Size and Depth Separation in Approximating Benign Functions with Neural Networks

Gal Vardi, Daniel Reichman, Toniann Pitassi, Ohad Shamir
Proceedings of Thirty Fourth Conference on Learning Theory, PMLR 134:4195-4223, 2021.

Abstract

When studying the expressive power of neural networks, a main challenge is to understand how the size and depth of the network affect its ability to approximate real functions. However, not all functions are interesting from a practical viewpoint: functions of interest usually have a polynomially-bounded Lipschitz constant, and can be computed efficiently. We call functions that satisfy these conditions “benign”, and explore the benefits of size and depth for approximation of benign functions with ReLU networks. As we show, this problem is more challenging than the corresponding problem for non-benign functions. We give complexity-theoretic barriers to showing depth-lower-bounds: Proving existence of a benign function that cannot be approximated by polynomial-sized networks of depth $4$ would settle longstanding open problems in computational complexity. It implies that beyond depth $4$ there is a barrier to showing depth-separation for benign functions, even between networks of constant depth and networks of nonconstant depth. We also study size-separation, namely, whether there are benign functions that can be approximated with networks of size $O(s(d))$, but not with networks of size $O(s’(d))$. We show a complexity-theoretic barrier to proving such results beyond size $O(d\log^2(d))$, but also show an explicit benign function, that can be approximated with networks of size $O(d)$ and not with networks of size $o(d/\log d)$. For approximation in the $L_\infty$ sense we achieve such separation already between size $O(d)$ and size $o(d)$. Moreover, we show superpolynomial size lower bounds and barriers to such lower bounds, depending on the assumptions on the function. Our size-separation results rely on an analysis of size lower bounds for Boolean functions, which is of independent interest: We show linear size lower bounds for computing explicit Boolean functions (such as set disjointness) with neural networks and threshold circuits.

Cite this Paper

BibTeX
@InProceedings{pmlr-v134-vardi21a, title = {Size and Depth Separation in Approximating Benign Functions with Neural Networks}, author = {Vardi, Gal and Reichman, Daniel and Pitassi, Toniann and Shamir, Ohad}, booktitle = {Proceedings of Thirty Fourth Conference on Learning Theory}, pages = {4195--4223}, year = {2021}, editor = {Belkin, Mikhail and Kpotufe, Samory}, volume = {134}, series = {Proceedings of Machine Learning Research}, month = {15--19 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v134/vardi21a/vardi21a.pdf}, url = {https://proceedings.mlr.press/v134/vardi21a.html}, abstract = {When studying the expressive power of neural networks, a main challenge is to understand how the size and depth of the network affect its ability to approximate real functions. However, not all functions are interesting from a practical viewpoint: functions of interest usually have a polynomially-bounded Lipschitz constant, and can be computed efficiently. We call functions that satisfy these conditions “benign”, and explore the benefits of size and depth for approximation of benign functions with ReLU networks. As we show, this problem is more challenging than the corresponding problem for non-benign functions. We give complexity-theoretic barriers to showing depth-lower-bounds: Proving existence of a benign function that cannot be approximated by polynomial-sized networks of depth $4$ would settle longstanding open problems in computational complexity. It implies that beyond depth $4$ there is a barrier to showing depth-separation for benign functions, even between networks of constant depth and networks of nonconstant depth. We also study size-separation, namely, whether there are benign functions that can be approximated with networks of size $O(s(d))$, but not with networks of size $O(s’(d))$. We show a complexity-theoretic barrier to proving such results beyond size $O(d\log^2(d))$, but also show an explicit benign function, that can be approximated with networks of size $O(d)$ and not with networks of size $o(d/\log d)$. For approximation in the $L_\infty$ sense we achieve such separation already between size $O(d)$ and size $o(d)$. Moreover, we show superpolynomial size lower bounds and barriers to such lower bounds, depending on the assumptions on the function. Our size-separation results rely on an analysis of size lower bounds for Boolean functions, which is of independent interest: We show linear size lower bounds for computing explicit Boolean functions (such as set disjointness) with neural networks and threshold circuits.} }
Endnote
%0 Conference Paper %T Size and Depth Separation in Approximating Benign Functions with Neural Networks %A Gal Vardi %A Daniel Reichman %A Toniann Pitassi %A Ohad Shamir %B Proceedings of Thirty Fourth Conference on Learning Theory %C Proceedings of Machine Learning Research %D 2021 %E Mikhail Belkin %E Samory Kpotufe %F pmlr-v134-vardi21a %I PMLR %P 4195--4223 %U https://proceedings.mlr.press/v134/vardi21a.html %V 134 %X When studying the expressive power of neural networks, a main challenge is to understand how the size and depth of the network affect its ability to approximate real functions. However, not all functions are interesting from a practical viewpoint: functions of interest usually have a polynomially-bounded Lipschitz constant, and can be computed efficiently. We call functions that satisfy these conditions “benign”, and explore the benefits of size and depth for approximation of benign functions with ReLU networks. As we show, this problem is more challenging than the corresponding problem for non-benign functions. We give complexity-theoretic barriers to showing depth-lower-bounds: Proving existence of a benign function that cannot be approximated by polynomial-sized networks of depth $4$ would settle longstanding open problems in computational complexity. It implies that beyond depth $4$ there is a barrier to showing depth-separation for benign functions, even between networks of constant depth and networks of nonconstant depth. We also study size-separation, namely, whether there are benign functions that can be approximated with networks of size $O(s(d))$, but not with networks of size $O(s’(d))$. We show a complexity-theoretic barrier to proving such results beyond size $O(d\log^2(d))$, but also show an explicit benign function, that can be approximated with networks of size $O(d)$ and not with networks of size $o(d/\log d)$. For approximation in the $L_\infty$ sense we achieve such separation already between size $O(d)$ and size $o(d)$. Moreover, we show superpolynomial size lower bounds and barriers to such lower bounds, depending on the assumptions on the function. Our size-separation results rely on an analysis of size lower bounds for Boolean functions, which is of independent interest: We show linear size lower bounds for computing explicit Boolean functions (such as set disjointness) with neural networks and threshold circuits.
APA
Vardi, G., Reichman, D., Pitassi, T. & Shamir, O.. (2021). Size and Depth Separation in Approximating Benign Functions with Neural Networks. Proceedings of Thirty Fourth Conference on Learning Theory, in Proceedings of Machine Learning Research 134:4195-4223 Available from https://proceedings.mlr.press/v134/vardi21a.html.

Related Material