Fawzi et al. propose an adaptive data augmentation scheme based on adversarial transformations similar to adversarial training. In particular, in each training iteration – and for each sample/batch – they compute an adversarial version by finding a transformation that maximizes the training loss. The transformation is usually constrained to a specific class of transformations – on MNIST, for example, they consider affine transformations. Additionally, only small transformations are considered, a constraint very similar to the perturbation constraint of adversarial examples. In order to find these adversarial transformations during training, they employ an iterative algorithm based on a local first-order approximation of the training loss on the transformed sample. This leads to a simple linear program that can be solved in each iteration. Overall, however, this results in a significant computational overhead during training (as does adversarial training, as well). Quantitative results on MNIST are encouraging; specifically, Figure 1 shows that the test loss converges faster when employing these adversarial transformations as data augmentation in contrast to random transformations.

https://i.imgur.com/OxuQCP9.png
Figure 1: Test loss for a model trained without data augmentation, the proposed adaptive data augmentation and random data augmentation, see paper for details.

Also find this summary at [davidstutz.de](https://davidstutz.de/category/reading/).

I like the idea proposed in this paper - training on a label-rich domain and transfer the representation to a label-limited domain, but would like to extend it to data more than faces such as transferring the object attributes.

A biofilm is a surface-associated population of microorganisms embedded in a matrix of extracellular polymeric substances. Biofilms are a major natural growth form of microorganisms and the cause of pervasive device-associated infection. This report focuses on the biofilm matrix of Candida albicans, the major fungal pathogen of humans. We report here that the C. albicans zinc-response transcription factor Zap1 is a negative regulator of a major matrix component, soluble β-1,3 glucan, in both in vitro and in vivo biofilm models. To understand the mechanistic relationship between Zap1 and matrix, we identified Zap1 target genes through expression profiling and full genome chromatin immunoprecipitation. On the basis of these results, we designed additional experiments showing that two glucoamylases, Gca1 and Gca2, have positive roles in matrix production and may function through hydrolysis of insoluble β-1,3 glucan chains. We also show that a group of alcohol dehydrogenases Adh5, Csh1, and Ifd6 have roles in matrix production: Adh5 acts positively, and Csh1 and Ifd6, negatively. We propose that these alcohol dehydrogenases generate quorum-sensing aryl and acyl alcohols that in turn govern multiple events in biofilm maturation. Our findings define a novel regulatory circuit and its mechanism of control of a process central to infection.

A biofilm is a surface-associated population of microbes that is embedded in a cement of extracellular compounds. This cement is known as matrix. The two main functions of matrix are to protect cells from their surrounding environment, preventing drugs and other stresses from penetrating the biofilm, and to maintain the architectural stability of the biofilm, acting as a glue to hold the cells together. The presence of matrix is a contributing factor to the high degree of resistance to antimicrobial drugs observed in biofilms. Because biofilms have a major impact on human health, and because matrix is such a pivotal component of biofilms, it is important to understand how the production of matrix is regulated. We have begun to address this question in the major human fungal pathogen Candida albicans. We found that the zinc-responsive regulatory protein Zap1 controls the expression of several genes important for matrix formation in C. albicans. These target genes encode glucoamylases and alcohol dehydrogenases, enzymes that probably govern the synthesis of distinct matrix constituents. The findings here offer insight into the metabolic processes that contribute to biofilm formation and indicate that Zap1 functions broadly as a negative regulator of biofilm maturation.

Similarity-preserving binary code obtained from quantization can efficiently accelerate the retrieval of large-scale image collections. This paper presents iterative quantization (ITQ) that iteratively minimizes quantization error by rotating the data before quantization. ITQ can be couple with any projection of data onto an orthogonal basis. Experiment results show outstanding performance on the retrieval of large-scale image collections especially when the length of binary code is short.
Technical details and results

Idea of ITQ

Figure 1 illustrate the idea of rotating data before quantization to reduce quantization error.

![](http://2.bp.blogspot.com/-riBFQ71aWKk/VRI4Aprwx5I/AAAAAAAAAxw/ggFrB14UGlw/s1600/toy.png)

Figure 1. Toy illustration of the proposed ITQ method
The data points (blue points) are quantized to the closest vertex of the binary cube. By rotating the data points to that shown in Figure 1 (c), the quantization error is reduced and the partitioning respects the structure of the cluster.

ITQ in unsupervised code learning

Given a set of n data points and let each data point be d dimension, the data matrix is denoted by

$X \in \mathbb{R}^{n \times d}$

The binary code matrix can be computed by 

$B = sgn(XW)$ 

$W \in \mathbb{R}^{d \times c}$

where W denotes the projection matrix computed by PCA in this case.

Let R denote any c x c orthogonal matrix. The quantization error after projection and rotation of the data matrix is denotes by

$Q(B,R) = ||B - VR||^{2}_F$

$V = XW$

The ITQ method minimizes the quantization error by seeking optimal R. Because of the quantization operator, the quantization error is not a smooth function and direct minimization of the quantization error is impractical. This paper propose optimizing R and B alternately like k-mean algorithm.  When R is fixed, B is computed by


When B is fixed, R is computed by

$B = sgn(VR)$

where S and "S hat" are the left-singular vector and right-singular vector of the matrix 
         
$B^TV$

Figure 2 shows the quantization error for learning a 32-bit ITQ code on the CIFAR dataset.

![](http://4.bp.blogspot.com/-F6wZGT5fl3E/VRJW5U2VvFI/AAAAAAAAAzg/aEt4DIDpV6E/s1600/quantization%2Berror.png)

Results

"PCA-ITQ" in the legend of the figures denote the proposed method.


![](http://3.bp.blogspot.com/-WbVup__p9CQ/VRJTBh7NKpI/AAAAAAAAAx8/DCUvfo5Ijuw/s1600/result1.png)

![](http://2.bp.blogspot.com/-dj6F3pd5WiM/VRJTBvF6G5I/AAAAAAAAAyA/WyW7He4TJOw/s1600/result2.png)

Demontis et al. study transferability of adversarial examples and data poisening attacks in the light of the targeted models gradients. In particular, they experimentally validate the following hypotheses: First, susceptibility to these attacks depends on the size of the model’s gradients; the higher the gradient, the smaller is the perturbation needed to increase the loss. Second, the size of the gradient depends on regularization. And third, the cosine between the target model’s gradients and the surrogate model’s gradients (trained to compute transferable attacks) influences vulnerability. These insights hold for both evasion and poisening attacks and are motivated by a simple linear Taylor expansion of the target model’s loss.

Also find this summary at [davidstutz.de](https://davidstutz.de/category/reading/).