Quantifying Epistemic and Aleatoric Uncertainty in 3D U-Net Segmentation

This work shows a derivation of a multinomial probability function and quantitative measures of the data and epistemic uncertainty as direct output of a 3D U-Net segmentation network. A set of T1 brain MRI images were downloaded from the Connectome Project and segmented using FMRIB's FAST algorithm to be used as ground truth. A 3D U-Net neural network was trained with sample sizes of 200, 500, and 898 T1 brain images using a loss function defined as the negative logarithm of the likelihood based on a derivation of the definition of the multinomial probability function. From this definition, the epistemic (model) and aleatoric (data) uncertainty equations were derived and used to quantify maps of the uncertainty in data prediction. The epistemic and aleatoric uncertainty decreased based on the increasing number of training data used to train the neural network. The neural network trained with 898 volumes resulted in uncertainty maps that were high primarily in the tissue boundary regions. The uncertainty was averaged over all test data (connectome and tumor separately) and the epistemic uncertainty showed a decreasing trend, as expected, with increasing numbers of data used to train the model. The aleatoric uncertainty showed a similar trend, but it was less obvious, which was also expected as the aleatoric uncertainty is not expected to be as dependent on the number of training data. The derived data and epistemic uncertainty equations from a multinomial probability distribution are applicable for all 2D and 3D neural networks.

International Real Estate Review

Financial globalization has enabled investors to allocate some of their portfolio assets to foreign countries and alternative assets. This environment has also created an increase in investment in international real estate especially in the emerging markets. In this study, we investigate whether foreign real estate investors outperform domestic investors after controlling for property specific characteristics. Using property level transaction data of Korea from 2003 to 2016, we also examine the characteristics of commercial real estate investment associated with the probability of an acquirer being a foreign investor versus a domestic investor. The binary and multinomial probability models are used to test our research hypothesis and the structural equation model is applied to find the determinants of the internal rate of return. The result reveals foreign investors perform better than domestic investors in a holding period analysis. Furthermore, the findings support that foreign direct and indirect real estate investments are statistically significant to the age of the building, corporate bond and exchange rates, growth domestic product growth, and the equity market movement in the domestic market.

Non-Equiprobable Statistical Analysis of Misfires and Partial Burns for Cycle-to-Cycle Control of Combustion Variability

Cycle-to-cycle combustion variability (CV) in spark ignition internal combustion engines is amplified at high levels of exhaust gas recirculation (EGR) by sporadic partial burn and misfire events. A non-equiprobable cycle classification method, based on the magnitude of the indicated mean effective pressure (IMEP), was developed to discern and study the deterministic and stochastic components of cyclic CV. The time series analysis of experimental combustion cycles suggested that the occurrence of high energy release cycles right after misfires is the only deterministic component between consecutive cycles. This predictable behavior results from the retained air and fuel from the incomplete combustion cycle to the next. On the other hand, this study shows that the occurrence of partial burn and misfire cycles is the product of the stochastic component of cyclic CV with statistical properties similar to a multinomial probability distribution. It is demonstrated that observation of partial burns can increase the probability of observing a misfire when the conditional probability is used as the metric. Based on these findings, future work will be able to use the observation of partial burns alone to control the upper bound on the probability of misfire events. To this end, different metrics are proposed to control directly and indirectly the probability of misfires, and their advantages and disadvantages for feedback combustion control are discussed.

Seasonal Variations in Mood and Behavior in Romanian Postgraduate Students

To our knowledge, this paper is the first to estimate seasonality of mood in a predominantly Caucasian sample, living in areas with hot summers and a relative unavailability of air conditioning. As a summer pattern of seasonal depression was previously associated with a vulnerability to heat exposure, we hypothesized that those with access to air conditioners would have a lower rate of summer seasonal affective disorder (SAD) compared to those without air conditioning. A convenience sample of 476 Romanian postgraduate students completed the Seasonal Pattern Assessment Questionnaire (SPAQ), which was used to calculate a global seasonality score (GSS) and to estimate the rates of winter- and summer-type SAD. The ratio of summer- vs. winter-type SAD was compared using multinomial probability distribution tests. We also compared the ratio of summer SAD in individuals with vs. without air conditioners. Winter SAD and winter subsyndromal SAD (S-SAD) were significantly more prevalent than summer SAD and summer S-SAD. Those with access to air conditioners had a higher, rather than a lower, rate of summer SAD. Our results are consistent with prior studies that reported a lower prevalence of summer than winter SAD in Caucasian populations. Finding an increased rate of summer SAD in the minority of those with access to air conditioners was surprising and deserves replication.