Comments on "Backpropagation algorithms for a broad class of dynamic networks".

Area of Science:

• Computational Neuroscience
• Machine Learning

Background:

• De Jesús proposed a general framework for dynamic neural networks, focusing on gradient and Jacobian calculations.
• The paper utilized backpropagation-through-time (BPTT) and real-time recurrent learning (RTRL) algorithms.
• Implementation of the proposed framework faced challenges due to identified errors.

Purpose of the Study:

• To identify and correct critical errors in De Jesús's framework for dynamic neural networks.
• To provide errata for gradient and Jacobian calculation methods.
• To facilitate accurate implementation of dynamic neural network algorithms.

Main Methods:

• Analysis of the original framework by De Jesús.
• Identification of specific errors in gradient and Jacobian calculations.
• Correction of algorithmic steps for BPTT and RTRL.

Main Results:

• Pinpointed inaccuracies in the original publication's methodology.
• Provided corrected formulas and explanations for gradient and Jacobian computations.
• Clarified the application of BPTT and RTRL in dynamic neural networks.

Conclusions:

• The corrections facilitate a more robust understanding and implementation of dynamic neural networks.
• Accurate gradient and Jacobian calculations are crucial for effective BPTT and RTRL.
• This commentary serves as a guide for researchers working with dynamic neural network frameworks.