Enforcing Imprecise Constraints on Generative Adversarial Networks for Emulating Physical Systems

Generative adversarial networks (GANs) were initially proposed to generate images by learning from a large number of samples. Recently, GANs have been used to emulate complex physical systems such as turbulent flows. However, a critical question must be answered before GANs can be considered trusted emulators for physical systems: do GANs-generated samples conform to the various physical constraints? These include both deterministic constraints (e.g., conservation laws) and statistical constraints (e.g., energy spectrum of turbulent flows). The latter have been studied in a companion paper (Wu et al., Enforcing statistical constraints in generative adversarial networks for modeling chaotic dynamical systems. Journal of Computational Physics. 406, 109209, 2020). In the present work, we enforce deterministic yet imprecise constraints on GANs by incorporating them into the loss function of the generator. We evaluate the performance of physics-constrained GANs on two representative tasks with geometrical constraints (generating points on circles) and differential constraints (generating divergence-free flow velocity fields), respectively. In both cases, the constrained GANs produced samples that conform to the underlying constraints rather accurately, even though the constraints are only enforced up to a specified interval. More importantly, the imposed constraints significantly accelerate the convergence and improve the robustness in the training, indicating that they serve as a physics-based regularization. These improvements are noteworthy, as the convergence and robustness are two well-known obstacles in the training of GANs.     

High resolution evoked potentials of cognition

Summary The precision of four methods of quantifying neuroelectric signals has been improved by increasing EEG spatial sampling, using up to 124 electrodes, and by accurate anatomical registration of the EEG with Magnetic Resonance Images (MRIs). One such method, equivalent dipole modeling, is a well-known form of source localization which is useful when the generator of the scalp recorded signal approximates a simple dipolar source, as is usually the case with early and mid-latency Evoked Potentials (EPs). Two methods of enhancing spatial detail which benefit from increased spatial sampling include the Laplacian Derivation and the Finite Element Deblurring method. The latter is a new technique which estimates the EP distribution at the superficial cortical surface. The fourth method, Evoked Potential Covariance, characterizes the spatiotemporal relationships among EP segments at different recording sites. This is useful when studying functional neural networks underlying higher cognitive functions. These methods are reviewed and examples of results of their application in recent experiments are presented.Supported by competitive grants from The National Institute of Mental Health, The National Institute of Neurological Diseases and Stroke, The Air Force Office of Scientific Research, The Office of Naval Research, The National Science Foundation NASA and The National Institute on Alcohol Abuse and Alcoholism of the United States Federal Government. Thanks to all members of the EEG Systems Laboratory and SAM Technology, past and present, for their vital contributions to the work presented here.This paper is based on an invited lecture given at the Tenth Anniversary Meeting of The Japanese Society for Brain Electromagnetic Topography held in Fukuoka, Japan in September, 1992.     

Widespread Horizontal Connections Arising from Layer 5/6 Border Inverted Cells in Rabbit Visual Cortex

Herein we describe the inverted cells [defined as those projection neurons having a major dendritic shaft abpially oriented (Bueno-López et al., Eur. J. Neurosci. , 3, 415, 1991)] originating a unique set of cortical connections characterized by extraordinarily widespread horizontal distribution. Single and multiple injections of wheatgerm agglutinin - horseradish peroxidase were made in areas 17 and 18 and the resulting retrograde labelling in the cortex was analysed. The findings were assessed in independent control experiments in which Fluoro-Gold was used as retrograde tracer. Following single injections in area 17 several separate patches of labelled cells comprising layers 2–6 were consistently found in area 18. In addition to these associational cells a number of labelled cells appeared at the layer 5/6 border but were distributed over most of the tangential extent of the visual occipital cortex. This widespread pattern was particularly striking in brains after multiple injections. In these brains a conspicuous band of labelled cells at the 5/6 border radiated from the injection sites, making up an apparently continuous horizontal sheet that intersected the striate - extrastriate boundary and merged with the patches of labelled cells in area 18 and beyond. Most of the cells in the 5/6 border band were inverted cells (82%; n = 2081). Injections in area 18 failed to produce such a widespread set of labelled cells in area 17. The functional significance of these connections furnished by the 5/6 border inverted cells remains to be determined, but their distribution would allow for convergent/divergent binding interactions both intra-areally (within area 17) and inter-areally (from area 18 to area 17).     

Stimulatory effect of ipriflavone on formation of bone-like tissue in rat bone marrow stromal cell culture

The effects of ipriflavone (IP) (10^–5 M) on bone formation were studied in stromal cells from the femoral bone marrow of young adult rats cultured for 21 days in the presence of -glycerophosphate and dexamethasone. Stereoscopic microscopy showed nodule formation after 14 days of culturing, and both the number and the size of the nodules increased with time. The alizarin-red-stained calcified area in the nodules in the IP group was nearly 4 times as large as that in the control after 21 days. Light and electron microscopy revealed the presence of many osteoblast-like cells with developed rough endoplasmic reticulum and Golgi apparatus in the nodules in the control group after 14 days, and a collagenous fibril network was seen among the cells. After 21 days, calcification of the dense collagenous fibril network and bone matrix-like tissue were observed in many nodules, resulting in the formation of bone-like tissue containing osteocyte-like cells. In the IP group, the collagenous fibril network area in the nodules was greater than that in the control after 14 days, and a further increase in both the dense collagenous fibril network area and calcified bone-like tissue area was observed after 21 days. These findings indicate that IP stimulates bone-like tissue formation in the rat bone marrow stromal cell culture, suggesting that the promotion of collagen production by osteoblasts is involved in the stimulation of bone-like tissue formation by IP.     

Effects of antipsychotic drugs on cytokine networks

It has been known since the 1950s that phenothiazines have immunomodulatory effects. This review summarizes recent evidence suggesting that antipsychotic drugs, in particular chlorpromazine and the atypical compound clozapine, influence the production of cytokines. Cytokines, organized in networks of related peptides with pleiotropic functions, are pivotal humoral mediators of infection and inflammation, and they play an important role in hematopoiesis and autoimmunity. Therefore, the effects of antipsychotic drugs on cytokine networks are important for the understanding of immune-mediated side effects of these drugs, e.g. agranulocytosis. In addition, modulation of cytokine production by antipsychotic agents suggests that these drugs might be useful for the treatment of diseases which primarily involve the immune system. Moreover, because cytokines are known to have numerous effects on the CNS, they may mediate effects of antipsychotic drugs on brain functions. Finally, the influence of antipsychotic drugs on cytokine networks is an important confounding factor in studies investigating disease-related immunopathology in psychiatric disorders. This review provides a synopsis of the data published on these topics and outlines future research perspectives.     

人工神经网络紫外光谱法测定氯芬待因片含量

目的:对紫外光谱重叠的氯芬待因片进行多组分不经分离的含量测定.方法:应用前向型误差反向传播人工神经网络对氯芬待因片进行含量测定.结果:3个模拟样品平均回收率和RSD分别为磷酸可待因98.68%,3.40%;双氯芬酸钠100.01%,0.94%.结论:该方法测定结果准确,性能良好,对于吸收光谱重叠的药物含量测定有一定的参考价值.     

Prognosis in node-negative primary breast cancer: a neural network analysis of risk profiles using routinely assessed factors.

BACKGROUND: The present study investigated complex time-dependent effects of routinely assessed factors on the risk of breast cancer recurrence over follow-up time, with a partial logistic artificial neural network (PLANN) model. PATIENTS AND METHODS: PLANN was applied to data from 1793 patients with node-negative breast cancer, not submitted to any adjuvant treatment and with a minimal potential follow-up of 10 years. RESULTS: The shape of the hazard function changed according to histology, which showed a time-dependent effect, partly modulated by estrogen receptors (ERs). Age and progesterone receptors (PgR) showed protective effects; the latter was more evident for short follow-up and high ER values. Tumour size and ER content showed time-dependent unfavourable effects at early and long follow-up times, respectively. Predicted values of disease recurrence probability at 2 years of follow-up showed that low steroid-receptor content, young age and large tumour size were associated with the highest risk of relapse. Although the oldest patients with high ER content seem to be those most protected overall, high risk predictions tend to spread also to higher steroid-receptor contents, intermediate ages and small tumour size, with an increase in follow-up time. CONCLUSION: PLANN with suitable visualisation techniques provided thorough insights into the dynamics of breast cancer recurrence for improving individual risk staging of node-negative breast cancer patients.     

焦炉立火道温度的智能集成软测量方法及其应用

在分析焦炉立火道温度特性的基础上,提出了一种基于线性回归(LR)和监督式分布神经网络(SDNN)的立火道温度智能集成软测量方法。通过特性分析提出了典型蓄热室的选取原则,并利用级数验证了选取的有效性。为反映蓄顶温度与立火道温度的关系,首先分别建立了一元、二元和十二元LR模型,并通过智能集成将3个模型的输出进行有机融合;然后在对样本监督式聚类后,利用SDNN获得各个子网的综合输出;最后由专家协调器协调LR和SDNN的输出,得到立火道温度的软测量值。实际运行结果验证了该方法的有效性。     

利用Alopex改进的粒子群优化算法及其在软测量建模中的应用

提出了一种利用A lopex算法改进的粒子群优化算法,并将其应用于神经网络的建模中。改进的粒子群优化算法改善了粒子群优化算法摆脱局部极小点的能力,对典型函数的测试和基于神经网络的软测量建模表明:改进算法的全局搜索能力有了显著提高,特别是对多峰函数能够有效地避免早熟收敛问题。     

The Basis of Hyperspecificity in Autism: A Preliminary Suggestion Based on Properties of Neural Nets

This article reviews a few key ideas about the representation of information in neural networks and uses these ideas to address one aspect of autism, namely, the apparent hyperspecificity that is often seen in autistic children's application of previously acquired information. Hyperspecificity is seen as reflecting a possible feature of the neural codes used to represent concepts in the autistic brain.