Implicit Regularization in Matrix Factorization
Implicit Regularization in Matrix Factorization
复制标题
DOI：
10.1109/ita.2018.8503198
复制
发表时间：
2017-05
期刊：
影响因子：
--
通讯作者：
Suriya Gunasekar;Blake E. Woodworth;Srinadh Bhojanapalli;Behnam Neyshabur;N. Srebro
中科院分区：
文献类型：
--
作者：
Suriya Gunasekar;Blake E. Woodworth;Srinadh Bhojanapalli;Behnam Neyshabur;N. Srebro
当在矩阵$ x $上优化不确定的二次目标时，我们研究隐式正则化，并在X分解方面进行梯度下降。我们猜想并提供了经验和理论上的证据，这些证据足够小，初始化足够小，足够接近起点，梯度下降，在梯度下降。完整的维分解会收敛到最小核范围解决方案。
We study implicit regularization when optimizing an underdetermined quadratic objective over a matrix $X$ with gradient descent on a factorization of X. We conjecture and provide empirical and theoretical evidence that with small enough step sizes and initialization close enough to the origin, gradient descent on a full dimensional factorization converges to the minimum nuclear norm solution.