Machine Learning

Crater identification with Deep Learning

I lead highly influential and internationally recognized research on the identification of impact craters and other features on solar system surfaces using computer vision techniques. My team and I pioneered the field by creating DeepMoon, the world's first convolution neural network capable of identifying craters on the Moon with human-like accuracy.

ACID

These techniques discussed above were refined over the years culminating in the Astrophysical Circles Detector (ACID), a MaskRCNN-based framework capable of correctly identifying craters, boulders, cyclones, and holes in astrophysical images. ACID is currently being used in collaboration with the European Space Agency to analyze data from the Rosetta space probe that visited comet 67P/C-G between 2015 and 2018. It is moreover the backbone of the path planning efforts for the Emirates Lunar Mission, where I lead the team in charge of mapping hazardeous Lunar surface objects.

Using AI to better understand galaxy formation

ACID was used to map quasi-circular regions of gas over and under-density in real and simulated galaxies. These features were then used to find the best fitting simulated galaxies to the observed ones. A separate classifier based on VGG16 was then trained and used to confirm the results obtained using ACID.

Predicting planetary systems stability

In 2016, I was part of a team that trained a machine learning model to classify the stability of planetary systems, using simulations-generated dataset. Our model has an accuracy higher than 87\%, and is a 1000 times faster than centuries old techniques based on celestial mechanics. It is now used to analyze data from NASA's probe TESS.

---

---

Planet Formation Theory

Chemical composition of giant planets

My past research focused on connecting the chemical composition of giant planets to their formation process, both for the solar system and exoplanets. During my PhD research I developed a volatile transport model to understand the distribution of water and CO in the disk, and used it to explain the chemical composition of Hot Jupiter WASP 12b, in addition to Uranus and Neptune (Ali-Dib et al., 2014a,b; Ali-Dib & Lakhlani, 2017b). I then developed a model for the transport and chemical evolution of deuterium in an FU Ori disk, and used it to explain the D/H ratio measurements in chondrites and comets (Ali-Dib et al., 2015). During my postdoctoral research I focused on end-to-end planet formation models to link a giant planet's chemical composition to its entire formation track. I adapted a model incorporating a protoplanetary disk model with photoevaporation, solids and gas accretion, type I and II migration, and simplified volatile chemistry. I then used it to make predictions for the oxygen abundance and core mass of Jupiter, to be compared with Juno's measurements, and to predict the C/O ratios of Hot Jupiters (Ali-Dib, 2017a; Ali-Dib and Lakhlani, 2017a; Ali-Dib, 2017b). I also used the same model to interpret the occurrence rate of giant exoplanets inside 100 days, and show how it is compatible with planets forming preferentially on snowlines (Ali-Dib, Johansen, & Huang, 2017).

Accretional processes

Limits on pebble accretion

In Ali-Dib & Thompson (2020) we investigated the fate of icy-rocky pebbles accreted by an embedded low mass protoplanetary core surrounded by a hydrostatic envelope, using independent analytical and numerical approaches. We first showed that pebble entering the envelope will quickly be sandblasted down to dust size. This will result in an increase in dust abundance, and thus in opacity. We showed that this will lead to a runaway expansion of the convective zone till it reaches the Bondi radius. Pebbles accreted beyond this point will get quickly destroyed and then ejected diffusively from the envelope, halting the protoplanet's growth. We interpret these results as constrains on the pebble accretion rate, but also on the nature of convection in these envelopes.

Effects of entropy advection on accreting Super Earths

A major open question in planet formation theory is what stops Super Earths from accreting enough gas to transform into Jupiters. In Ali-Dib, Cumming, & Lin we investigated the role of disk entropy advection into the envelopes of gas accreting Super Earths. Using a 1D model that allows us to follow this process over long timescales, we showed that this effect pushes the radiative-convective boundary inward till it reaches the core. At this point the envelope is isothermal and fully radiative, quenching cooling and thus halting gas accretion.

Role of collisions in the formation of cold sub-Saturns

Giant collisions are widely thought to play a major role in planet formation. In Ali-Dib, Cumming, & Lin we investigate their effects on near-critical protoplanetary envelopes. Using an analytical approach we showed that impactors only 5% the mass of the combined core-envelope can completely strip the atmosphere inside-out through Eddington-like thermal winds. Assuming that collisions are more likely to take place during the dissipation of the protoplanetary disk into a lower density transition disk, the protoplanet will not have enough time to re-accrete a thick envelope. We proposed that this mechanism can explain the formation of cold sub-Saturns.

Astrophysical Dynamics

Origins of Kepler-419

In Petrovich, Wu & Ali-Dib (2019) and Ali-Dib & Petrovich (2020) we investigated the dynamical origins of Kepler-419. This system consists of two coplanar super-Jovian planets with eccentricities of 0.2 and 0.8 and apses that librate around anti-alignment. We showed how this unlikely architecture is a natural consequence of embedding the two planets in the inner cavity of an adiabatically dissipating massive disk, a common scenario during planet formation.

A color-eccentricity correlation in the Kuiper Belt

In Ali-Dib et al. (2021) we investigated the relation between the colors and orbits of Kuiper Belt objects. By reanalyzing the dataset of Marsset el. al (2019), we discovered a statistically significant paucity of red objects in the scattered disk. We then investigated the origins of this correlation by modeling the formation of the Kuiper belt through N-body simulations accounting for Uranus and Neptune's outward migration. We concluded that it is mainly due to the absence of sweeping second and third order mean motion resonances in the formation zone of red objects.