How to perform automated data augmentation
Juan David Tuta Botero
Data Science | Machine Learning | Artificial Intelligence
Every day of our lives we see regularly how machine learning takes more and more prominence in the different economic, social, and even cultural sectors. A derivation of this means that our models not only become more robust and sophisticated but also require more and better training data. The fact that collecting relevant data has generated a vast market from which many of the most important companies of the last century have leveraged does not mean that this has become an easy task, in many cases the amount of information available is not enough and in this article, we are going to discuss some of how we can solve this problem.
Why Data Augmentation?
Modern machine learning models, such as deep neural networks, may have billions of parameters and require massive labeled training datasets which are often not available. The technique of artificially expanding labeled training datasets known as data augmentation has quickly become critical for combating this data scarcity problem. Today, data augmentation is used as a secret sauce in nearly every state-of-the-art model for image classification and is becoming increasingly common in other modalities such as natural language understanding as well.
Practical Methods of Learnable Data Augmentations
Learnable data augmentation is promising, in that it allows us to search for more powerful parameterizations and compositions of transformations. Perhaps the biggest difficulty with automating data augmentation is how to search over the space of transformations. This can be prohibitive due to a large number of transformation functions and associated parameters in the search space. How can we design learnable algorithms that explore the space of transformation functions efficiently and effectively, and find augmentation strategies that can outperform human-designed heuristics? In response to the challenge, we highlight a few recent methods below.
Transformation Adversarial Networks for Data Augmentations (TANDA)
To address this problem, TANDA proposes a framework to learn augmentations, which models data augmentations as sequences of Transformation Functions (TFs) provided by users. For example, these might include?“rotate 5 degrees”?or?“shift by 2 pixels”. At the core, this framework consists of two components first?learning a TF sequence generator?that results in useful augmented data points, and second?using the sequence generator?to augment training sets for a downstream model. In particular, the TF sequence generator is trained to produce realistic images by having to fool a discriminator network, following the GANs framework. The underlying assumption here is that the transformations would either lead to realistic images or indistinguishable garbage images that are off the manifold. As shown in Figure 1, the objective for the generator is to produce sequences of TFs such that the augmented data point can fool the discriminator; whereas the objective for the discriminator is to produce values close to 1 for data points in the original training set and values close to 0 for augmented data points.
领英推荐
Data Augmentations for Model Patching
Most machine learning research carried out today is still solving fixed tasks. However, in the real world, machine learning models in deployment can fail due to unanticipated changes in data distribution. This raises the concerning question of how we can move from model building to model maintenance in an adaptive manner. In our latest work, we propose model patching the first framework that exploits data augmentation to mitigate the performance issues of a flawed model in deployment.
Class-conditional Learned Augmentations for Model Patching (CLAMP)
The conceptual framework of model patching consists of two stages.
We propose CLAMP, an instantiation of our first end-to-end model patching framework. They combine a novel consistency regularizer with a robust training objective that is inspired by recent work of Group Distributionally Robust Optimization (GDRO). They extend GDRO to a class conditional training objective that jointly optimizes for the worst-subgroup performance in each class. CLAMP is able to balance the performance of subgroups within each class, reducing the performance gap by up to?24x. On a skin cancer detection dataset ISIC, CLAMP improves robust accuracy by?11.7% compared to the robust training baseline. Through visualization, we also show in Figure 5 that CLAMP successfully removes the model’s reliance on the spurious feature (colorful bandages), shifting its attention to the skin lesion's true feature of interest.
Bibliography