Publications

Although models exist that predict human response times (RTs) in tasks such as target search and visual discrimination, the development of image-computable predictors for scene understanding time remains an open challenge. Recent advances in vision-language models (VLMs), which can generate scene descriptions for arbitrary images, combined with the availability of quantitative metrics for comparing linguistic descriptions, offer a new opportunity to model human scene understanding. We hypothesize that the primary bottleneck in human scene understanding and the driving source of variability in response times across scenes is the interaction between the foveated nature of the human visual system and the spatial distribution of task-relevant visual information within an image. Based on this assumption, we propose a novel image-computable model that integrates foveated vision with VLMs to produce a spatially resolved map of scene understanding as a function of fixation location (Foveated Scene Understanding Map, or F-SUM), along with an aggregate F-SUM score. This metric correlates with average (N=17) human RTs (r=0.47) and number of saccades (r=0.51) required to comprehend a scene (across 277 scenes). The F-SUM score also correlates with average (N=16) human description accuracy (r=-0.56) in time-limited presentations. These correlations significantly exceed those of standard image-based metrics such as clutter, visual complexity, and scene ambiguity based on language entropy. Together, our work introduces a new image-computable metric for predicting human response times in scene understanding and …

Background: Measurement of cortisol in saliva is used to quantify “biological stress,” as reflected in HPA axis activity. It is “noisy,” however, and strict guidelines are needed to optimally collect it. We have now carefully and simultaneously collected both saliva and hair cortisol, and lab-based probes of HPA reactivity and regulatory dynamics, yielding 180+ HPA axis data points across 6 days of home saliva collection, a dex suppression test, 4 lab probes (TSST, ACTH stimulation, dex-CRH, and metyrapone), and two hair collections.Methods: From this extensive biological data base, we have captured the full 180+ HPA data points on a single slide for each of 127 participants. We want to share these slides, to show our ISPNE audience more raw HPA data than they ever have seen before, in a form that highlights the consistencies and inconsistencies of our familiar methods, and the immense complexity of the patterns …

Background: Saliva cortisol has been used extensively to identify links between psychosocial factors and “biological stress.” However, few studies have attempted to determine the biological meaning of HPA measures from saliva, in terms of the HPA regulatory signaling dynamics that shape health consequences. This study utilized established lab probes of HPA neurobiology to add biological meaning to field friendly cortisol measures from saliva.Methods: 180+ HPA axis data points were collected from 142 participants, including 9 saliva samples/day over 6 days and these “laboratory” tests: Dexamethasone suppression test (negative feedback), dexamethasone/CRH stimulation test (feedback and pituitary sensitivity), ACTH stimulation test (adrenal sensitivity), metyrapone test (central drive), and TSST (psychosocial stress reactivity). Diurnal curve components examined include cortisol awakening response (CAR …

Background: Hair cortisol is used to track longitudinal, integrated cortisol secretion over time (a month or more). Like saliva cortisol, it has been linked to numerous psychosocial stressors. As with saliva measures, it has been little studied in relation to HPA regulatory signaling dynamics that shape health consequences. This study utilized laboratory probes of HPA neurobiology to study the biological meaning of saliva cortisol measures and to link hair cortisol levels to both saliva cortisol and lab probes.Methods: Saliva cortisol was comprehensively sampled over 6 days within a month, and hair samples were obtained to assess cortisol levels over that month. Over the same month, the following “laboratory” tests were also performed: The dexamethasone suppression test (negative feedback), the dexamethasone/CRH stimulation test (feedback and pituitary sensitivity), the ACTH stimulation test (adrenal sensitivity), the …

During the 2017 NBA playoffs, Celtics coach Brad Stevens was faced with a difficult decision when defending against the Cavaliers: "Do you double and risk giving up easy shots, or stay at home and do the best you can?" It's a tough call, but finding a good defensive strategy that effectively incorporates doubling can make all the difference in the NBA. In this paper, we analyze double teaming in the NBA, quantifying the trade-off between risk and reward. Using player trajectory data pertaining to over 643,000 possessions, we identified when the ball handler was double teamed. Given these data and the corresponding outcome (i.e., was the defense successful), we used deep reinforcement learning to estimate the quality of the defensive actions. We present qualitative and quantitative results summarizing our learned defensive strategy for defending. We show that our policy value estimates are predictive of points per possession and win percentage. Overall, the proposed framework represents a step toward a more comprehensive understanding of defensive strategies in the NBA.

This study applies dynamical and statistical modeling techniques to quantify the proliferation and popularity of trending hashtags on Twitter. Using time-series data reflecting actual tweets in New York City and San Francisco, we present estimates for the dynamics (i.e., rates of infection and recovery) of several hundred trending hashtags using an epidemic modeling framework coupled with Bayesian Markov Chain Monte Carlo (MCMC) methods. This methodological strategy is an extension of techniques traditionally used to model the spread of infectious disease. Using SIR-type models, we demonstrate that most hashtags are marginally infectious, while very few emerge as “trending”. In doing so we illustrate that hashtags can be grouped by infectiousness, possibly providing a method for quantifying the trendiness of a topic.

Created in 2006, Twitter is an online social networking service in which users share and read 140-character messages called Tweets. The site has approximately 288 million monthly active users who produce about 500 million Tweets per day. This study applies dynamical and statistical modeling strategies to quantify the spread of information on Twitter. Parameter estimates for the rates of infection and recovery are obtained using Bayesian Markov Chain Monte Carlo (MCMC) methods. The methodological strategy employed is an extension of techniques traditionally used in an epidemiological and biomedical context (particularly in the spread of infectious disease). This study, which addresses information spread, presents case studies pertaining to the prevalence of several “trending” topics on Twitter over time. The study introduces a framework to compare information dynamics on Twitter based on the topical area as well as a framework for the prediction of topic prevalence. Additionally, methodological and results-based comparisons are drawn between the spread of information and the spread of infectious disease.

Building upon previous studies that highlight considerable overlap in the influential factors of both obesity and depression, we employ a structural model to investigate the direct and indirect impacts of behavioural and demographic factors on obesity. We use new body mass index (BMI) to calculate the obesity status and find a significant relationship between an individual’s depression status and his/her obesity status. The results and simulations imply that demographic and behavioural factors can significantly influence the obesity status both directly and indirectly through their impact on depression. Therefore, this study suggests that models which do not account for these various pathways of influence are most likely misrepresenting the impact of these factors on obesity.

This paper investigates the socioeconomic determinants of obesity (as measured by BMI) in the United States. Logistic regression is employed on cross-sectional data from the 2011-2012 National Health and Nutrition Examination Survey (NHANES). The results show that, in general, holding other factors constant, individuals with a college diploma are less likely to be obese than those with a lesser education (except in extreme cases), married individuals are more likely to be obese than those that are not married, and females are more likely to be obese than males. Additionally, compared to white persons, Black and Hispanic persons have a greater probability of being obese, while Asians have a significantly lower probability of being obese. These findings are supported by the broader literature, in which different empirical techniques are often utilized.

In this paper, we explore a variety of models attempting to explain the pollution-income relationship (PIR). There has been much literature addressing the notion of an environmental Kuznets curve (EKC). Many researchers find an EKC relationship for certain pollutants, while others do not find evidence of an EKC relationship. There is also literature formally critiquing the EKC. We employ cross-sectional, panel, and time-series analysis to add insight into the relationship between economic growth and environmental degradation, a research area that is far from consensual and that has practical implications. We ultimately find that the clearest case of an EKC effect in our study arises in the analysis of organic water pollution, while there is modest evidence suggesting an EKC effect with regard to CO2, NO, and methane. We also present ample evidence suggesting an anti-EKC effect for PM10. Our analysis causes us to question the existence of an EKC effect throughout the environment in general.