In recent years, Convolutional Neural Networks (CNNs) have become the state-of-the-art for object recognition in computer vision. Typically, a CNN consists of several convolutional layers, followed by two fully-connected layers. An intuition behind this is that the convolutional layers learn a better representation of the input data, and the fully connected layers then learn to classify this representation based into a set of labels.
However, before CNNs started to dominate, Support Vector Machines (SVMs) were the state-of-the-art. So it seems sensible to say that an SVM is still a stronger classifier than a two-layer fully-connected neural network. Therefore, I am wondering why state-of-the-art CNNs tend to use the fully connected layers for classification rather than an SVM? In this way, you would have the best of both worlds: a strong feature representation, and a strong classifier, rather than a strong feature representation but only a weak classifier…
Any ideas?
Best Answer
What is an SVM, anyway?
I think the answer for most purposes is “the solution to the following optimization problem”: $$ \begin{split} \operatorname*{arg\,min}_{f \in \mathcal H} \frac{1}{n} \sum_{i=1}^n \ell_\mathit{hinge}(f(x_i), y_i) \, + \lambda \lVert f \rVert_{\mathcal H}^2 \\ \ell_\mathit{hinge}(t, y) = \max(0, 1 - t y) ,\end{split} \tag{SVM} $$ where $\mathcal H$ is a reproducing kernel Hilbert space, $y$ is a label in $\{-1, 1\}$, and $t = f(x) \in \mathbb R$ is a “decision value”; our final prediction will be $\operatorname{sign}(t)$. In the simplest case, $\mathcal H$ could be the space of affine functions $f(x) = w \cdot x + b$, and $\lVert f \rVert_{\mathcal H}^2 = \lVert w \rVert^2 + b^2$. (Handling of the offset $b$ varies depending on exactly what you’re doing, but that’s not important for our purposes.) In the ‘90s through the early ‘10s, there was a lot of work on solving this particular optimization problem in various smart ways, and indeed that’s what LIBSVM / LIBLINEAR / SVMlight / ThunderSVM / ... do. But I don’t think that any of these particular algorithms are fundamental to “being an SVM,” really.
Now, how do we train a deep network? Well, we try to solve something like, say, $$ \begin{split} \operatorname*{arg\,min}_{f \in \mathcal F} \frac1n \sum_{i=1}^n \ell_\mathit{CE}(f(x_i), y) + R(f) \\ \ell_\mathit{CE}(p, y) = - y \log(p) - (1-y) \log(1 - p) ,\end{split} \tag{$\star$} $$ where now $\mathcal F$ is the set of deep nets we consider, which output probabilities $p = f(x) \in [0, 1]$. The explicit regularizer $R(f)$ might be an L2 penalty on the weights in the network, or we might just use $R(f) = 0$. Although we could solve (SVM) up to machine precision if we really wanted, we usually can’t do that for $(\star)$ when $\mathcal F$ is more than one layer; instead we use stochastic gradient descent to attempt at an approximate solution.
If we take $\mathcal F$ as a reproducing kernel Hilbert space and $R(f) = \lambda \lVert f \rVert_{\mathcal F}^2$, then $(\star)$ becomes very similar to (SVM), just with cross-entropy loss instead of hinge loss: this is also called kernel logistic regression. My understanding is that the reason SVMs took off in a way kernel logistic regression didn’t is largely due to a slight computational advantage of the former (more amenable to these fancy algorithms), and/or historical accident; there isn’t really a huge difference between the two as a whole, as far as I know. (There is sometimes a big difference between an SVM with a fancy kernel and a plain linear logistic regression, but that’s comparing apples to oranges.)
So, what does a deep network using an SVM to classify look like? Well, that could mean some other things, but I think the most natural interpretation is just using $\ell_\mathit{hinge}$ in $(\star)$.
One minor issue is that $\ell_\mathit{hinge}$ isn’t differentiable at $\hat y = y$; we could instead use $\ell_\mathit{hinge}^2$, if we want. (Doing this in (SVM) is sometimes called “L2-SVM” or similar names.) Or we can just ignore the non-differentiability; the ReLU activation isn’t differentiable at 0 either, and this usually doesn’t matter. This can be justified via subgradients, although note that the correctness here is actually quite subtle when dealing with deep networks.
An ICML workshop paper – Tang, Deep Learning using Linear Support Vector Machines, ICML 2013 workshop Challenges in Representation Learning – found using $\ell_\mathit{hinge}^2$ gave small but consistent improvements over $\ell_\mathit{CE}$ on the problems they considered. I’m sure others have tried (squared) hinge loss since in deep networks, but it certainly hasn’t taken off widely.
(You have to modify both $\ell_\mathit{CE}$ as I’ve written it and $\ell_\mathit{hinge}$ to support multi-class classification, but in the one-vs-rest scheme used by Tang, both are easy to do.)
Another thing that’s sometimes done is to train CNNs in the typical way, but then take the output of a late layer as "features" and train a separate SVM on that. This was common in early days of transfer learning with deep features, but is I think less common now.
Something like this is also done sometimes in other contexts, e.g. in meta-learning by Lee et al., Meta-Learning with Differentiable Convex Optimization, CVPR 2019, who actually solved (SVM) on deep network features and backpropped through the whole thing. (They didn't, but you can even do this with a nonlinear kernel in $\mathcal H$; this is also done in some other "deep kernels" contexts.) It’s a very cool approach – one that I've also worked on – and in certain domains this type of approach makes a ton of sense, but there are some pitfalls, and I don’t think it’s very applicable to a typical "plain classification" problem.