GPA: 4.75 / 5.00

Relevant Coursework: Eigenvalue of Random Matrices (A), Nonlinear Dynamics and Chaos (A), Fast Methods for Partial Differential Equations (A)

GPA: 5.00 / 5.00

Relevant Coursework: Parallel Computing & Scientific Machine Learning (A), Optimization Methods (A), Numerical Methods for Partial Differential Equations (A+), Introduction to Numerical Methods (A)

GPA: 3.89 / 4.00, rank 1 / 137, honored as Weiming Bachelor (top 1%)

Relevant Coursework: Introduction to Computation (97), Data Structure and Algorithms (96), Computational Physics (92), Ordinary Differential Equations (92), Mathematical Method in Physics (95), Advanced Mathematics I, II (92, 96), Advanced Algebra I, II (92, 94)

GPA: 4.00 / 4.00

Relevant Coursework: Introduction to Probability (A+), Applied Numerical Methods (A)

Advisor: Qilin He, Jiaqi Jiang (Platform Architecture)

• Worked on machine learning models and differentiable solvers.

Advisor: Yuan Lin (Deep Learning Compiler Team)

• Investigated the heuristics of choosing kernels and compilation parameters during GEMM + epilogue fusion, and improved the optimal tactic hit by 70%
• Accelerated the layer fusion compilation by 40%, by a combination of caching, multi-threading and reducing trial compilations

Advisor: Christopher Rackauckas & Alan Edelman

• Developing higher-order forward-mode automatic differentiation (AD) algorithms that scale linearly with the order, suitable for differential models like ODEs and PDEs
• Synthesizing code that is compiler-friendly and compatible with reverse-mode AD libraries like Zygote.jl, by aggressively specializing with compile-time type information
• Deriving higher-order differentiation rules automatically from first-order chain rules with symbolic computation and metaprogramming in Julia

Advisor: Christopher Rackauckas & Alan Edelman

• Joined the Enzyme project (enzyme.mit.edu), an automatic differentiation framework based on source code transformation at LLVM intermediate representation (IR) level, which can differentiate through all languages with a LLVM backend (e.g. Julia, C++, Fortran)
• Synthesized derivatives of BLAS/LAPACK kernels with generated kernels and calls to other BLAS/LAPACK kernels, and performed extensive optimizations based on linear algebra relations
• Outperformed other high-level AD frameworks in Julia with 1.3× speed on a linear algebra benchmark set

Advisor: Weinan E & Linfeng Zhang

• Modeled the exchange-correlation density functional in generalized Kohn-Sham theory with deep neural networks and descriptors from density matrices
• Established a comprehensive theory for using physical quantities data that depends on the functional minimization result to train the functional model, in other words, addressed the “differentiate through argmin” problem
• Implemented the training process with multiple types of physical quantity data, such as energy band structure and dipole moment
• Improved the accuracy and generalization performance of the model, obtained an average energy error of 0.06 kcal/mol on a test set that includes 1200 water molecule configurations labeled with SCAN0 functional (48% less than previous methods)

Advisor: Teresa Head-Gordon & Lin Lin

• Investigated reactive chemial systems with fluctuating charges described with differential-algebraic equations
• Developed an extended Lagrangian scheme to replace the algebraic part with differential dynamics of an extended system, thereby eliminating the expensive charge-equilibration step (i.e. algebraic equation solving) in simulation
• Proved the correctness of this scheme both theoretically and practically (the modified simulation still reproduced statistic and dynamic properties of that system)