My photo.


Po-Wei Wang

resume [pdf]

I am a fourth-year PhD student in the Machine Learning Department at Carnegie Mellon University, advised by Prof. Zico Kolter. Before that, I was an undergrad in the CSIE Department at National Taiwan University, working with Prof. Chih-Jen Lin on convergence properties of SVMs. My interests cover both theories and applications for convex/nonconvex optimization.

Research Experience

  • Global linear convergence for non-strongly convex problems [2]

    • The first global linear convergence rate for first-order methods on non-strongly convex problems.
    • Applied to cyclic coordinate descent methods for dual SVC and SVR, published in JMLR.
    • Solved the open problem of convergence rate of the Gauss-Seidel method on PSD matrices.
  • Global optimal convergence of low-rank SDPs on spherical manifolds [9] [11]

    • The first low-rank semidefinite method that provably converges to a global optimum without assumptions.
    • Random initialization, linear convergence in neighborhoods, scales to millions of vertices.
    • Orders of magnitude faster than other existing methods in experiments.
  • The common-direction solver for linear classification [6] [5]

    • Reuse gradient information to reduce communication time in distributed optimization.
    • Strictly decreasing, optimal linear convergence, and local quadratic convergence.
    • Outperform state-of-the-art first- and second-order methods in experiments.
  • Polynomial optimization methods for matrix factorization [8]

    • Solve bi-direction line-search in matrix factorization exactly via polynomial optimization.
    • Apply Durand-Kerner method in numerical optimization to accelerate the process.
    • Achieve empirical speedup and lower objective values in benchmarks.
  • Disciplined convex optimization by proximal and epigraph projection [4] [3]

    • Solve the conic program in {\sf cvxpy} by fast proximal and epigraph projection operators.
    • New algorithms for the implicit dual problem and piece-wise linear functions.
    • Orders of magnitude faster than existing general-purpose optimization solvers.



  1. SATNet: Bridging deep learning and logical reasoning using a differentiable satisfiability solver [code] [slide] [poster] [Sudoku dataset, parity dataset]
    Po-Wei Wang, Priya L. Donti, Bryan Wilder, Zico Kolter. ICML, 2019 (Best paper Honorable mention)
  2. Low-rank semidefinite programming for the MAX2SAT problem [code] [slide]
    Po-Wei Wang, J. Zico Kolter. AAAI, 2019
  3. Realtime query completion via deep language models [code]
    Po-Wei Wang, Huan Zhang, Vijai Mohan, Inderjit S. Dhillon, and J. Zico Kolter. SIGIR Workshop on eCommerce, 2018
  4. The Mixing method: low-rank coordinate descent for semidefinite programming with diagonal constraints [code] [poster]
    Po-Wei Wang, Wei-Cheng Chang, and J. Zico Kolter. arXiv preprint, 2017.
  5. Polynomial optimization methods for matrix factorization
    Po-Wei Wang, Chun-Liang Li, and J. Zico Kolter. AAAI, 2017.
  6. The Mixing method for Maxcut-SDP problem
    Po-Wei Wang and J. Zico Kolter. NIPS LHDS Workshop, 2016.
  7. Limited-memory common-directions method for distributed optimization and its application on empirical risk minimization [supp and exp code] [Implemented in Distributed LIBLINEAR]
    Ching-pei Lee, Po-Wei Wang, Weizhu Chen, and Chih-Jen Lin. SDM, 2017.
  8. The Common-directions Method for Regularized Empirical Risk Minimization [supp and exp code] [slide]
    Po-Wei Wang, Ching-pei Lee, and Chih-Jen Lin. JMLR, 2019.
  9. Epigraph projections for fast general convex programming [github] [slide]
    Po-Wei Wang, Matt Wytock, and J. Zico Kolter. ICML, 2016.
  10. Convex programming with fast proximal and linear operators [github]
    Matt Wytock, Po-Wei Wang and J. Zico Kolter. NIPS Optimization Workshop, 2015.
  11. Iteration Complexity of Feasible Descent Methods for Convex Optimization [slide]
    Po-Wei Wang and Chih-Jen Lin. JMLR, 15(2014), 1523-1548.
  12. Support Vector Machines in Data Classification: Algorithms and Applications.
    Po-Wei Wang and Chih-Jen Lin. CRC Press, 2014.
(See my Google Scholar for other publications.)

Past Projects


ITRS Test 3 (with Scott Tsai)

We created a web-based Linux environment with QEMU, python, and AJAX, in which students could run vim, gcc, python, and various tools. In addition, students can submit their code to a central judge server on appengine to evaluate their answer. The system is designed for "Programming 101" courses.

Robocup NTU PAL (with members in NTU RPAL Lab)

We rewrite the vision, behavior, and motion system for Nao robot from BHuman codebase so that the robots can share localization information and perform wide-angle kick without moving too far. As a result, we won the 3rd place in the standard platform league.

MIA (with Chun-Liang Li)

The MIA system uses two PS3Eyes to perform triangulation on feature points, and one camera to record the video in H264. We wrote a 3D SLAM library to perform self-localization and consistent mapping of objects. It was a term project for VFX course.

LLDVK (with Logan Chien)

LLDVK is a translator which translates C codes into Android bytecodes (dalvik). It is built on the LLVM compiler infrastructure, and it served as a term project for the compiler course.


I have been a reviewer for Journal of Machine Learning Research (JMLR), Data Minining and Knowledge Discovery (DAMI), IEEE Conference on Decision and Control (CDC), Neural Information Processing Systems (NIPS), International Conference on Machine Learning (ICML), IEEE Transactions on Knowledge and Data Engineering (TKDE), and Neurocomputing.


  • Scholarship from Amazon for works on deep query completion.
  • Our team NTU RobotPal earned the 3rd place in Robocup Standard Platform League.
  • Our team are Champions for both tracks in the Yahoo KDD Cup 2011 on Music Rating Prediction.
  • NTU Presidential Awards (x2), which is given to the top 5% students each semester.
  • Undergraduate scholarship from Macronix Education Foundation.
  • Grant and prize from the National Science Council for Creative Undergraduate Research.