An Adaptive Deep Learning Optimization Method Based on Radius of Curvature

An adaptive clamping method (SGD-MS) based on the radius of curvature is designed to alleviate the local optimal oscillation problem in deep neural network, which combines the radius of curvature of the objective function and the gradient descent of the optimizer. The radius of curvature is considered as the threshold to separate the momentum term or the future gradient moving average term adaptively. In addition, on this basis, we propose an accelerated version (SGD-MA), which further improves the convergence speed by using the method of aggregated momentum. Experimental results on several datasets show that the proposed methods effectively alleviate the local optimal oscillation problem and greatly improve the convergence speed and accuracy. A novel parameter updating algorithm is also provided in this paper for deep neural network.

Discussion on the damping factor and improvement of the Levenberg-Marquardt method

The ill-posed problem is the key obstacle to obtain the accurate inversion results in the geophysical inversion field, and the Levenberg-Marquardt1, 2(hereinafter referred to as the L-M method) method has been widely used as it can effectively improve the ill-posed problems. However, the inversion results obtained by the L-M method are usually stable but incorrect, the reason is that the damping factor in the L-M method is difficult to solve, and it is usually approximated with a positive constant by experience or through some fitting methods. This paper uses the binary gravity model to demonstrate that the damping factor in the L-M method cannot be regarded as a positive constant only, it should have the following characteristics: (i) the damping factor is a vector, not just a constant; (ii) the values of the vector are composed of both positive and negative constants, not just positive constants; (iii) the corresponding value in the vector is close or equal to ∞ when the corresponding density block’s value is close or equal to zero. Even if the above characteristics have been found in the L-M method, it is difficult to reasonably estimate the damping factor as the damping factor oscillate severely due to the third characteristic, and the improved L-M method is proposed which effectively avoids the damping factor’s severe oscillation problem. The strategy of obtaining the reasonable damping factor is given finally.

Improved FAST Device for Inherent Safety of Oxide-Fueled Sodium-Cooled Fast Reactors

The floating absorber for safety at transient (FAST) was proposed as a solution for the positive coolant temperature coefficient in sodium-cooled fast reactors (SFRs). It is designed to insert negative reactivity in the case of coolant temperature rise or coolant voiding in an inherently passive way. The use of the original FAST design showed effectiveness in protecting the reactor core during some anticipated transients without scram (ATWS) events. However, oscillation behaviors of power due to refloating of the absorber module in FAST were observed during other ATWS events. In this paper, we propose an improved FAST device (iFAST), in which a constraint is imposed on the sinking (insertion) limit of the absorber module in FAST. This provides a simple and effective solution to the power oscillation problem. Here, we focus on an oxide fuel-loaded SFR that is characterized by a more negative Doppler reactivity coefficient and higher operating temperature than the metallic-loaded SFR cores. The study is carried out for the 1000 MWth advanced burner reactor with an oxide fuel-loaded core during postulated ATWS events that are unprotected transient over power, unprotected loss of flow, and unprotected loss of the heat sink. It was found that the iFAST device has promising potentials for protecting the oxide SFR core during the various studied ATWS events.