Publication
Loss landscape: SGD has a better view. (2020).
CBMM-Memo-107.pdf (1.03 MB)
Typos and small edits, ver11 (955.08 KB)
Small edits, corrected Hessian for spurious case (337.19 KB)
Compositional sparsity of learnable functions. Bulletin of the American Mathematical Society 61, 438-456 (2024).
Visual Cortex and Deep Networks: Learning Invariant Representations. 136 (The MIT Press, 2016). at <https://mitpress.mit.edu/books/visual-cortex-and-deep-networks>
Double descent in the condition number. (2019).
Fixing typos, clarifying error in y, best approach is crossvalidation (837.18 KB)
Incorporated footnote in text plus other edits (854.05 KB)
Deleted previous discussion on kernel regression and deep nets: it will appear, extended, in a separate paper (795.28 KB)
correcting a bad typo (261.24 KB)
Deleted plot of condition number of kernel matrix: we cannot get a double descent curve (769.32 KB)
From Marr’s Vision to the Problem of Human Intelligence. (2021).
CBMM-Memo-118.pdf (362.19 KB)
I-theory on depth vs width: hierarchical function composition. (2015).
cbmm_memo_041.pdf (1.18 MB)
Theory II: Landscape of the Empirical Risk in Deep Learning. (2017).
CBMM Memo 066_1703.09833v2.pdf (5.56 MB)
Theoretical issues in deep networks. Proceedings of the National Academy of Sciences 201907369 (2020). doi:10.1073/pnas.1907369117
PNASlast.pdf (915.3 KB)
Compositional Sparsity of Learnable Functions. (2024).
This is an update of the AMS paper (230.72 KB)
]