Secure Deep Learning Engineering: A Road Towards Quality Assurance of Intelligent Systems

Speaker:  Yang Liu – Singapore, Singapore
Topic(s):  Software Engineering and Programming

Abstract

In company with massive data explosion and powerful computational hardware enhancement, deep learning (DL) has recently achieved substantial strides in cutting edge intelligent applications, ranging from virtual assistant (e.g., Alex, Siri), art design, autonomous vehicles, to medical diagnoses-tasks that until a few years ago could be done only by humans. DL has become the innovation driving force of many next generation’s technologies. We have been witnessing on the increasing trend of industry stakeholders’ continuous investment on DL based intelligent system, penetrating almost every application domain, revolutionizing industry manufacturing as well as reshaping our daily life. However, current DL system development still lacks systematic engineering guidance, quality assurance standards, as well as mature toolchain support. The magic box, such as DL training procedure and logic encoding (as high dimensional weight matrices and complex neural network structures), further poses challenges to interpret and understand behaviors of derived DL systems. The latent software quality and security issues of current DL systems, already started emerging out as the major vendors, rush in pushing products with higher intelligence (e.g., Google/Uber car accident, Alexa and Siri could be manipulated with hidden command.

To bridge the pressing industry demand and future research directions, we perform a large-scale study on the most-recent curated 223 relevant works on deep learning engineering from a software quality assurance perspective. Based on this, we further conduct a consecutive set of ongoing work towards addressing the current challanges in quality, reliability, and security assurance of general-purpose intelligent systems. This talk not only provides a high-level vision of secure deep learning engineering (SDLE), from the quality assurance perspective, accompanied by a state-of-the-art literature curation, as well as the state-of-the-art solutions. We hope this work facilitates drawing the attention of the software and system engineering community on necessity and demands of quality assurance for SDLE, which altogether lays down the foundations and conquers technical barriers towards constructing robust and high-quality DL applications.

About this Lecture

Number of Slides:  60
Duration:  45 minutes
Languages Available:  Chinese (Simplified), English
Last Updated: 

Request this Lecture

To request this particular lecture, please complete this online form.

Request a Tour

To request a tour with this speaker, please complete this online form.

All requests will be sent to ACM headquarters for review.