From explanation to intervention: Interactive knowledge extraction from Convolutional Neural Networks used in radiology.

Deep Learning models such as Convolutional Neural Networks (CNNs) are very effective at extracting complex image features from medical X-rays. However, the limited interpretability of CNNs has hampered their deployment in medical settings as they failed to gain trust among clinicians. In this work, we propose an interactive framework to allow clinicians to ask what-if questions and intervene in the decisions of a CNN, with the aim of increasing trust in the system. The framework translates a layer of a trained CNN into a measurable and compact set of symbolic rules. Expert interactions with visualizations of the rules promote the use of clinically-relevant CNN kernels and attach meaning to the rules. The definition and relevance of the kernels are supported by radiomics analyses and permutation evaluations, respectively. CNN kernels that do not have a clinically-meaningful interpretation are removed without affecting model performance. By allowing clinicians to evaluate the impact of adding or removing kernels from the rule set, our approach produces an interpretable refinement of the data-driven CNN in alignment with medical best practice.

Extracting Relational Explanations From Deep Neural Networks: A Survey From a Neural-Symbolic Perspective.

The term "explainable AI" refers to the goal of producing artificially intelligent agents that are capable of providing explanations for their decisions. Some models (e.g., rule-based systems) are designed to be explainable, while others are less explicit "black boxes" for which their reasoning remains a mystery. One example of the latter is the neural network, and over the past few decades, researchers in the field of neural-symbolic integration (NSI) have sought to extract relational knowledge from such networks. Extraction from deep neural networks, however, has remained a challenge until recent years in which many methods of extracting distinct, salient features from input or hidden feature spaces of deep neural networks have been proposed. Furthermore, methods of identifying relationships between these features have also emerged. This article presents examples of old and new developments in extracting relational explanations in order to argue that the latter have analogies in the former and, as such, can be described in terms of long-established taxonomies and frameworks presented in early neural-symbolic literature. We also outline potential future research directions that come to light from this refreshed perspective.

Has Income Segregation Really Increased? Bias and Bias Correction in Sample-Based Segregation Estimates.

Several recent studies have concluded that residential segregation by income in the United States has increased in the decades since 1970, including a significant increase after 2000. Income segregation measures, however, are biased upward when based on sample data. This is a potential concern because the sampling rate of the American Community Survey (ACS)-from which post-2000 income segregation estimates are constructed-was lower than that of the earlier decennial censuses. Thus, the apparent increase in income segregation post-2000 may simply reflect larger upward bias in the estimates from the ACS, and the estimated trend may therefore be inaccurate. In this study, we first derive formulas describing the approximate sampling bias in two measures of segregation. Next, using Monte Carlo simulations, we show that the bias-corrected estimators eliminate virtually all of the bias in segregation estimates in most cases of practical interest, although the correction fails to eliminate bias in some cases when the population is unevenly distributed among geographic units and the average within-unit samples are very small. We then use the bias-corrected estimators to produce unbiased estimates of the trends in income segregation over the last four decades in large U.S. metropolitan areas. Using these corrected estimates, we replicate the central analyses in four prior studies on income segregation. We find that the primary conclusions from these studies remain unchanged, although the true increase in income segregation among families after 2000 was only half as large as that reported in earlier work. Despite this revision, our replications confirm that income segregation has increased sharply in recent decades among families with children and that income inequality is a strong and consistent predictor of income segregation.

Widespread Elevational Occurrence of Antifungal Bacteria in Andean Amphibians Decimated by Disease: A Complex Role for Skin Symbionts in Defense Against Chytridiomycosis.

Emerging infectious disease is a growing threat to global health, and recent discoveries reveal that the microbiota dwelling on and within hosts can play an important role in health and disease. To understand the capacity of skin bacteria to protect amphibian hosts from the fungal disease chytridiomycosis caused by Batrachochytrium dendrobatidis (Bd), we isolated 192 bacterial morphotypes from the skin of 28 host species of frogs (families Bufonidae, Centrolenidae, Hemiphractidae, Hylidae, Leptodactylidae, Strabomantidae, and Telmatobiidae) collected from the eastern slopes of the Peruvian Andes (540-3,865 m a.s.l.) in the Kosñipata Valley near Manu National Park, a site where we previously documented the collapse of montane frog communities following chytridiomycosis epizootics. We obtained isolates through agar culture from skin swabs of wild frogs, and identified bacterial isolates by comparing 16S rRNA sequences against the GenBank database using BLAST. We identified 178 bacterial strains of 38 genera, including 59 bacterial species not previously reported from any amphibian host. The most common bacterial isolates were species of Pseudomonas, Paenibacillus, Chryseobacterium, Comamonas, Sphingobacterium, and Stenotrophomonas. We assayed the anti-fungal abilities of 133 bacterial isolates from 26 frog species. To test whether cutaneous bacteria might inhibit growth of the fungal pathogen, we used a local Bd strain isolated from the mouthparts of stream-dwelling tadpoles (Hypsiboas gladiator, Hylidae). We quantified Bd-inhibition in vitro with co-culture assays. We found 20 bacterial isolates that inhibited Bd growth, including three isolates not previously known for such inhibitory abilities. Anti-Bd isolates occurred on aquatic and terrestrial breeding frogs across a wide range of elevations (560-3,695 m a.s.l.). The inhibitory ability of anti-Bd isolates varied considerably. The proportion of anti-Bd isolates was lowest at mid-elevations (6%), where amphibian declines have been steepest, and among hosts that are highly susceptible to chytridiomycosis (0-14%). Among non-susceptible species, two had the highest proportion of anti-Bd isolates (40 and 45%), but one common and non-susceptible species had a low proportion (13%). In conclusion, we show that anti-Bd bacteria are widely distributed elevationally and phylogenetically across frog species that have persisted in a region where chytridiomycosis emerged, caused a devastating epizootic and continues to infect amphibians.

Heterotrophy mitigates the response of the temperate coral Oculina arbuscula to temperature stress.

Anthropogenic increases in atmospheric carbon dioxide concentration have caused global average sea surface temperature (SST) to increase by approximately 0.11°C per decade between 1971 and 2010 - a trend that is projected to continue through the 21st century. A multitude of research studies have demonstrated that increased SSTs compromise the coral holobiont (cnidarian host and its symbiotic algae) by reducing both host calcification and symbiont density, among other variables. However, we still do not fully understand the role of heterotrophy in the response of the coral holobiont to elevated temperature, particularly for temperate corals. Here, we conducted a pair of independent experiments to investigate the influence of heterotrophy on the response of the temperate scleractinian coral Oculina arbuscula to thermal stress. Colonies of O. arbuscula from Radio Island, North Carolina, were exposed to four feeding treatments (zero, low, moderate, and high concentrations of newly hatched Artemia sp. nauplii) across two independent temperature experiments (average annual SST (20°C) and average summer temperature (28°C) for the interval 2005-2012) to quantify the effects of heterotrophy on coral skeletal growth and symbiont density. Results suggest that heterotrophy mitigated both reduced skeletal growth and decreased symbiont density observed for unfed corals reared at 28°C. This study highlights the importance of heterotrophy in maintaining coral holobiont fitness under thermal stress and has important implications for the interpretation of coral response to climate change.

Temperature Regimes Impact Coral Assemblages along Environmental Gradients on Lagoonal Reefs in Belize.

Coral reefs are increasingly threatened by global and local anthropogenic stressors such as rising seawater temperature, nutrient enrichment, sedimentation, and overfishing. Although many studies have investigated the impacts of local and global stressors on coral reefs, we still do not fully understand how these stressors influence coral community structure, particularly across environmental gradients on a reef system. Here, we investigate coral community composition across three different temperature and productivity regimes along a nearshore-offshore gradient on lagoonal reefs of the Belize Mesoamerican Barrier Reef System (MBRS). A novel metric was developed using ultra-high-resolution satellite-derived estimates of sea surface temperatures (SST) to classify reefs as exposed to low (lowTP), moderate (modTP), or high (highTP) temperature parameters over 10 years (2003 to 2012). Coral species richness, abundance, diversity, density, and percent cover were lower at highTP sites relative to lowTP and modTP sites, but these coral community traits did not differ significantly between lowTP and modTP sites. Analysis of coral life history strategies revealed that highTP sites were dominated by hardy stress-tolerant and fast-growing weedy coral species, while lowTP and modTP sites consisted of competitive, generalist, weedy, and stress-tolerant coral species. Satellite-derived estimates of Chlorophyll-a (chl-a) were obtained for 13-years (2003-2015) as a proxy for primary production. Chl-a concentrations were highest at highTP sites, medial at modTP sites, and lowest at lowTP sites. Notably, thermal parameters correlated better with coral community traits between site types than productivity, suggesting that temperature (specifically number of days above the thermal bleaching threshold) played a greater role in defining coral community structure than productivity on the MBRS. Dominance of weedy and stress-tolerant genera at highTP sites suggests that corals utilizing these two life history strategies may be better suited to cope with warmer oceans and thus may warrant protective status under climate change.