This paper introduces the concept of canonical neural models, which are simplified representations of families of neural models that capture universal neurocomputational properties. The authors argue that studying canonical models allows for the derivation of results that are independent of specific model details, as these models can be transformed into canonical forms through continuous variable changes. This approach is particularly useful when analyzing complex neural systems where the exact form of the equations is not known. The paper discusses various types of neural models, including weakly connected networks, bistable and hysteretic dynamics, small amplitude oscillations, large amplitude oscillations, and excitable systems. For each of these, the authors derive canonical models that retain key features of the original systems. For example, the canonical phase model is shown to capture universal computational abilities such as oscillatory associative memory, regardless of the specific nature of the oscillators or the equations describing them. The paper also presents theorems that demonstrate the convergence properties of these canonical models, such as the Cohen-Grossberg-Hopfield convergence theorem and the synchronization theorem for oscillatory neural networks. These theorems show that canonical models can exhibit similar computational properties to more detailed models, even when the exact form of the original equations is not known. The authors conclude that canonical models provide a rigorous way to study the universal properties of neural systems, even when only partial information about the dynamics is available. They emphasize that the derivation of canonical models is an art rather than a science, as there is no general algorithm for doing so. However, much success has been achieved in deriving canonical models for weakly connected networks of neurons near bifurcations. The paper also discusses the importance of understanding the dynamics of different types of neural systems, including excitable systems, and how canonical models can be used to study these systems.This paper introduces the concept of canonical neural models, which are simplified representations of families of neural models that capture universal neurocomputational properties. The authors argue that studying canonical models allows for the derivation of results that are independent of specific model details, as these models can be transformed into canonical forms through continuous variable changes. This approach is particularly useful when analyzing complex neural systems where the exact form of the equations is not known. The paper discusses various types of neural models, including weakly connected networks, bistable and hysteretic dynamics, small amplitude oscillations, large amplitude oscillations, and excitable systems. For each of these, the authors derive canonical models that retain key features of the original systems. For example, the canonical phase model is shown to capture universal computational abilities such as oscillatory associative memory, regardless of the specific nature of the oscillators or the equations describing them. The paper also presents theorems that demonstrate the convergence properties of these canonical models, such as the Cohen-Grossberg-Hopfield convergence theorem and the synchronization theorem for oscillatory neural networks. These theorems show that canonical models can exhibit similar computational properties to more detailed models, even when the exact form of the original equations is not known. The authors conclude that canonical models provide a rigorous way to study the universal properties of neural systems, even when only partial information about the dynamics is available. They emphasize that the derivation of canonical models is an art rather than a science, as there is no general algorithm for doing so. However, much success has been achieved in deriving canonical models for weakly connected networks of neurons near bifurcations. The paper also discusses the importance of understanding the dynamics of different types of neural systems, including excitable systems, and how canonical models can be used to study these systems.