We investigate jointly modelling age–year-specific rates of various causes of death in a multinational setting. We apply multi-output Gaussian processes (MOGPs), a spatial machine learning method, to smooth and extrapolate multiple cause-of-death mortality rates across several countries and both genders. To maintain flexibility and scalability, we investigate MOGPs with Kronecker-structured kernels and latent factors. In particular, we develop a custom multi-level MOGP that leverages the gridded structure of mortality tables to efficiently capture heterogeneity and dependence across different factor inputs. Results are illustrated with datasets from the Human Cause-of-Death Database (HCD). We discuss a case study involving cancer variations in three European nations and a US-based study that considers eight top-level causes and includes comparison to all-cause analysis. Our models provide insights into the commonality of cause-specific mortality trends and demonstrate the opportunities for respective data fusion.

The Lee–Carter (LC) model is a basic approach to forecasting mortality rates of a single population. Although extensions of the LC model to forecasting rates for multiple populations have recently been proposed, the structure of these extended models is hard to justify and the models are often difficult to calibrate, relying on customised optimisation schemes. Based on the paradigm of representation learning, we extend the LCmodel to multiple populations using neural networks, which automatically select an optimal model structure. We fit this model to mortality rates since 1950 for all countries in the Human Mortality Database and observe that the out-of-sample forecasting performance of the model is highly competitive.

In order to investigate the origin of multiple populations in globular clusters (GCs), we have constructed new chemical evolution models for proto-GCs where the supernova blast waves undergo blowout without expelling the ambient gas. Chemical enrichments in our models are then dictated by the winds of massive stars together with the asymptotic-giant-branch stars ejecta. We find that the observed Na-O anti-correlation can be reproduced when multiple episodes of starburst and enrichment are allowed to continue in proto-GCs. The “mass budget problem” is mostly resolved by our models without ad-hoc assumptions on star formation efficiency, initial mass function, and significant loss of first-generation stars. Interestingly, ages and chemical abundances predicted by this chemical evolution model are in good agreements with those independently obtained from our stellar evolution model for the horizontal-branch. We also discuss observational evidence for the GC-like multiple populations in the Milky Way bulge.

Recent investigations of multiple stellar populations in globular clusters (GCs) suggest that the horizontal-branch (HB) morphology and mean period of type ab RR Lyrae variables are mostly sensitive to helium abundance, while the star formation timescale has the greatest effect on our chemical evolution model constructed to reproduce the Na-O anti-correlation of GCs. Therefore, by combining the results from synthetic HB model with those from chemical evolution model, we could put better constraints on star formation history and chemical evolution in GCs with multiple populations. From such efforts made for four GCs, M4, M5, M15, and M80, we find that consistent results can be obtained from these two independent models.

The origin of the chemical anomalies in star clusters is still an open question, although much effort has been employed both from a theoretical and observational point of view. The exploration of the dependence of such multiple stellar populations based on certain cluster properties (e.g. mass, age, metallicity) has represented a compelling line of investigation so far. Here I report an overview of the results obtained from our latest surveys aimed at characterising the phenomenon of chemical variations in star clusters that are much younger with respect to the ancient globular clusters. The fundamental question we are asking is whether these abundance patterns are only restricted to the old massive clusters; and if not, is there a difference between young and old objects?

The Small Magellanic Cloud (SMC) presents us with a unique opportunity to study in detail the effect of environmental processes (interaction with the LMC and the Milky Way) on its star formation history. With the 6.5m Magellan Telescope at the Las Campanas Observatory in Chile we have acquired deep B and I images in four 0.44 degree fields covering a large part of the main body of the SMC, yielding accurate photometry for 1,068,893 stars down to ~24th magnitude, with a spatial resolution of 0.201 arcsec/pixel. Colour-magnitude diagrams and luminosity functions (corrected for completeness) have been constructed, yielding significant new results that indicate at least two discrete star formation events around 2.7 and 4-5 Gyr ago.

A number of scenarios have been put forward to explain the origin of the chemical anomalies (and resulting complex colour-magnitude diagrams) observed in globular clusters (GCs), namely the AGB, Fast Rotating Massive Star, Very Massive Star, and Early Disc Accretion scenarios. We compare the predictions of these scenarios with a range of observations (including young massive clusters (YMCs), chemical patterns, and GC population properties) and find that all models are inconsistent with observations. In particular, YMCs do not show evidence for multiple epochs of star-formation and appear to be gas free by an age of ~ 3 Myr. Also, the chemical patterns displayed in GCs vary from one to the next in such a way that cannot be reproduced by standard nucleosynthetic yields. Finally, we show that the “mass budget problem” for the scenarios cannot be solved by invoking heavy cluster mass loss (i.e. that clusters were 10-100 times more massive at birth) as this solution makes basic predictions about the GC population that are inconsistent with observations. We conclude that none of the proposed scenarios can explain the multiple population phenomenon, hence alternative theories are needed.