Centers & Labs

RIKEN Center for Advanced Intelligence Project

Nonconvex Learning Theory Team

Team Leader: Takafumi Kanamori (Ph.D.)
Takafumi  Kanamori(Ph.D.)

The research team aims to develop machine learning algorithms using non-convex optimization problems and its theoretical foundations. Most of current learning algorithms are formalized as convex optimization problems. Though the convexity is advantageous for optimization, it is not necessarily preferable from the standpoint of statistical properties such as robustness and bias-reduction of estimators. The optimization of non-convex functions, however, encounters computational difficulty. We challenge to develop learning algorithm using non-convex optimization beyond the scope of convexity and to establish a theoretical foundation to analyze learning methods with non-convexity.

Main Research Field

Computer Science

Related Research Fields

Mathematics

Research Subjects

  • Theoretical analysis of learning algorithms using non-convex optimization
  • Statistical inference for large-scale models using divergence measures
  • Extension of multimodal information integration and information-transfer learning

Selected Publications

Papers with an asterisk(*) are based on research conducted outside of RIKEN.
  1. *T. Kanamori:
    “Efficiency Bound of Local Z-Estimators on Discrete Sample Spaces”
    Entropy, vol. 18, no. 7, pp. 273-287 (2016).
  2. *T. Kanamori, and H. Fujisawa:
    “Robust Estimation under Heavy Contamination using Unnormalized Models”
    Biometrika, vol. 102, no. 3, pp. 559-572 (2015).
  3. *A. Takeda, S. Fujiwara, and T. Kanamori:
    “Extended Robust Support Vector Machine Based on Financial Risk Minimization”
    Neural Computation, vol. 26, num. 11, pp. 2541-2569 (2014).
  4. *T. Kanamori , and H. Fujisawa:
    “Affine Invariant Divergences associated with Proper Composite Scoring Rules and their Applications”
    Bernoulli, vol. 20, No. 4, pp. 2278-2304 (2014).
  5. *T. Kanamori, and A. Takeda:
    “A Numerical Study of Learning Algorithms on Stiefel Manifold”
    Computational Management Science, vol. 11, Issue 4, pp 319-340 (2014).
  6. *A. Takeda, and T. Kanamori:
    “Using Financial Risk for Analyzing Generalization Performance of Machine Learning Models”
    Neural Networks, vol. 57, pp. 29-38 (2014).
  7. *T. D. Nguyen, M. C. du Plessis, T. Kanamori, and M. Sugiyama:
    “Constrained Least-Squares Density-Difference Estimation”
    IEICE Transactions on Information and Systems, vol. E97-D, no. 7, pp. 1822-1829 (2014).
  8. *T. Kanamori:
    “Scale-Invariant Divergences for Density Functions”
    Entropy, vol 16(5), pp. 2611-2628 (2014).
  9. *T. Kanamori, and M. Sugiyama:
    “Statistical Analysis of Distance Estimators with Density Differences and Density Ratios”
    Entropy, vol. 16 (2), pp. 921-942 (2014).
  10. *T. Kanamori, and A. Ohara:
    “A Bregman extension of quasi-Newton updates II: analysis of robustness properties”
    Journal of Computational and Applied Mathematics, vol. 253, pp. 104-122 (2013).

Contact information

Department of Computer Science and Mathematical Informatics, Nagoya University,
Furocho,
Chikusaku, Nagoya,
464-8601, Japan

Email: kanamori [at] is.nagoya-u.ac.jp