High Level Of Interpretability Research Articles

A novel pipeline employing deep multi-attention channels network for the autonomous detection of metastasizing cells through fluorescence microscopy

Metastasis driven by cancer cell migration is the leading cause of cancer-related deaths. It involves significant changes in the organization of the cytoskeleton, which includes the actin microfilaments and the vimentin intermediate filaments. Understanding how these filament change cells from normal to invasive offers insights that can be used to improve cancer diagnosis and therapy. We have developed a computational, transparent, large-scale and imaging-based pipeline, that can distinguish between normal human cells and their isogenically matched, oncogenically transformed, invasive and metastasizing counterparts, based on the spatial organization of actin and vimentin filaments in the cell cytoplasm. Due to the intricacy of these subcellular structures, manual annotation is not trivial to automate. We used established deep learning methods and our new multi-attention channel architecture. To ensure a high level of interpretability of the network, which is crucial for the application area, we developed an interpretable global explainable approach correlating the weighted geometric mean of the total cell images and their local GradCam scores. The methods offer detailed, objective and measurable understanding of how different components of the cytoskeleton contribute to metastasis, insights that can be used for future development of novel diagnostic tools, such as a nanometer level, vimentin filament-based biomarker for digital pathology, and for new treatments that significantly can increase patient survival.

Psychological factors enhanced heterogeneous learning interactive graph knowledge tracing for understanding the learning process.

With the rapid expansion of online education, there is a burgeoning interest within the EdTech space to offer tailored learning experiences that cater to individual student's abilities and needs. Within this framework, knowledge tracing tasks have garnered considerable attention. The primary objective of knowledge tracing is to develop a model that assesses a student's proficiency in a particular skill based on their historical performance in exercises, enabling predictions regarding the likelihood of correct responses in future exercises. While existing knowledge tracing models often incorporate information such as students' exercise answering history and skill mastery level, they frequently overlook the students' mental states during the learning process. This paper addresses this gap by introducing a novel psychological factors-enhanced heterogeneous learning interactive graph knowledge tracing model (Psy-KT). This model delineates the interactions among students, exercises, and skills through a heterogeneous graph, supplementing it with four psychological factors that capture students' mental states during the learning process: frustration level, confusion level, concentration level, and boredom level. In the modeling of students' learning processes, we incorporate the forgetting curve and construct relevant cognitive parameters from the features. Additionally, we employ the Item Response Theory (IRT) model to predict students' performance in answering exercises at the subsequent time step. This model not only delves into the psychological aspects of students during the learning process but also integrates the simulation of forgetting, a natural phenomenon in the learning journey. The inclusion of cognitive parameters enhances the description of changes in students' abilities throughout the learning process. This dual focus allows for a more comprehensive understanding of students' learning behaviors while providing a high level of interpretability for the model. Empirical validation of the Psy-KT model is conducted using four publicly available datasets, demonstrating its superior performance in predicting students' future performance. Through rigorous experimentation, the integration of psychological and forgetting factors in the Psy-KT model not only improves predictive accuracy but also enables educators to offer more targeted tutoring and advice, enhancing the overall efficacy of the learning experience.

CrossFuse-XGBoost: accurate prediction of the maximum recommended daily dose through multi-feature fusion, cross-validation screening and extreme gradient boosting.

In the drug development process, approximately 30% of failures are attributed to drug safety issues. In particular, the first-in-human (FIH) trial of a new drug represents one of the highest safety risks, and initial dose selection is crucial for ensuring safety in clinical trials. With traditional dose estimation methods, which extrapolate data from animals to humans, catastrophic events have occurred during Phase I clinical trials due to interspecies differences in compound sensitivity and unknown molecular mechanisms. To address this issue, this study proposes a CrossFuse-extreme gradient boosting (XGBoost) method that can directly predict the maximum recommended daily dose of a compound based on existing human research data, providing a reference for FIH dose selection. This method not only integrates multiple features, including molecular representations, physicochemical properties and compound-protein interactions, but also improves feature selection based on cross-validation. The results demonstrate that the CrossFuse-XGBoost method not only improves prediction accuracy compared to that of existing local weighted methods [k-nearest neighbor (k-NN) and variable k-NN (v-NN)] but also solves the low prediction coverage issue of v-NN, achieving full coverage of the external validation set and enabling more reliable predictions. Furthermore, this study offers a high level of interpretability by identifying the importance of different features in model construction. The 241 features with the most significant impact on the maximum recommended daily dose were selected, providing references for optimizing the structure of new compounds and guiding experimental research. The datasets and source code are freely available at https://github.com/cqmu-lq/CrossFuse-XGBoost.

Generalizing the Gurson model using symbolic regression and transfer learning to relax inherent assumptions

To generate material models with fewer limiting assumptions while maintaining closed-form, interpretable solutions, we propose using genetic programming based symbolic regression (GPSR), a machine learning (ML) approach that describes data using free-form symbolic expressions. To maximize interpretability, we start from an analytical, derived material model, the Gurson model for porous ductile metals, and systematically relax inherent assumptions made in its derivation to understand each assumption’s contribution to the GPSR model forms. We incorporate transfer learning methods into the GPSR training process to increase GPSR efficiency and generate models that abide by known mechanics of the system. The results show that regularizing the GPSR fitness function is critical for generating physically valid models and illustrate how GPSR allows a high level of interpretability compared with other ML approaches. The method of systematic assumption relaxation allows the generation of models that address limiting assumptions found in the Gurson model, and the symbolic forms allow conjecture of decreased material strength due to void interaction and non-symmetric void shapes.

Re-interpreting rules interpretability

AbstractTrustworthy machine learning requires a high level of interpretability of machine learning models, yet many models are inherently black-boxes. Training interpretable models instead—or using them to mimic the black-box model—seems like a viable solution. In practice, however, these interpretable models are still unintelligible due to their size and complexity. In this paper, we present an approach to explain the logic of large interpretable models that can be represented as sets of logical rules by a simple, and thus intelligible, descriptive model. The coarseness of this descriptive model and its fidelity to the original model can be controlled, so that a user can understand the original model in varying levels of depth. We showcase and discuss this approach on three real-world problems from healthcare, material science, and finance.

Enhanced classification and regression tree (CART) by genetic algorithm (GA) and grid search (GS) for flood susceptibility mapping and assessment

Despite the high level of interpretability of the Classification and Regression Tree (CART) as one of the widely-used data mining models, its results usually are less accurate than other models. To address the low accuracy of the CART model, this paper developed two new and enhanced CART-based algorithms that can provide both high interpretability and high accuracy for flood susceptibility mapping. In the developed models, various hyperparameters of the CART were explicitly tuned using the genetic algorithm (GA) and grid search (GS) methods. Eleven variables and 222 flood locations were used in Sardabroud watershed in Iran for modelling. The results showed area under the receiver operating characteristic curve (AUROC) values of 0.884, 0.927 and 0.908, respectively, for the CART, CART-GS and CART-GA models in the validation run. The results showed that the developed models can produce the output that satisfies both accuracy and interpretability criteria.

Pupil responses to implied motion in figurative and abstract paintings.

Motion can be perceived in static images, such as photos and figurative paintings, representing realistic subjects in motion, with or without directional information (e.g., motion blur or speed lines). Motion impression can be achieved even in non-realistic static images such as motion illusions and abstract paintings. It has been shown that visual motion processing affects the diameter of the pupil, responding differently to real, illusory, and implied motion in photographs (IM). It has been suggested that these different effects might be due to top-down modulations from different cortical areas underlying their processing. It is worthwhile to investigate pupillary response to figurative paintings, since they require an even higher level of interpretation than photos representing the same kind of subjects, given the complexity of cognitive processes involved in the aesthetic experience. Also, pupil responses to abstract paintings allows to study the effect of IM perception in representations devoid of real-life motion cues. We measured pupil responses to IM in figurative and abstract artworks depicting static and dynamic scenes, as rated by a large group of individuals not participating in the following experiment. Since the pupillary response is modulated by the subjective image interpretation, a motion rating test has been used to correct individual pupil data according to whether participants actually perceived the presence of motion in the paintings. Pupil responses to movies showing figurative and abstract subjects, and to motion illusions were also measured, to compare real and illusory motion with painted IM. Movies, both figurative and abstract, elicit the largest pupillary dilation of all static stimuli, whereas motion illusions cause the smallest pupil size, as previously shown. Interestingly, pupil responses to IM depend on the paintings’ style. Figurative paintings depicting moving subjects cause more dilation than those representing static figures, and pupil size increases with the strength of IM, as already found with realistic photos. The opposite effect is obtained with abstract artworks. paintings depicting motion produce less dilation than those depicting stillness. In any case, these results reflect the individual subjective perception of dynamism, as the very same paintings can induce opposite responses in observer which interpreted it as static or dynamic. Overall, our data show that pupil size depends on high-level interpretation of motion in paintings, even when they do not represent real-world scenes. Our findings further suggest that the pupil is modulated by multiple top-down cortical mechanisms, involving the processing of motion, attention, memory, imagination, and other cognitive functions necessary for enjoying a complete aesthetic experience.

Environment humidity and temperature prediction in agriculture using Mamdani inference systems

<span>This paper presents the results of a humidity and temperature prediction model in the environment for agriculture, using diffuse sets and optimizing their parameters by heuristic methods, such as genetic algorithms, and exact methods such as Quasi-Newton. It has been identified that non-specialized users could have difficulties in understanding the system dynamics and the behavior of variables over time. The aim of this research is obtain models with a high level of interpretability and accuracy that allows predicting the temperature and humidity values for the environment. The use of fuzzy logic to present a solution has great advantages as this system is highly rated for interpretability. Furthermore, by relating the obtained values for environment humidity and temperature to qualitative categories as high, medium or low, it allows non-specialized users to have a better understanding of the system dynamics. Two optimization techniques are applied to two different diffuse sets that allow the prediction of the humidity and temperature. It is found that the best implementation involves a Mamdani fuzzy inference system optimized with Quasi-Newton algorithm that uses a set of initial values attained through a previous optimization process with a genetic algorithm.</span>

A Semi-Supervised Multi-Task Learning Approach for Predicting Short-Term Kidney Disease Evolution.

Kidney Disease (KD) may hide complex causes and is associated with a tremendous socio-economic impact. Timely identification and management from the first level of medical care represent the most effective strategy to address the growing global burden sustainably. Clinical practice guidelines suggest utilizing estimated Glomerular Filtration Rate (eGFR) for routine evaluation within a screening purpose. Accordingly, the analysis of Electronic Health Records (EHRs) using Machine Learning techniques offers great opportunities to monitor and predict the eGFR trend over time. This paper aims to propose a novel Semi-Supervised Multi-Task Learning (SS-MTL) approach for predicting short-term KD evolution on multiple General Practitioners' EHR data. We demonstrated that the SS-MTL approach can (i) capture the eGFR temporal evolution by imposing a temporal relatedness between consecutive time windows and (ii) exploit useful information from unlabeled patients when labeled patients are less numerous with a gain of up to 4.1% in terms of Recall. This situation reflects the real-case scenario, where available labeled samples are limited, but those unlabeled much more abundant. The SS-MTL approach, also given the high level of interpretability, might be the ideal candidate in general practice to get integrated within a decision support system for KD screening purposes.

Fusing Mobile Phone and Travel Survey Data to Model Urban Activity Dynamics

A key issue to understand urban system is to characterize the activity dynamics in a city—when, where, what, and how activities happen in a city. To better understand the urban activity dynamics, city-wide and multiday activity participation sequence data, namely, activity chain as well as suitable spatiotemporal models, are needed. The commonly used household travel survey data in activity analysis suffers from limited sample size and temporal coverage. The emergence of large-scale spatiotemporal data in urban areas, such as mobile phone data, provides a new opportunity to infer urban activities and the underlying dynamics. However, the challenge is the absence of labeled activity information in mobile phone data. Consequently, how to fuse the useful information in household survey data and mobile phone data to build city-wide, multiday, and all-time activity chains becomes an important research question. Moreover, the multidimension structure of the activity data (e.g., location, start time, duration, type) makes the extraction of spatiotemporal activity patterns another difficult problem. In this study, the authors first introduce an activity chain inference model based on tensor decomposition to infer the missing activity labels in large-scale and multiday activity data, and then develop a spatiotemporal event clustering model based on DBSCAN, called STE-DBSCAN, to identify the spatiotemporal activity patterns. The proposed approaches achieved good accuracy and produced patterns with a high level of interpretability.

A Fast Inference and Type-Reduction Process for Constrained Interval Type-2 Fuzzy Systems

Constrained interval type-2 (CIT2) fuzzy sets have been introduced to preserve interpretability when moving from type-1 to interval type-2 (IT2) membership functions. Although they can be used to produce type-2 fuzzy systems with enhanced explainability, so far, the latter comes at the expense of high computational cost. Specifically, the exhaustive type-reduction method for CIT2 Mamdani systems has been shown to be too slow to be used in practical applications and even the current approximation procedure is much slower than modern type-reduction algorithms used for IT2 fuzzy sets. In this article, a novel type-reduction procedure for CIT2 sets is presented, based on the concept of switch indices. The algorithm is applied on a real-world classification problem and compared to other type-reduction approaches used in IT2 and CIT2 systems. In the case studies presented, the new algorithm is significantly faster than the exhaustive and sampling CIT2 approaches while keeping the high level of interpretability of the type-reduction operation that characterizes CIT2 fuzzy sets.

GSIC: A New Interpretable System for Knowledge Exploration and Classification

Machine learning and data mining techniques have been developed rapidly in recent times. In tasks such as classification, machine learning techniques have been shown to equal to and even surpass human performance. However, high performance models are usually complex, opaque and have low interpretability thus making it difficult to explain the underlying behaviors of those models that lead to the final outcomes. In many domains such as medicine and healthcare, interpretability is one of the most important factors when considering the adoption of those models. In this paper, we propose a two-stage binary classification system applicable for healthcare (or general) data that benefits from a high level of interpretability and can at the same time achieve the results comparable to commonly used classification techniques. The motivation behind the proposed system is the lack of effective classification methods for handling data generated by various distributions (such as healthcare or banking data) that can harmonize both performance and interpretability perspectives. In this work, we tackle the problem by applying divide and conquer strategy on a new disentangled representation of the underlying data. The merit of our system is evaluated by a classification experiment with a wide range of real data and popular transparent and black-box models. Furthermore, a use case in data of sepsis patients staying in the ICU (Intensive Care Unit) is depicted to prove the interpretability of the proposed model.

Optimization strategies for Microgrid energy management systems by Genetic Algorithms

Grid-connected Microgrids (MGs) have a key role for bottom-up modernization of the electric distribution network forward next generation Smart Grids, allowing the application of Demand Response (DR) services, as well as the active participation of prosumers into the energy market. To this aim, MGs must be equipped with suitable Energy Management Systems (EMSs) in charge to efficiently manage in real time internal energy flows and the connection with the grid. Several decision making EMSs are proposed in literature mainly based on soft computing techniques and stochastic models. The adoption of Fuzzy Inference Systems (FISs) has proved to be very successful due to their ease of implementation, low computational run time cost, and the high level of interpretability with respect to more conventional models. In this work we investigate different strategies for the synthesis of a FIS (i.e. rule based) EMS by means of a hierarchical Genetic Algorithm (GA) with the aim to maximize the profit generated by the energy exchange with the grid, assuming a Time Of Use (TOU) energy price policy, and at the same time to reduce the EMS rule base system complexity. Results show that the performances are just 10% below to the ideal (optimal) reference solution, even when the rule base system is reduced to less than 30 rules.

Geographically Modeling and Understanding Factors Influencing Transit Ridership: An Empirical Study of Shenzhen Metro

Ridership analysis at the local level has a pivotal role in sustainable urban construction and transportation planning. In practice, urban rail transit (URT) ridership is affected by complex factors that vary across the urban area. The aim of this study is to model and explore the factors that impact metro station ridership in Shenzhen, China from a local perspective. The direct demand model, which uses ordinary least squares (OLS) estimation, is the most widely used method of ridership modeling. However, OLS estimation assumes parametric stability. This study investigates the use of a direct demand model on the basis of geographically weighted regression (GWR) to model the local relationships between metro station ridership and potential influencing factors. Real-world Shenzhen Metro smart card data are used to test and verify the applicability and performance of the model. The results show that GWR performs better than OLS estimation in terms of both model fitting and spatial interpretation. The GWR model demonstrates a high level of interpretability regarding the spatial distribution and variation of each coefficient, and thus can provide insights for decision-makers into URT ridership and its complex factors from a local perspective.

An improved fuzzy-differential evolution approach applied to classification of tumors in liver CT scan images.

Fuzzy inference systems have been frequently used in medical diagnosis for managing uncertainty sources in the medical images. In addition, fuzzy systems have high level of interpretability because of using linguistic terms for knowledge representation through the reasoning process. The evolutionary algorithms can be applied for optimization of the systems. This article takes advantages of differential evolutionary algorithm to search the problem space more intelligently using the knowledge of distance vector of the candidate solutions. For this, a hybrid fuzzy-DE model has been proposed for the problem of classification of the Haptic metastasis tumors through information obtained from the features measured in the CT scan images by radiologists. The hybrid fuzzy-DE model was evaluated using a real liver cancer dataset obtained from the Noor medical imaging center in Tehran. The results of the hybrid proposed models were compared with the diagnosis of the specialists, the results reveal that the proposed fuzzy-DE model has high capability for diagnosis of the hepatic metastasis tumors with an accuracy of 99.24% with 95% confidence interval (98.32 100) in terms of area under the ROC curve. The proposed model outperforms the multilayer perceptron (MLP) neural network and fuzzy-genetic algorithm (GA). The proposed model improves the trade-offs between the accuracy and interpretability by providing a model with high accuracy using less input variables. The hybrid fuzzy-DE model is promising to assist the medical specialists for early diagnosis of liver cancer and save more people lives.

Machine Learning Models Identify Multimodal Measurements Highly Predictive of Transdiagnostic Symptom Severity for Mood, Anhedonia, and Anxiety

BackgroundInsights from neuroimaging-based biomarker research have not yet translated into clinical practice. This translational gap may stem from a focus on diagnostic classification, rather than on prediction of transdiagnostic psychiatric symptom severity. Currently, no transdiagnostic, multimodal predictive models of symptom severity that include neurobiological characteristics have emerged. MethodsWe built predictive models of 3 common symptoms in psychiatric disorders (dysregulated mood, anhedonia, and anxiety) from the Consortium for Neuropsychiatric Phenomics dataset (N = 272), which includes clinical scale assessments, resting-state functional magnetic resonance imaging (MRI), and structural MRI measures from patients with schizophrenia, bipolar disorder, and attention-deficit/hyperactivity disorder and healthy control subjects. We used an efficient, data-driven feature selection approach to identify the most predictive features from these high-dimensional data. ResultsThis approach optimized modeling and explained 65% to 90% of variance across the 3 symptom domains, compared to 22% without using the feature selection approach. The top performing multimodal models retained a high level of interpretability that enabled several clinical and scientific insights. First, to our surprise, structural features did not substantially contribute to the predictive strength of these models. Second, the Temperament and Character Inventory scale emerged as a highly important predictor of symptom variation across diagnoses. Third, predictive resting-state functional MRI connectivity features were widely distributed across many intrinsic resting-state networks. ConclusionsCombining resting-state functional MRI with select questions from clinical scales enabled high prediction of symptom severity across diagnostically distinct patient groups and revealed that connectivity measures beyond a few intrinsic resting-state networks may carry relevant information for symptom severity.

Hierarchical Tensor SOM Network for Multilevel–Multigroup Analysis

The aim of this work is to develop a visualization method for multilevel---multigroup analysis based on a multiway nonlinear dimensionality reduction. The task of the method is to visualize what kinds of members each group is composed and to visualize the similarity between the groups in terms of probability distribution of constituent members. To achieve the task, the proposed method consists of hierarchically coupled tensor self-organizing maps, corresponding to the member/group level. This architecture enables more flexible analysis than ordinary parametric multilevel analysis, as it retains a high level of interpretability supported by strong visualization. We applied the proposed method to one benchmark dataset and two practical datasets: one is the survey data on the football players belonging to different teams and the other is the employee survey data belonging to different departments in a company. Our method successfully visualizes the types of the members that constitute each group as well as visualizes the differences or similarities between the groups.

Minimax Probability TSK Fuzzy System Classifier: A More Transparent and Highly Interpretable Classification Model

When an intelligent model is used for medical diagnosis, it is desirable to have a high level of interpretability and transparent model reliability for users. Compared with most of the existing intelligence models, fuzzy systems have shown a distinctive advantage in their interpretabilities. However, how to determine the model reliability of a fuzzy system trained for a recognition task is still an unsolved problem at present. In this study, a minimax probability Takagi–Sugeno–Kang (TSK) fuzzy system classifier called MP-TSK-FSC is proposed to train a fuzzy system classifier and determine the model reliability simultaneously. For the proposed MP-TSK-FSC, a lower bound of correct classification can be presented to the users to characterize the reliability of the trained fuzzy classifier. Thus, the obtained classifier has the distinctive characteristics of both a high level of interpretability and transparent model reliability inherited from the fuzzy system and minimax probability learning strategy, respectively. Our experiments on synthetic datasets and several real-world datasets for medical diagnosis have confirmed the distinctive characteristics of the proposed method.

Gatunkowe pogranicza – samurajskie filmy Hideo Goshy

The paper is dedicated to the intertextual and polemic threads in jidai-geki films of Hideo Gosha – especially The Sword of the Beast ( Kedamono no ken , 1965), The Secret of the Urn ( Tange Sazen: Hien iaigiri , 1966) and Official Gold ( Goyokin , 1969). Those works are treated here as emblematic of the transformations which occurred in the Japanese genre cinema of 1960s, when the samurai film (similarly to its American counterpart, the western) entered the phase of revision and reinterpretation of its fundamental ideological, aesthetic and philosophical values. This tendency, which had been anticipated by pre-war dissidents of the samurai films, such as Buntaro Futagawa or Sadao Yamanaka, and elevated by Masaki Kobayashi in 1960s, became a point of departure for further explorations for directors like Kihachi Okamoto, Eiichi Kudo or, last but not least, Hideo Gosha. Gosha’s works are characterised by a certain type of double-coding: his well-crafted, attractive adventure stories have all the features that distinguish the so-called “audience-friendly” films – and yet, on the higher level of interpretation, they appear as high quality genre pastiches, which deconstruct the national stereotypes imprinted in genre typing.

The Language of Extraversion

To understand the impact of personality, one needs to know how personality differences manifest themselves in language use. The present study investigates the link between extraversion and language abstraction. Participants’ spontaneous verbal utterances in face-to-face interactions were analyzed for language abstraction by applying the linguistic category model, which distinguishes predicate types that convey information in concrete or interpretative manner. We also applied the Linguistic Inquiry and Word Count (LIWC) program to relate several word categories to extraversion and language abstraction. Results show significant positive correlations between extraversion and both language abstraction and self-reported level of interpretation. Language abstraction was also linked to LIWC variables (e.g., articles, numbers) previously shown to be related to extraversion. The findings suggest that the verbal style of extraverts is characterized by a higher level of abstract interpretation, whereas introverts tend to stick to concrete facts.