A working notebook · 2025–present
Yufan Xia¹
where AI meets HPC, & HPC meets chemistry.
I am a PhD student in Computer Science at the University of Melbourne, advised by Giuseppe Barca and Adrian Pearce. I build machine learning tools for scientific computing — from neural network force fields to HPC-optimized linear algebra — to accelerate computational chemistry and materials discovery.
Fig. 1
Recent entries
Nov 2025
Nov 2025
Started PhD in Computer Science at the University of Melbourne.3
Fig. 2
Working figures
HPC · Machine Learning
Fig. 1.1
2024
ADSALA
ML-driven runtime tuning for BLAS Level 3 across multi-core CPUs. Predicts the optimal thread count from matrix shape and architecture — 1.5×–3.0× over default MKL/BLIS.
1.5–3.0×
speedup
BLAS L3
all kernels
IPDPS 24
best poster
Computational Chemistry · Neural Networks
Fig. 2.2
2026
NN-xTB
Hamiltonian-preserving neural augmentation of GFN2-xTB. Predicts parameter shifts so semi-empirical quantum chemistry reaches DFT-level accuracy at near-xTB cost.
4
WTMAD-2 kcal/mol
6×
lower error vs xTB
Nat. Comm.
in principle
Research interests
- Machine Learning in Scientific Computing
- Neural Network Force Fields
- High Performance Computing Optimization
- Quantum Chemistry and Computing
- Computational Materials Science
Education
Ph.D. in Computer Science
The University of Melbourne
2025-Present
MPhil. in Chemistry
The Chinese University of Hong Kong
2022-2024
Awards
- Best Poster Award at IEEE IPDPS 2024
- Hong Kong Government Scholarship, 2024
- RDMA Silver Medal, 2023