…papers that tried to re-analyze the NEOWISE observation data in competition with the NEOWISE group. One senior scientist told me outright that he just gave up on using the NEOWISE data set — he couldn’t reproduce the NEOWISE results, and he wasn’t confident enough to challenge the NEOWISE group.
… up with different explanations as to why deep learning optimization is surprisingly easy with SGD. One of the more compelling arguments it that in a high-dimensional space, one is more likely to find a saddle point rather than a local valley. There will always be sufficient dimensions with gradients that point to an escape route.