[link]
In the literature of adversarial examples, there’s this (to me) constant question: is it the case that adversarial examples are causing the model to objectively make a mistake, or just displaying behavior that is deeply weird, and unintuitive relative to our sense of what these models “should” be doing. A lot of the former question seems to come down to arguing over about what’s technically “out of distribution”, which has an occasional angels-dancing-on-a-pin quality, but it’s pretty unambiguously clear that the behavior displayed in this paper is weird, and beyond what I naively expected a network to be able to be manipulated to do. The goal these authors set for themselves is what they call “reprogramming” of a network; they want the ability to essentially hijack the network’s computational engine to perform a different task, predicting a different set of labels, on a different set of inputs than the ones the model was trained on. For example, one task they perform is feeding in MNIST images at the center of a bunch of (what appear to be random, but are actually carefully optimized) pixels, and getting a network that can predict MNIST labels out the other end. Obviously, it’s not literally possible to change the number of outputs that a network produces once it’s trained, so the authors would arbitrarily map ImageNet outputs to MNIST categories (like, “when this model predicts Husky, that actually means the digit 7”) and then judge how well this mapped output performs as a MNIST classifier. I enjoyed the authors’ wry commentary here about the arbitrariness of the mapping, remarking that “a ‘White Shark’ has nothing to do with counting 3 squares in an image, and an ‘Ostrich’ does not at all resemble 10 squares”. https://i.imgur.com/K02cwK0.png This paper assumes a white box attack model, which implies visibility of all of the parameters, and ability to directly calculate gradients through the model. So, given this setup of a input surrounded by modifiable pixel weights, and a desire to assign your “MNIST Labels” correctly, this becomes a straightforward optimization problem: modify the values of your input weights so as to maximize your MNIST accuracy. An important point to note here is that the same input mask of pixel values is applied for every new-task image, and so these values are optimized over a full training set of inserted images, the way that normal weights would be. One interesting observation the authors make is that, counter to the typical setup of adversarial examples, this attack would not work with a fully linear model, since you actually need your “weights” to interact with your “input”, which is different each time, but these are both just different areas of your true input. This need to have different regions of input determine how other areas of input are processed isn’t possible in a linear model where each input has a distinct impact on the output, regardless of other input values. By contrast, when you just need to optimize a single perturbation to get the network to jack up the prediction for one class, that can be accomplished by just applying a strong enough bias everywhere in the input, all pointing in the same direction, which can be added together linearly and still get the job done. The authors are able to perform MNIST and the task of “count the squares in this small input” to relatively high levels of accuracy. They perform reasonably on CIFAR (as well as a fully connected network, but not as well as a convnet). They found that performance was higher when using a pre-trained ImageNet, relative to just random weights. There’s some suggestion made that this implies there’s a kind of transfer learning going on, but honestly, this is weird enough that it’s hard to say. https://i.imgur.com/bj2MUnk.png They were able to get this reprogramming work on different model structures, but, fascinatingly, saw distinctive patterns to the "weight pixels" they needed to add to each model structure, with ResNet easily differentiable from Inception. One minor quibble I have with the framing of this paper - which I overall found impressive, creative, and well-written - is that I feel like it’s stretching the original frame of “adversarial example” a bit too far, to the point of possible provoking confusion. It’s not obvious that the network is making a mistake, per se, when it classifies this very out-of-distribution input as something silly. I suppose, in an ideal world, we may want our models to return to something like a uniform-over-outputs state of low confidence when predicting out of distribution, but that’s a bit different than seeing a gibbon in a picture of a panda. I don’t dispute the authors claim that the behavior they’re demonstrating is a vulnerability in terms of its ability to let outside actors “hijack” networks compute, but I worry we might be overloading the “adversarial example” to cover too many types of network failure modes.
Your comment:
|
[link]
Elsayed et al. use universal adversarial examples to reprogram neural networks in order to perform different tasks. In particular, e.g., on ImageNet, an adversarial example $\delta = \tanh(W \cdot M)$ is computed where $M$ is a mask image (see Figure 1, in the paper the mask image essentially embeds a smaller image into an ImageNet-sized image) and $W$ is the adversarial perturbation itself (note that the notaiton was changed slightly for simplification). The hyperbolic tangent constraints the adversarial example, also called adversarial program, to the valid range of $[-1,1]$ as used in most ImageNet networks. The adversarial program is comuted by minimizing $\min_W (-\log P(h(\tilde{y}) | \tilde{x}) + \lambda \|W\|_2^2$. Here, $h$ is a function mapping the labels of the target taks (e.g., MNIST classification) to the $1000$ labels of the ImageNet classification task (e.g., using the first ten labels, or assigning multiple ImageNet labels to one MNIST label). Essentially, this means minimizing the cross entropy loss of the new task (with new labels) to solve for the adversarial program. Examples of adversarial programs for different tasks and architectures are shown in Figure 1. https://i.imgur.com/hPLQn9m.png Figure 1: Adversarial programs for different ImageNet architectures (columns) and tasks (rows). Counting refers to a simple task of counting white squares (see paper). Interestingly, these adversarial programs are able to achieve quite high accuracy on tasks such as MNIST and CIFAR10. Additionally, the authors found that this is only possible for trained networks, not for networks with random weights (although, this might also have other reasons). Also find this summary at [davidstutz.de](https://davidstutz.de/category/reading/). |