CAN DATA SCIENTISTS TRICK DEEP MACHINE LEARNING ALGORITHMS? Posted on : Aug 18 - 2020

Fine-tuning deep learning models just got easier with the black box adversarial reprogramming (BAR) technique.

When data scientists mention AI and machine learning models, the hot topic of discussion always revolves around not having enough training samples to fine-tune the deep learning models. Consequently, they rely on transfer learning to subsequently fine-tune pre-train deep learning models to increase a model’s accuracy.

To make data scientists work a lot easier, at the International Conference on Machine Learning (ICML) scientists at IBM research and Taiwan’s National Tsing Hua University unrevealed the Black Box Adversarial Reprogramming (BAR) touted as an alternative repurposing technique which turns the weakness of deep neural networks into a strength.

Explaining BAR, the research paper presented at the paper read, “Black Box Adversarial Reprogramming repurposes a well-trained black-box ML model for solving different ML tasks, especially in the scenario with scarce data and constrained resources. The rationale lies in exploiting high-performance but unknown ML models to gain learning capability for transfer learning. Using zeroth order optimization and multi-label mapping techniques, BAR can reprogram a black-box ML model solely based on its input-output responses without knowing the model architecture or changing any parameter. More importantly, in the limited medical data setting, on autism spectrum disorder classification, diabetic retinopathy detection, and melanoma detection tasks, BAR outperforms state-of-the-art methods and yields comparable performance to the vanilla adversarial reprogramming method requiring complete knowledge of the target ML model. BAR also outperforms baseline transfer learning approaches by a significant margin, demonstrating cost-effective means and new insights for transfer learning”.

Understanding Black Box Adversarial Reprogramming (BAR)

Adversarial Reprogramming still falls into the category of the white box deep learning models which do not answer the persistent question that can data scientists expand transfer learning to black-box ML models which have only the input (data samples) and output model responses (predictions) observable.

To bridge this gap raised, black-box adversarial reprogramming (BAR), addresses to reprogram a deployed ML model for black-box transfer learning through black-box setting and taking data scarcity and resource constraint into consideration.

The findings of the research study are presented as follows-

• The researchers have proposed BAR, considered to be a novel approach to reprogram BlackBox ML models for transfer learning. BAR is the first work which expands transfer learning to the black-box setting without finetuning the pre-trained model.

• The performance of BAR is evaluated using three different medical imaging tasks for transfer learning which are taken from ImageNet models which are a set of pre-trained image database or dataset of over 14 million images. These images range from (a) autism spectrum disorder (ASD) classification; (b) diabetic retinopathy (DR) detection; and (c) melanoma detection.

• The results show that the BAR method consistently outperforms state-of-the-art methods and improves the accuracy of the finetuning approach by a significant margin.

• The researches state that in terms of the total expenses, it only costs less than $24 US dollars to reprogram these two APIs for ASD classification. Source