Two complementary approaches to supersymmetry at the CERN Large Hadron Collider are presented. The first is a grand combination of 22 ATLAS Run-1 searches, reinterpreted in the 19-dimensional phenomenological MSSM to provide a comprehensive picture of the state of supersymmetry constraints — the first such summary from an experimental collaboration. The results incorporate dark matter, heavy flavour, and precision electroweak observables, and were featured in the CERN Courier. The second is a direct search for new physics in events with large jet multiplicities arising from gluino pair production, using 13 TeV data and novel jet substructure techniques to extend limits on the gluino mass up to 1600 GeV.

In 2014 the MEGS-A instrument aboard NASA's Solar Dynamics Observatory malfunctioned, resulting in the loss of solar EUV measurements critcal for space weather prediction. We develop a hybrid deep learning architecture to "virtualize" the EVE experiment by combining a linear component with a CNN (EfficientNet backbone) and using solar imagery data from AIA and magnetograms from HMI. Our model improves upon the state-of-the-art for solar irradiance prediction.

A deep-learning approach to searching for new physics using the novel tensor-attention mechanism to solve the combinatorial problem of matching jets to partons. The first paper to pioneer this new method in an ATLAS analysis, setting the strongest limits on R-parity-violating supersymmetry models in some regions of parameter space.

A scalable, parallelised climate modelling tool that systematically explores hypothetical planetary atmospheres on Earth-like planets around solar-type stars. By varying the abundances of biologically mediated gases (O₂, CO₂, H₂O, CH₄, H₂, N₂), the tool generates datasets of steady-state atmospheres now publicly available on the NASA Exoplanet Archive, enabling habitability assessments and ML applications in atmospheric science.

An expoloration of physics-informed machine learning techniques to solve combinatorial problems relevant to searching for new fundamental particles with LHC data. Using a neural network that incorporates a Lorentz-equivariant layer to extract high-dimensional correlations, we demonstrate significant improvement over traditional methods on the benchmark problem of jet-to-parton assignment in R-parity-violating SUSY decays.

A highly novel approach to searching for new physics by exploiting charge-flavour symmetry, which should be conserved in the Standard Model. Using 139 fb⁻¹ of 13 TeV data, the search provides model-independent constraints on the cross-section ratio at the 2% level, and sets limits on RPV smuons (up to 640 GeV) and scalar leptoquarks (below 1880 GeV). This paper demostrates the first use of the powerful charge-flavour symmetry technique in an LHC analysis, and improved sensitivity of searches to new physics to never seen before levels.

As part of the Phase-II upgrade of the ATLAS detector, a new all-silicon inner tracker will be constructed. This paper documents the extensive prototyping programme that produced around 100 barrel-type modules using the ABC130/HCC130 chip set, establishing the assembly procedures and quality control protocols to be used across multiple production sites for the eventual construction of 18,000 modules.

A study of a novel tracker design for the proposed Future Circular Collider (FCC-hh), which would face extreme pileup of ~1000 interactions per bunch crossing. A triplet of tracking layers is shown to reconstruct tracks with close to 100% efficiency and only 0.5% fake rate for high-momentum tracks, achieving 50% signal acceptance at a 100 kHz trigger rate.

A search for new physics in events containing 11 or more high-momentum jets, using 36.1 fb⁻¹ of 13 TeV data collected by the ATLAS detector. The analysis employs novel techniques including large-radius jet mass sums and b-tagging requirements. This was the most sophisticated search for supersymmetry in the multijet series at the time of publication.

A search for new phenomena in events with high jet multiplicity and missing transverse momentum using 3.2 fb⁻¹ of data collected in 2015. This was the first search for supersymmetry to be published with 13 TeV data from the LHC.

An update of the multijet search for new physics using 18.2 fb⁻¹ of data, introducing jet substructure techniques to enhance the sensitivity of the search to new physics models.

A combination of 22 Run-1 searches for new physics, providing the most comprehensive summary of ATLAS sensitivity to supersymmetry. The results were interpreted in the 19-parameter phenomenological MSSM, making this the first study of its kind from an experimental collaboration.

A comprehensive summary of constraints on R-parity-violating supersymmetry using a combination of ATLAS analyses and the full Run-1 dataset.