Update README.md (to include CNN description)

README.md

@@ -3,22 +3,45 @@
 [![Build Status](https://github.com/mgyoo86/TroyonBetaNN.jl/actions/workflows/CI.yml/badge.svg?branch=main)](https://github.com/mgyoo86/TroyonBetaNN.jl/actions/workflows/CI.yml?query=branch%3Amain)
 
 
-
-
 <!-- ABOUT THE PROJECT -->
 ## About The Package
-This package uses a pre-trained neural network (NN) to predict Troyon (no-wall) $\beta_N$ limits for a given equilibrium specified by sampling a few points on 2D plasma boundary and 1D profiles of the equilibrium safety factor and plasma pressure.
-Note that the NN model was developed and trained by Yueqiang Liu, *et al.*, where the MHD stability was analyzed by MARS-F code (see [[Y.Q. Liu, *et al.*, *PPCF* (2020)]](https://doi.org/10.1088/1361-6587/ab6f56) for details).
-This package employs the pre-trained NN model and provides some useful interfaces to make it compatible with the IMAS data structure and FUSE ecosystem.
-
-
-### Quick Examples
+This package uses pre-trained neural network (NN) models to predict Troyon (no-wall) $\beta_\mathrm{N}$ limits for a given equilibrium specified by sampling a few points on 2D plasma boundary and 1D profiles of the equilibrium safety factor and plasma pressure.
+There are two different NN models with either MLP or CNN architectures, whose training dataset were generated with the [CHEASE](https://doi.org/10.1016/0010-4655(92)90167-W) and [MARS-F](https://doi.org/10.1063/1.1287744) codes for the equilibrium and MHD stability, respectively.
+This package employs the pre-trained NN models and provides some useful interfaces to make it compatible with the IMAS data structure and [**FUSE**](https://github.com/ProjectTorreyPines/FUSE.jl) ![GitHub](https://i.sstatic.net/tskMh.png) ecosystem.
+
+## About NN models
+Note that the following two models are each valid for a specific range of plasma parameters (e.g., aspect ratio and shape).
+For example, the MLP model can cover a wider range of aspect ratio ($A\in[1.3, 4.0]$), while the CNN model only targets a specific aspect ratio ($A \sim 2.74$) since it is designed for the HL-2M tokamak.
+On the other hand, the MLP model can only hanlde the positive triangularity ($\sigma>0$), but the CNN model can deal with a range of triangularity ($\sigma \in [-0.6, 0.8]$), including the negative triangularity.
+
+### 1. MLP (Multi-Layer Perceptron) model
+Developed and trained by Yueqiang Liu, *et al.*. (see [[Y.Q. Liu, *et al.*, *PPCF* (2020)]](https://doi.org/10.1088/1361-6587/ab6f56) for details).
+This model can cover a wider range of plasma parameters
+* Input
+    * 2D boundary shape
+    * 1D safety factor
+    * 1D plasma pressure
+* Output
+    * Troyon $\beta_\mathrm{N}$ limits for $n=(1,2,3)$ toroidal modes
+
+### 2. CNN (Convolutional neural network) model
+Developed and trained by Yifei Zhao, *et al.*. (see [[Y.F. Zhao, *et al.*, *PPCF* (2022)]](https://doi.org/10.1088/1361-6587/ac4524) for details)
+* Input
+    * 2D boundary shape
+    * 1D safety factor
+    * 1D plasma pressure
+    * Internal inductance ($l_i$)
+    * Pressure Peaking Factor (PPF)
+* Output
+    * Troyon $\beta_\mathrm{N}$ limit for $n=1$ toroidal modes
+
+
+
+## Quick Examples
 ```julia
 # Assuming that "FUSE" and "dd" are already in your scope
 
-using TroyonBetaNN
-
-TBNN = TroyonBetaNN
+import TroyonBetaNN as TBNN
 
 # The following function cacluates Troyon limits for all equilibrium time_slices in dd,
 # and returns Vector{Troyon_Data}, which has all information about the result
@@ -62,19 +85,34 @@ Sample points for a FPP equilibrium:
 
 
 
-
-
-
 <!-- CONTACT -->
 ## Contact
 Min-Gu Yoo [![Linkedin](https://i.sstatic.net/gVE0j.png)](https://www.linkedin.com/in/min-gu-yoo-704773230) (General Atomics)  [email protected] \
 Yueqiang Liu (General Atomics) [email protected]
 
 
 
-<!-- ACKNOWLEDGMENTS -->
+## ACKNOWLEDGMENTS
+We acknowledge Yeifei Zhao (<[email protected]>) for graciously providing the trained CNN models for this project.
+
 ## References
 
-* Yueqiang Liu, Lang Lao, Li Li, and A D Turnbull, *Plasma Phys. Control. Fusion* **62** (2020) 045001 \
+* **[MLP model]**\
+ Yueqiang Liu, Lang Lao, Li Li, and A D Turnbull, *Plasma Phys. Control. Fusion* **62** (2020) 045001 \
    ["Neural network based prediction of no-wall βN limits due to ideal external kink instabilities"](https://doi.org/10.1088/1361-6587/ab6f56)
-<!-- <p align="right">(<a href="#readme-top">back to top</a>)</p> -->
+
+* **[CNN model]**\
+ Y F Zhao, Y Q Liu, S Wang, G Z Hao, Z X Wang, Z Y Yang, B Li, J X Li, H T Chen, M Xu, and X R Duan,  *Plasma Phys. Control. Fusion* **64** (2022) 045010 \
+["Neural network based fast prediction of βN limits in HL-2M"](https://doi.org/10.1088/1361-6587/ac4524)
+
+* **[CHEASE code]**\
+H. Lütjens, A. Bondeson, A. Roy, *Comput. Phys. Commun.* **69**  (1992) 287 \
+["Axisymmetric MHD equilibrium solver with bicubic Hermite elements"](https://doi.org/10.1016/0010-4655(92)90167-W)
+
+* **[MARS-F code]**\
+Y. Q. Liu, A. Bondeson, C. M. Fransson, B. Lennartson, C. Breitholtz,
+*Phys. Plasmas* **7** (2000) 3681
+["Feedback stabilization of nonaxisymmetric resistive wall modes in tokamaks. I. Electromagnetic model"](https://doi.org/10.1063/1.1287744)
+
+
+<!-- <p align="right">(<a href="#readme-top">back to top</a>)</p> --j>