of brand-level (focus of our study), as opposed to category-level (focus of previous studies), dependencies in household demand. Finally, we propose an easily interpretable visual representation – Largess and Free-Ride Plot – of cross-category price elasticities that summarizes the differential abilities of brands to influence, or be influenced by, brands in the complementary category.

Grain size normalization: a case for post extraction normalization and inclusion of selective extraction procedures

The spatial mapping of trace metal and/or nutrient elemental concentrations in surface is often aided by normalization approaches that allow comparison of texturally, geochemically, and mineralogically dissimilar sites. Frequently, this is a grain‐size normalization at < 62.5 µm before chemical analysis. However, material that is > 62.5 µm (generally geochemically inert) may have fine‐grained coatings that bind components of interest and/or may consist of aggregates of fine‐grained sediment. The misplacement of these essential components within the coarse fraction of size‐normalized sediments may skew data and subsequent geochemical interpretations, especially when these components make up a substantial proportion of the sediment metal and/or nutrient binding capacity. In this article, sedimentary contaminant data were normalized using size normalization (wet sieving < 62.5 µm) and a post‐extraction normalization (PEN). PEN is a size‐based normalization procedure devised to minimize the misplacement of fine‐grained coatings/aggregates by applying a size and mineralogical normalization to inert digestion residuals, post‐chemical extraction. The data show that PEN provides a more ecologically and diagenetically interpretable representation of partitioning of elements between sediments in a freshwater reservoir. In addition, PEN samples are more rapidly collected, and consequently, PEN provides a cheaper alternative to traditional size normalization. In addition, application of the PEN allows the collection of data not only relevant to ecological health, but also relevant sedimentary diagenetic mobility data using a single sediment sample.

Selective information enhancement learning for creating interpretable representations in competitive learning

In this paper, we propose a new information-theoretic method, called "selective information enhancement learning," to explicitly interpret final representations created by learning. More specifically, we aim to make class boundaries obtained by learning as overt as possible by picking up the small number of important variables. The variable selection is performed by information enhancement in which mutual information between input patterns and competitive units is measured, while focusing upon a specific input variable. When this information is larger, the importance of the variable is higher. With selected and important variables, a network is retrained by free energy minimization. With this free energy minimization, we can obtain connection weights by considering the importance of specific variables. We applied the method to an artificial data problem, the Senate problem and the voting attitude problem, all of which are easily obtained for purposes of reproduction. Experimental results for all three problems showed that clear class boundaries could be obtained with a smaller number of variables. In addition, we could observe that a smaller number of input variables tended to have the majority of information on input patterns. This tendency became more explicit when the network size was large.

Supposed maximum information for comprehensible representations in SOM

In this paper, we propose a new information-theoretic method to simplify the computation of information and to unify several methods in one framework. The new method is called “supposed maximum information,” used to produce humanly comprehensible representations in competitive learning by taking into account the importance of input units. In the new learning method, by supposing the maximum information of input units, the actual information of input units is estimated. Then, the competitive network is trained with the estimated information in input units. The method is applied not to pure competitive learning, but to self-organizing maps, because it is easy to demonstrate visually how well the new method can produce more interpretable representations. We applied the method to three well-known sets of data, namely, the Kohonen animal data, the SPECT heart data and the voting data from the machine learning database. With these data, we succeeded in producing more explicit class boundaries on the U-matrices than did the conventional SOM. In addition, for all the data, quantization and topographic errors produced by our method were lower than those by the conventional SOM.

Registration, atlas estimation and variability analysis of white matter fiber bundles modeled as currents

This paper proposes a generic framework for the registration, the template estimation and the variability analysis of white matter fiber bundles extracted from diffusion images. This framework is based on the metric on currents for the comparison of fiber bundles. This metric measures anatomical differences between fiber bundles, seen as global homologous structures across subjects. It avoids the need to establish correspondences between points or between individual fibers of different bundles. It can measure differences both in terms of the geometry of the bundles (like its boundaries) and in terms of the density of fibers within the bundle. It is robust to fiber interruptions and reconnections. In addition, a recently introduced sparse approximation algorithm allows us to give an interpretable representation of the fiber bundles and their variations in the framework of currents.First, we used this metric to drive the registration between two sets of homologous fiber bundles of two different subjects. A dense deformation of the underlying white matter is estimated, which is constrained by the bundles seen as global anatomical landmarks. By contrast, the alignment obtained from image registration is driven only by the local gradient of the image.Second, we propose a generative statistical model for the analysis of a collection of homologous bundles. This model consistently estimates prototype fiber bundles (called template), which capture the anatomical invariants in the population, a set of deformations, which align the geometry of the template to that of each subject and a set of residual perturbations. The statistical analysis of both the deformations and the residuals describe the anatomical variability in terms of geometry (stretching, torque, etc.) and “texture” (fiber density, etc.).Third, this statistical modeling allows us to simulate new synthetic bundles according to the estimated variability. This gives a way to interpret the anatomical features that the model detects consistently across the subjects. This may be used to better understand the bias introduced by the fiber extraction methods and eventually to give anatomical characterization of the normal or pathological variability of fiber bundles.