Michael F. Zimmer
The paper [1] by Liu, Madhavan, and Tegmark sought to use machine learning methods to elicit known conservation laws for several systems. However, in their example of a damped 1D harmonic oscillator they made seven serious errors, causing both their method and result to be incorrect. In this Comment, those errors are reviewed.
Michael F. Zimmer
This paper begins with a dynamical model that was obtained by applying a machine learning technique (FJet) to time-series data; this dynamical model is then analyzed with Lie symmetry techniques to obtain constants of motion. This analysis is performed on both the conserved and non-conserved cases of the 1D and 2D harmonic oscillators. For the 1D oscillator, constants are found in the cases where the system is underdamped, overdamped, and critically damped. The novel existence of such a constant for a non-conserved model is interpreted as a manifestation of the conservation of energy of the {\em total} system (i.e., oscillator plus dissipative environment). For the 2D oscillator, constants are found for the isotropic and anisotropic cases, including when the frequencies are incommensurate; it is also generalized to arbitrary dimensions. In addition, a constant is identified which generalizes angular momentum for all ratios of the frequencies. The approach presented here can produce {\em multiple} constants of motion from a {\em single}, generic data set.
Michael F. Zimmer
The problem of determining the underlying dynamics of a system when only given data of its state over time has challenged scientists for decades. In this paper, the approach of using machine learning to model the updates of the phase space variables is introduced; this is done as a function of the phase space variables. (More generally, the modeling is done over functions of the jet space.) This approach (named FJet) allows one to accurately replicate the dynamics, and is demonstrated on the examples of the damped harmonic oscillator, the damped pendulum, and the Duffing oscillator; the underlying differential equation is also accurately recovered for each example. In addition, the results in no way depend on how the data is sampled over time (i.e., regularly or irregularly). It is demonstrated that a regression implementation of FJet is similar to the model resulting from a Taylor series expansion of the Runge-Kutta (RK) numerical integration scheme. This identification confers the advantage of explicitly revealing the function space to use in the modeling, as well as the associated uncertainty quantification for the updates. Finally, it is shown in the undamped harmonic oscillator example that the stability of the updates is stable $10^9$ times longer than with $4$th-order RK (with time step $0.1$).
Michael F. Zimmer
It has long been a goal to efficiently compute and use second order information on a function ($f$) to assist in numerical approximations. Here it is shown how, using only basic physics and a numerical approximation, such information can be accurately obtained at a cost of ${\cal O}(N)$ complexity, where $N$ is the dimensionality of the parameter space of $f$. In this paper, an algorithm ({\em VA-Flow}) is developed to exploit this second order information, and pseudocode is presented. It is applied to two classes of problems, that of inverse kinematics (IK) and gradient descent (GD). In the IK application, the algorithm is fast and robust, and is shown to lead to smooth behavior even near singularities. For GD the algorithm also works very well, provided the cost function is locally well-described by a polynomial.
Michael F. Zimmer
The purpose of this paper is to improve upon existing variants of gradient descent by solving two problems: (1) removing (or reducing) the plateau that occurs while minimizing the cost function, (2) continually adjusting the learning rate to an "ideal" value. The approach taken is to approximately solve for the learning rate as a function of a trust metric. When this technique is hybridized with momentum, it creates an especially effective gradient descent variant, called NeogradM. It is shown to outperform Adam on several test problems, and can easily reach cost function values that are smaller by a factor of $10^8$, for example.
Michael F. Zimmer
A different parametrization of the hyperplanes is used in the neural network algorithm. As demonstrated on several autoencoder examples it significantly outperforms the usual parametrization, reaching lower training error values with only a fraction of the number of epochs. It's argued that it makes it easier to understand and initialize the parameters.