線形分離可能な場合:Soudry et al., 2018 • 線形分離不可能な場合:Ji and Telgarsky, 2019 • 一般の損失への拡張とさらなる特徴づけ:Ji et al., 2020 ニューラルネットに対しても拡張されつつある(Lyu and Li, 2019; Ji and Telgarsky, 2020; Ji and Telgarsky, 2020) Soudry, D., Hoffer, E., Nacson, M. S., Gunasekar, S., & Srebro, N. (2018). The implicit bias of gradient descent on separable data. The Journal of Machine Learning Research, 19(1), 2822–2878. Ji, Z., & Telgarsky, M. (2019). The implicit bias of gradient descent on nonseparable data. Conference on Learning Theory, 1772–1798. Ji, Z., Dudı́k, M., Schapire, R. E., & Telgarsky, M. (2020). Gradient descent follows the regularization path for general losses. Conference on Learning Theory, 2109–2136. Lyu, K., & Li, J. (2019). Gradient descent maximizes the margin of homogeneous neural networks. arXiv preprint arXiv:1906.05890. Ji, Z., & Telgarsky, M. (2020). Directional convergence and alignment in deep learning. Advances in Neural Information Processing Systems, 33, 17176–17186. Ji, Z., & Telgarsky, M. (2020). Directional convergence and alignment in deep learning. Advances in Neural Information Processing Systems, 33, 17176–17186. 24/91