Search for: All records

In this paper, we propose conjugate energy-based models (CEBMs), a new class of energy-based models that define a joint density over data and latent variables. The joint density of a CEBM decomposes into an intractable distribution over data and a tractable posterior over latent variables. CEBMs have similar use cases as variational autoencoders, in the sense that they learn an unsupervised mapping from data to latent variables. However, these models omit a generator network, which allows them to learn more flexible notions of similarity between data points. Our experiments demonstrate that conjugate EBMs achieve competitive results in terms of image modelling, predictive power of latent space, and out-of-domain detection on a variety of datasets. 

Variational autoencoders (VAEs) optimize an objective that comprises a reconstruction loss (the distortion) and a KL term (the rate). The rate is an upper bound on the mutual information, which is often interpreted as a regularizer that controls the degree of compression. We here examine whether inclusion of the rate term also improves generalization. We perform rate-distortion analyses in which we control the strength of the rate term, the network capacity, and the difficulty of the generalization problem. Lowering the strength of the rate term paradoxically improves generalization in most settings, and reducing the mutual information typically leads to underfitting. Moreover, we show that generalization performance continues to improve even after the mutual information saturates, indicating that the gap on the bound (i.e. the KL divergence relative to the inference marginal) affects generalization. This suggests that the standard spherical Gaussian prior is not an inductive bias that typically improves generalization, prompting further work to understand what choices of priors improve generalization in VAEs. 

The construction industry still leads the world as one of the sectors with the most work-related injuries and worker fatalities. Recent studies show that both a state of mindfulness and various personality traits contribute to individuals’ safety and work performance. This study examines the relationship between mindfulness and personality by measuring the mindfulness state of individuals against their personality traits. To achieve this objective, data were collected from a sample of 55 undergraduate students at George Mason University. Scores from the Big Five Inventory were ranked by each traits’ score (independent variable) and split into three groups: high, moderate, and low scores. The corresponding mindfulness scores (dependent variable) were analyzed to determine the relationship between high/low personality traits and mindfulness. Comparing the high/low groups using statistical analyses showed that three of the five personality traits—conscientiousness, agreeableness, and neuroticism—significantly correlate with higher mindfulness scores of individuals. As mindfulness has been shown to increase individual safety and work performance and to reduce stress, the results of this study help inform future work into translating personality and mindfulness characteristics into factors that predict specific elements of unsafe human behaviors. 

Cognitive processes have been found to contribute substantially to the human errors that lead to construction accidents. Working memory—a cognitive system with a limited capacity that is responsible for temporarily holding information available for processing—plays an important role in reasoning and decision-making. Since eye movements indicate where a worker directs his/her attention, tracking such movements provides a practical way to measure workers’ attention and comprehension of construction hazards. As a departure in construction industry research, this study correlates attentional allocation with working memory to assess workers’ situation awareness under different scenarios that expose workers to various hazards. To achieve this goal, this study merges research linking eye movements and workers’ attention with research focused on working-memory load and decision making and evaluates what, how, and where a worker distributes his/her attention while performing a task under different working-memory loads. Path analysis models then examined the direct and indirect effect of different working-memory loads on hazard identification performance. The independent variable (working-memory load) is linked to the dependent variable (hazard identification) through the set of mediators (attention metrics). The results showed that the high-memory load condition delayed workers’ hazard identification. The findings of this study emphasize the important role working memory plays in determining how and why workers in dynamic work environments fail to detect, comprehend, and/or respond to physical risks. 

One of the main contributors to the human errors that lead to catastrophic injuries in the construction workplace is the failure to identify hazards as a result of poor attention or cognitive lapses. To address this safety concern, the present study used eye-tracking technology to assess how the association between work experience and hazard identification may be mediated due to inattention. A mediation analysis was conducted and tested using a bias-corrected bootstrapping technique with 5000 resamples. The results estimate the direct and indirect effects of work experience on the hazard identification skills of construction workers observing varying hazardous conditions. The results of the mediation analysis confirm that inattention—demonstrated via inattentiveness toward hazards—mediates the relationship between work experience and hazard identification. Specifically, though work experience and dwell time positively correlate with hazard identification, the direct effect of work experience on hazard identification is attenuated with the inclusion of the mediator variables in the model, thus suggesting attentional impairment offsets the benefits of work experience. The outcomes of this study will enable researchers and safety practitioners to harness real-time eye-movement patterns to identify the precursors of cognitive failure, deficient attentional allocation, and poor visual search strategies, all of which may put workers at risk on construction sites. The results also facilitate the provision of personalized safety feedback to workers and the design of training interventions that will address unique performance deficiencies in workers to prevent the human errors that cause injuries in dynamic environments.