This web application provides estimates of the age of a star/star population using neural networks trained on stellar evolution models. It is based on Boin et al. 2026 (to be published in A&A). It is a web implementation of the NEST python package.
The neural networks were trained on the BaSTI, Dartmouth, MIST, PARSEC, Geneva and YaPSI isochrones. Don't use the YaPSI isochrones, they are broken right now. The neural networks are simple Multi-Layer Perceptrons with 4 hidden layers of 32,64,64 & 32 neurons, trained for 200 epochs. The training data size, i.e. the total number of points in the isochrone sets, depends on the model used.
They take as input the [M/H], σ([M/H]), MG, σ(MG), BP-RP & σ(BP-RP) of a star and return the age of the star and its distribution. The age estimate is given in terms of the mean, median and mode of the distribution, while the age uncertainty given is computed as the standard deviation in the age distribution.
You can click anywhere on the Color-Magnitude Diagram to set the MG & BP-RP of a star. You can also directly input its value using the input tags on the right. Changing any input will automatically trigger the computation of a new age estimate.
For each stellar evolution model, 3 sets of isochrones are used to display in the CMD at [M/H]=-2.0,-1.0 & 0.0 dex. Depending on the metallicity chosen, the closest set will be used.
The Number of samples input sets the number of Monte Carlo realisations used to compute the age distribution. 1000 is usually good enough to get a smooth distribution.
The Import csv file button lets you import a star population to get a global age estimate/distribution. Your csv file should have at least the following columns : [M/H], MG & BP-RP. For the uncertainties, you can either have additional columns named e(M/H), e(MG) & e(BP-RP) or directly input global uncertainties in the σ([M/H]), σ(MG) & σ(BP-RP) inputs.
Once you imported a stellar population and its age has been estimated, you can export the results as a csv using the Export csv file button. The csv file contains the original quantities for each star, with the median, mode, mean and standard deviation of the individual age distributions added.
The age distribution for a star population displays the sum of all the individual Monte Carlo realisations summed up.
Hovering over a specific age bin will higlight stars whose age estimate is within (fully opaque) or close (semi-transparent) to that bin. You can click on a bin to select it and investigate the CMD further without needing to be hovering. A checkbox lets you unselect the bin.
The code for this web application, along with the weights and biases of the trained neural networks, can be found here. Feel free to contribute.