Student Posters Repository

The L-band (1.4 GHz) microwave radiometer provides soil moisture (SM) information limited to around 5 cm soil depth. Deeper information (up to 10 cm) can be obtained using low frequency sensors (P-band: 0.3-1 GHz) with reduced effects of surface roughness and vegetation. The present study explored the capability of P-band and/or L-band singly or in combination, via direct assimilation of brightness temperature (Tb) into the Joint UK Land Environment Simulator (JULES) land surface model. JULES was driven by ERA-5 (ECMWF Reanalysis v5) meteorological forcing data and calibrated model parameters for bare soil. The assimilation framework consists of a radiative transfer model to convert simulated SM to Tb and an Ensemble Kalman Filter to generate an observation-corrected SM trajectory. This framework was first validated with an open loop experiment in a synthetic environment over Cora Lynn, Victoria, Australia for the period of 9th May to 14th June, 2019. Assimilation experiments with synthetic observations were then set up to investigate the sensitivity of i) number of ensembles, ii) observation error, iii) incidence angle, iv) assimilation interval, and iv) frequency bands. The diagnostics (Kalman gain and Jacobians) showed that P band was more sensitive to the deeper layers as compared to L-band. The results also showed substantial improvement in the soil moisture analysis state in both the dry and wet period of the study when both L- and P-band Tbs were assimilated. Further study will include investigating improvement in soil moisture estimates when using real field observations and assimilating Tb with multiple incidence angles. The development of global synthetic data for P- and L-band simulation would be performed using high-performance computing. This will be done to compare the results of the P-band Tb assimilation with the SMAP data in L-band and P-band AirMOSS data. 

Keywords: Ensemble Kalman filter, Tb assimilation, P-band, JULES Land surface model, Radiative Transfer Model

Quantum machine learning is in a period of rapid development and discovery, however it still lacks the resources and diversity of computational models of its classical complement. With the growing difficulties of classical models requiring extreme hardware and power solutions, and quantum models being limited by noisy intermediate-scale quantum (NISQ) hardware, there is a growing niche of solving both problems at once. In this work we introduce a new software model for quantum neuromorphic computing - a quantum leaky integrate-and-fire (QLIF) neuron, implemented as a compact high-fidelity quantum circuit, requiring only 2 rotation gates and no CNOTs. We use these neurons as building blocks in the construction of a quantum spiking neural network (QSNN), and a quantum spiking convolutional neural network (QSCNN), as the first of their kind. We apply these models to the MNIST, Fashion-MNIST, and KMNIST datasets for a full comparison with other classical and quantum models. We find that the proposed models perform competitively, with comparative accuracy, with efficient scaling and fast computation.

This project leverages topological constraints to automate and accelerate a new algorithm to compute a property of materials known as “spontaneous electric polarization.” This property is used to identify a class of materials known as “ferroelectrics,” which have applications in neuromorphic computing, non-volatile memory devices, and tunable capacitors. Beyond identifying ferroelectrics, the algorithm I am developing has the potential to efficiently generate high-quality training data for machine learning tasks, including exploring structure-property relationships to discover novel materials with nontrivial topological phases. This algorithm takes as input electronic wavefunctions from ab initio electronic structure codes, which solve high-dimensional eigenvalue equations to obtain electronic probability distributions based on a material's nuclear coordinates. On this poster, I demonstrate how enforcing topological constraints across the electronic wavefunctions during our calculation enables us to circumvent a numerical branch selection issue that plagues the current state-of-the-art procedure to calculate spontaneous electric polarization. I also show that this algorithm drastically reduces the computational resources necessary for computing this quantity. By automating and accelerating this calculation, we can increase the range, scope, and size of materials for which spontaneous electric polarization can be calculated given the capacity of modern HPC resources.

A significant fraction (>20%) of the stars in the Universe are not alone but gravitationally bound to other stars in binary systems. Stars in binaries interact through various processes, exchanging mass and influencing each other’s evolution. The interplay between stellar evolution and binary interactions can produce rare and peculiar objects, such as binary black holes that can merge later on due to the emission of gravitational waves. To study the properties of these rare populations, millions to billions of binary systems need to be simulated. Detailed stellar evolution and hydrodynamics codes are too computationally expensive for this task. Rapid population synthesis codes overcome this issue by using analytic and semi-analytic formalism to account for stellar and binary evolution. In the last three years, I developed a new, innovative, and flexible population synthesis code: SEVN.

In classical planet formation theory, planetary migration was not considered, and it was assumed that planets formed in situ (e.g., Hayashi et al., 1982; 1985). However, it is known that assuming in situ formation poses challenges in explaining the formation of Uranus and Neptune at the outer edges of the solar system within the age of the solar system (Levison & Stewart 2001). In addition, since the discovery of the first exoplanet in 1995, over 5000 exoplanets have been observed (Zhu & Dong 2021). Among these, there are many planets, such as Hot Jupiters with orbital periods of just a few days which cannot be explained without considering planetary migration.

The leading mechanisms of planetary migration include Type I migration, where planets migrate due to gravitational interactions with a gas disk, and planetesimal-driven migration (PDM), resulting from gravitational scattering between a planet and planetesimals. In Type I migration, planets typically lose their angular momentum and drift inwards, a phenomenon known as the “planet migration problem” (Ward 1986; Tanaka et al., 2002). Conversely, in PDM, it has been suggested that planets can universally migrate outwards by gaining angular momentum from planetesimals (Ida et al., 2000). However, due to the high computational cost, previous studies of PDM ignored gas drag and gravitational interactions among planetesimals, leaving it unclear whether planets can migrate dynamically through PDM (e.g., Kirsh et al., 2009; Minton & Levison 2014). 

In this study, we used a parallel N-body simulation code for planetary system formation, GPLUM (Ishigaki et al., 2021), and the supercomputer Fugaku to conduct the self-consistent global N-body simulation of PDM in which gravitational interactions among planetesimals, gas drag, and Type-I migration are taken into account. In our simulations, we placed a single protoplanet within a planetesimal disk and conducted a total of 570 simulations with varying parameters such as the mass of the protoplanet, the number of particles, and the presence of gas drag and Type I migration to statistically investigate the impact of PDM on the planet migration process.

Through our simulations, we found that a fair fraction of protoplanets are capable of migrating outward within the disk despite experiencing inward kick due to Type I torque. Furthermore, our results show that PDM can cause dynamic migration of protoplanets within the disk, even at a much smaller mass ratio between a protoplanet and planetesimals than previously thought. Our results suggest a solution to the longstanding “planet migration problem” and indicate that planets can grow while migrating actively inwards and outwards within the disk.

In this presentation, we will present the results of our self-consistent N-body simulations of PDM and discuss the effect of PDM on the process of planet migration.

Severe alcoholic hepatitis (sAH), a life-threatening liver condition, has a high short-term mortality and limited treatment options. The use of corticosteroids is currently the standard of care (SOC) therapy in these patients with an increased risk of infections. While research indicates a strong gut-liver association and potential use of microbial modulation as a treatment option for this disease, there is not enough microbiome-focused evidence to validate this claim. 

 Faecal Microbiota Transplantation (FMT) has shown some benefits in small uncontrolled studies in this cohort of patients. This study concerns a comprehensive investigation of changes in the gut microbiome of sAH patients before and after FMT using comparative and functional metagenomics. Utilizing high-performance computing (HPC) in microbial genomics optimizes storage, accelerates processing, and reduces computation time. 

We introduce CORTEX, an algorithmic framework designed for large-scale brain simulation. Leveraging the computational capacity of the Fugaku Supercomputer, CORTEX maximizes available problem size and processing performance. Our primary innovation, Indegree Sub-Graph Decomposition, along with a suite of parallel algorithms, facilitates efficient domain decomposition by segmenting the global graph structure into smaller, identically structured sub-graphs. This segmentation allows for parallel processing of synaptic interactions without inter-process dependencies, effectively eliminating data racing at the thread level without necessitating mutexes or atomic operations. Additionally, this strategy enhances the overlap of communication and computation. Benchmark tests conducted on spiking neural networks, characterized by biological parameters, have demonstrated significant enhancements in both problem size and simulation performance, surpassing the capabilities of the current leading open-source solution, the NEST Simulator. Our work offers a powerful new tool for the field of neuromorphic computing and understanding brain function.