Because in high dimensional space there are partial saddle points, not local min... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		jph00 on May 7, 2017 \| parent \| context \| favorite \| on: Deep Learning’s Impact on Image Processing, Mathem... Because in high dimensional space there are partial saddle points, not local minima. Intuitively obvious, and shown theoretically in the last 12 months.

reader5000 on May 7, 2017 | [–]

Are you referencing https://arxiv.org/abs/1412.0233?

soVeryTired on May 8, 2017 | [–]

Shown theoretically under fairly artificial assumptions.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact