Rationalization for explainable NLP: a survey.

Recent advances in deep learning have improved the performance of many Natural Language Processing (NLP) tasks such as translation, question-answering, and text classification. However, this improvement comes at the expense of model explainability. Black-box models make it difficult to understand the internals of a system and the process it takes to arrive at an output. Numerical (LIME, Shapley) and visualization (saliency heatmap) explainability techniques are helpful; however, they are insufficient because they require specialized knowledge. These factors led rationalization to emerge as a more accessible explainable technique in NLP. Rationalization justifies a model's output by providing a natural language explanation (rationale). Recent improvements in natural language generation have made rationalization an attractive technique because it is intuitive, human-comprehensible, and accessible to non-technical users. Since rationalization is a relatively new field, it is disorganized. As the first survey, rationalization literature in NLP from 2007 to 2022 is analyzed. This survey presents available methods, explainable evaluations, code, and datasets used across various NLP tasks that use rationalization. Further, a new subfield in Explainable AI (XAI), namely, Rational AI (RAI), is introduced to advance the current state of rationalization. A discussion on observed insights, challenges, and future directions is provided to point to promising research opportunities.

Toward sustainable culture media: Using artificial intelligence to optimize reduced-serum formulations for cultivated meat.

When considering options for future foods, cell culture approaches are at the fore, however, culture media to support the process has been identified as a significant contributor to the overall global warming potential (GWP) and cost of cultivated meat production. To address this issue, an artificial intelligence-based approach was applied to simultaneously optimize the GWP, cost, and cell growth rate of a reduced-serum culture media formulation for a zebrafish (ZEM2S cell line) cultivated meat production system. Response surface methodology (RSM) was used to design the experiments, with seven components - IGF, FGF, TGF, PDGF, selenium, ascorbic acid, and serum - selected as independent variables, given their influence on culture media performance. Radial basis function (RBF) neural networks and genetic algorithm (GA) were applied for prediction of dependent variables, and optimization of the culture media formulation, respectively. The results indicated that the developed RBF could accurately predict the GWP, cost and growth rate, with a model efficiency of 0.98. Subsequently, the three developed RBF neural networks predictive models were used as the inputs for a multi-objective genetic algorithm, and the optimal quantities of the independent variables were determined using a multi-objective optimization algorithm. The suggested RSM + RBF + GA framework in this study could be applied to sustainably optimize serum-free media development, identifying the combination of media ingredients that balances yield, environmental impact, and cost for various cultivated meat cell lines.

Clustering by periodontitis-associated factors: A novel application to NHANES data.

BACKGROUND: Unsupervised clustering is a method used to identify heterogeneity among groups and homogeneity within a group of patients. Without a prespecified outcome entry, the resulting model deciphers patterns that may not be disclosed using traditional methods. This is the first time such clustering analysis is applied in identifying unique subgroups at high risk for periodontitis in National Health and Nutrition Examination Surveys (NHANES 2009 to 2014 data sets using >500 variables. METHODS: Questionnaire, examination, and laboratory data (33 tables) for >1,000 variables were merged from 14,072 respondents who underwent clinical periodontal examination. Participants with ≥6 teeth and available data for all selected categories were included (N = 1,222). Data wrangling produced 519 variables. k-means/modes clustering (k = 2:14) was deployed. The optimal k-value was determined through the elbow method, formula = ∑ (xi2 ) - ((∑ xi )2 /n). The 5-cluster model showing the highest variability (63.08%) was selected. The 2012 Centers for Disease Control and Prevention/American Academy of Periodontology (AAP) and 2018 European Federation of Periodontology/AAP periodontitis case definitions were applied. RESULTS: Cluster 1 (n = 249) showed the highest prevalence of severe periodontitis (43%); 39% self-reported "fair" general health; 55% had household income <$35,000/year; and 48% were current smokers. Cluster 2 (n = 154) had one participant with periodontitis. Cluster 3 (n = 242) represented the greatest prevalence of moderate periodontitis (53%). In Cluster 4 (n = 35) only one participant had no periodontitis. Cluster 5 (n = 542) was the systemically healthiest with 77% having no/mild periodontitis. CONCLUSION: Clustering of NHANES demographic, systemic health, and socioeconomic data effectively identifies characteristics that are statistically significantly related to periodontitis status and hence detects subpopulations at high risk for periodontitis without costly clinical examinations.

Measuring outcomes in healthcare economics using Artificial Intelligence: With application to resource management.

The quality of service in healthcare is constantly challenged by outlier events such as pandemics (i.e., Covid-19) and natural disasters (such as hurricanes and earthquakes). In most cases, such events lead to critical uncertainties in decision-making, as well as in multiple medical and economic aspects at a hospital. External (geographic) or internal factors (medical and managerial) lead to shifts in planning and budgeting, but most importantly, reduce confidence in conventional processes. In some cases, support from other hospitals proves necessary, which exacerbates the planning aspect. This paper presents three data-driven methods that provide data-driven indicators to help healthcare managers organize their economics and identify the most optimum plan for resources allocation and sharing. Conventional decision-making methods fall short in recommending validated policies for managers. Using reinforcement learning, genetic algorithms, traveling salesman, and clustering, we experimented with different healthcare variables and presented tools and outcomes that could be applied at health institutes. Experiments are performed; the results are recorded, evaluated, and presented.

Preventive healthcare policies in the US: solutions for disease management using Big Data Analytics.

Data-driven healthcare policy discussions are gaining traction after the Covid-19 outbreak and ahead of the 2020 US presidential elections. The US has a hybrid healthcare structure; it is a system that does not provide universal coverage, albeit few years ago enacted a mandate (Affordable Care Act-ACA) that provides coverage for the majority of Americans. The US has the highest health expenditure per capita of all western and developed countries; however, most Americans don't tap into the benefits of preventive healthcare. It is estimated that only 8% of Americans undergo routine preventive screenings. On a national level, very few states (15 out of the 50) have above-average preventive healthcare metrics. In literature, many studies focus on the cure of diseases (research areas such as drug discovery and disease prediction); whilst a minority have examined data-driven preventive measures-a matter that Americans and policy makers ought to place at the forefront of national issues. In this work, we present solutions for preventive practices and policies through Machine Learning (ML) methods. ML is morally neutral, it depends on the data that train the models; in this work, we make the case that Big Data is an imperative paradigm for healthcare. We examine disparities in clinical data for US patients by developing correlation and imputation methods for data completeness. Non-conventional patterns are identified. The data lifecycle followed is methodical and deliberate; 1000+ clinical, demographical, and laboratory variables are collected from the Centers for Disease Control and Prevention (CDC). Multiple statistical models are deployed (Pearson correlations, Cramer's V, MICE, and ANOVA). Other unsupervised ML models are also examined (K-modes and K-prototypes for clustering). Through the results presented in the paper, pointers to preventive chronic disease tests are presented, and the models are tested and evaluated.