Search for: All records

Knowledge distillation is a simple but powerful way to transfer knowledge between a teacher model to a student model. Existing work suffers from at least one of the following key limitations in terms of direction and scope of transfer which restrict its use: all knowledge is transferred from teacher to student regardless of whether or not that knowledge is useful, the student is the only one learning in this exchange, and typically distillation transfers knowledge only from a single teacher to a single student. We formulate a novel form of knowledge distillation in which many models can act as both students and teachers which we call cooperative distillation. The models cooperate as follows: a model (the student) identifies specific deficiencies in it's performance and searches for another model (the teacher) who encodes learned knowledge into instructional virtual instances via counterfactual instance generation. Because different models may have different strengths and weaknesses, all models can act as either students or teachers (cooperation) when appropriate and only distill knowledge in areas specific to their strengths (focus). Since counterfactuals as a paradigm are not tied to any specific algorithm, we can use this method to distill knowledge between learners of different architectures, algorithms, and even feature spaces. We demonstrate our approach not only outperforms baselines such as transfer learning, self-supervised learning, and multiple knowledge distillation algorithms on several datasets, but it can also be used in settings where the aforementioned techniques cannot. 

Deep learning holds a great promise of revolutionizing healthcare and medicine. Unfortunately, various inference attack models demonstrated that deep learning puts sensitive patient information at risk. The high capacity of deep neural networks is the main reason behind the privacy loss. In particular, patient information in the training data can be unintentionally memorized by a deep network. Adversarial parties can extract that information given the ability to access or query the network. In this paper, we propose a novel privacy-preserving mechanism for training deep neural networks. Our approach adds decaying Gaussian noise to the gradients at every training iteration. This is in contrast to the mainstream approach adopted by Google's TensorFlow Privacy, which employs the same noise scale in each step of the whole training process. Compared to existing methods, our proposed approach provides an explicit closed-form mathematical expression to approximately estimate the privacy loss. It is easy to compute and can be useful when the users would like to decide proper training time, noise scale, and sampling ratio during the planning phase. We provide extensive experimental results using one real-world medical dataset (chest radiographs from the CheXpert dataset) to validate the effectiveness of the proposed approach. The proposed differential privacy based deep learning model achieves significantly higher classification accuracy over the existing methods with the same privacy budget. 

Computer-aided diagnosis (CAD) systems must constantly cope with the perpetual changes in data distribution caused by different sensing technologies, imaging protocols, and patient populations. Adapting these systems to new domains often requires significant amounts of labeled data for re-training. This process is labor-intensive and time-consuming. We propose a memory-augmented capsule network for the rapid adaptation of CAD models to new domains. It consists of a capsule network that is meant to extract feature embeddings from some high-dimensional input, and a memory-augmented task network meant to exploit its stored knowledge from the target domains. Our network is able to efficiently adapt to unseen domains using only a few annotated samples. We evaluate our method using a large-scale public lung nodule dataset (LUNA), coupled with our own collected lung nodules and incidental lung nodules datasets. When trained on the LUNA dataset, our network requires only 30 additional samples from our collected lung nodule and incidental lung nodule datasets to achieve clinically relevant performance (0.925 and 0.891 area under receiving operating characteristic curves (AUROC), respectively). This result is equivalent to using two orders of magnitude less labeled training data while achieving the same performance. We further evaluate our method by introducing heavy noise, artifacts, and adversarial attacks. Under these severe conditions, our network’s AUROC remains above 0.7 while the performance of state-of-the-art approaches reduce to chance level