A Trust Region Method for the Optimization of Noisy Functions .
Proposed and implemented a new trust region algorithm under noise; proved for global convergence.
Using mathematics to solve problems.
Optimization, Quantitative Research, Machine Learning and Beyond.
Pursuing my PhD in Applied Mathematics degree under the supervision of National Academy of Engineering member, Prof. Jorge Nocedal .
- We are particularly interested in developing
machine learning and nonlinear optimization algorithms and softwares.
- We are developing algorithms for training Deep Neural Networks and handeling noise in PDE robust design problems.
- Previously worked on Wavelets and applications in Finance, Numerical Partial Differential Equations and Computational Fluid Machanics in Biophysics.
Quant Strategist Intern at Virtu and Applied Scientist Intern at Amazon AWS AI Research Labs.
- Experience in High Frequency Trading alpha signal research and hedge fund style investing.
- Experience in Industry Research Labs machine learning research.
Recent News
- I will present our work on robust SQP algorithms in a invited talk at the INFORMS Optimization Conference in Houston, TX in March.
- Our paper, 'Noise-Tolerant Optimization Methods for the Solution of A Robust Design Problem' has recently been released online, [link] .
- Our paper, 'A trust region method for noisy unconstrained optimization' has recently appeared in the Journal of Mathematical Programming, [link] .
- I chaired a few sessions and presented our work at SIAM Conference on Optimization (OP23) in Seattle
- I co-taught a PhD level class, Convex Optimization this Spring
Recent Work
A Stochastic Line Search Algorithm for Training of Deep Neural Networks.
Implemented a stochastic line search algorithm for training of deep neural networks under PyTorch.
Emprical Comparison of Machine Learning algorithms for Image Classification.
Implemented SGD, ADAM, AdaGrad and SVRG on Gisette data using logistic regression. Performed scale-invariant analysis.
Instabilities in Active Contracting Networks
Produced a large Chebyshev spetral 2-D numerical PDE solver for a fluid mechanics model in biophysics. Derived a conservation law, performed stablity analysis.
Periodic Migration in a Physical Model of 1D Cell on Micro-patterns
Programmed finite difference and fourier spectral PDE solvers for solving a Phase-Field reaction-diffusion equation. Demonstrated various emergent phenomena.
Modeling International Links Between Interbank Offered Rates Through A Wavelet Analysis Approach
Modeled and analyzed the causality between interbank swap rates LIBOR, SHIBOR and MosPrime using a novel wavalet analysis approach.
Work in progress
- We are working on a new class of machine learning method, more details will be shared once NeurIPS and arXiv submissions are made.
- We are comparing noisy optimization settings for machine learning with PDE optimizations and related topics. Manuscript to appear soon online.
- We are developing constrained optimzation algorithms which are robust under noise. We are also developing Derivative Free Optimization solvers.
- We are always looking for interesting ML and other problems where optimization can help. Feel free to reach out to us.
We primarily work in Deep Learning, but we are exploring ways to branch out.
Software and Technical Skills
- Python, C, Matlab.
- PyTorch, Tensorflow, scikit-learn, Keras, JAX, LAPACK/LAPACKe, Pandas, BLAS, OpenMP, Slurm.
- Basic Knowledge in CUDA, SQL, Tableau, OpenRefine.
About Me
- I'm a PC builder and tech enthusiast. I liquid cooled my PC with fully customized loops.
- Recently completed Mensa IQ challenge. Scored 142, at 99.7 percentile, close to 3 standard deviation above average. (I don't buy it. lol)
- I'm an amateur photographer. My favoriate lens is my 85mm f/1.4 .
- A friend recently (April, 2023) asked Google's Bard (ChatGPT rival) if Bard knew about me and here's the (some accurate and others not so much) response:
