Primary pulmonary hypertension (PPH) is a progressive disease of unknownetiology usually followed by death within 5 years after diagnosis. Although heart-lung or lung transplantation is nowoffered to patients with advancedPPH,adequate criteria assessing an accurate prediction of life expectancy in PPHhas been difficult to establish. The aims of this study were to identify the characteristic features associated with a poor prognosis in patients with PPH, and to attempt to establish an individual prognostic index that predicts with great accuracy survival or death ofPPH after one year, thereby helping to define criteria for patient selection for transplantation. In 1991, a retrospective nation-wide survey on PPHwas conducted in Japan, and the clinical and cardiorespiratory variables of223 PPH cases (female; 144, male; 79) in the period from 1980-1990 were obtained. The mean pulmonary arterial pressure (Ppa) was 57.5 ± 17.2 mmHg(mean ± SD), and the overall median survival time was 32.5 months since the first diagnostic catheterization. The characteristic features of 61 patients who died within one year of catheterization (Nonsurvivors group) were compared to 141 patients whosurvived one year or more from the time of catheterization (Survivors group). Amongseveral clinical and cardiorespiratory variables, heart rate, Ppa, right atrial pressure (Pra), stroke volume index (SI), pulmonary vascular resistance, and partial pressure of carbon dioxide (PaCO2) were significantly different between the two groups. As the independent factors, Ppa, Pra, SI, and PaCO2were selected for the multiple logistic analysis. Using a 0.7 probability cut-point to separate Nonsurvivors from Survivors, 84.6% of Nonsurvivors and Survivors could be correctly predicted from this logistic regression equation. Predictive equations like the present preliminary one can be used in the future to better assess life expectancy in patients with PPHin whomtransplantation will be considered. (Internal Medicine 38: 12-16, 1999)

The plateau phenomenon, wherein the loss value stops decreasing during the process of learning, has been reported by various researchers. The phenomenon was actively inspected in the 1990s and found to be due to the fundamental hierarchical structure of neural network models. Then, the phenomenon has been thought of as inevitable. However, the phenomenon seldom occurs in the context of recent deep learning. There is a gap between theory and reality. In this paper, using statistical mechanical formulation, we clarified the relationship between the plateau phenomenon and the statistical property of the data learned. It is shown that the data whose covariance has small and dispersed eigenvalues tend to make the plateau phenomenon inconspicuous.

No abstract

The dynamics of supervised learning play a main role in deep learning, which takes place in the parameter space of a multilayer perceptron (MLP). We review the history of supervised stochastic gradient learning, focusing on its singular structure and natural gradient. The parameter space includes singular regions in which parameters are not identifiable. One of our results is a full exploration of the dynamical behaviors of stochastic gradient learning in an elementary singular network. The bad news is its pathological nature, in which part of the singular region becomes an attractor and another part a repulser at the same time, forming a Milnor attractor. A learning trajectory is attracted by the attractor region, staying in it for a long time, before it escapes the singular region through the repulser region. This is typical of plateau phenomena in learning. We demonstrate the strange topology of a singular region by introducing blow-down coordinates, which are useful for analyzing the natural gradient dynamics. We confirm that the natural gradient dynamics are free of critical slowdown. The second main result is the good news: the interactions of elementary singular networks eliminate the attractor part and the Milnor-type attractors disappear. This explains why large-scale networks do not suffer from serious critical slowdowns due to singularities. We finally show that the unit-wise natural gradient is effective for learning in spite of its low computational cost.

This research work is aimed to develop autonomous bio-monitoring mobile robots, which are capable of tracking and measuring patients' motions, recognizing the patients' behavior based on observation data, and providing calling for medical personnel in emergency situations in home environment. The robots to be developed will bring about cost-effective, safe and easier at-home rehabilitation to most motor-function impaired patients (MIPs). In our previous research, a full framework was established towards this research goal. In this research, we aimed at improving the human activity recognition by using contour data of the tracked human subject extracted from the depth images as the signal source, instead of the lower limb joint angle data used in the previous research, which are more likely to be affected by the motion of the robot and human subjects. Several geometric parameters, such as, the ratio of height to weight of the tracked human subject, and distance (pixels) between centroid points of upper and lower parts of human body, were calculated from the contour data, and used as the features for the activity recognition. A Hidden Markov Model (HMM) is employed to classify different human activities from the features. Experimental results showed that the human activity recognition could be achieved with a high correct rate.