“The kind of math that would be necessary to understand papers like [1][2][3][4]. [1]: https://arxiv.org/abs/1703.04933 [2]: https://arxiv.org/abs/1701.07875 [3]: https://arxiv.org/abs/1706.01350 [4]: https://arxiv.org/abs/1512.04860”
Sharp Minima Can Generalize For Deep Nets
Despite their overwhelming capacity to overfit, deep learning architectures tend to generalize relatively well to unseen data, allowing them to be deployed in practice. However, explaining why this is the case is still an open area of research. One standing hypothesis that is gaining popularity, e.g. Hochreiter & Schmidhuber (1997); Keskar et al. (2017), is that the flatness of minima of the loss function found by stochastic gradient based methods results in good generalization. This paper argues that most notions of flatness are problematic for deep models and can not be directly applied to explain generalization. Specifically, when focusing on deep networks with rectifier units, we can exploit the particular geometry of parameter space induced by the inherent symmetries that these architectures exhibit to build equivalent models corresponding to arbitrarily sharper minima. Furthermore, if we allow to reparametrize a function, the geometry of its parameters can change drastically without affecting its generalization properties.
候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“Sorry, my last sentence > Otherwise, models can achieve very good generalization. was probably unclear. When we talk about generalization, we typically have to assume a lack of distributional shift. Otherwise, if the "truly unseen data" really is drawn from a different distribution (say one where all cats are now called "dogs" and all dogs are now called…”