Generative models have drawn attention in recent years for their ability to synthesize previously unseen instances from complex data distributions. These models can yield surprising results that reflect an ability to combine semantic features from training data, e.g., images of an armchair in the shape of an avocado.

In this talk, we explore the potential of such models to support software testing. To be useful in software testing, such models must produce data that is realistic and diverse relative to the input space of the system under test. Moreover, they must be controllable, so that input generation can target specific regions of the input space to enable focused testing.

We provide an overview of latent-space generative models and how they can support software testing. More specifically, we describe recent work defining test coverage criteria, test input generation methods, and methods that can control the generative process to produce data that is consistent with preconditions thereby allowing oracles encoding postconditions to be used.

Bio: Matthew B. Dwyer is the Robert Thomson Distinguished Professor in the Department of Computer Science at the University of Virginia and an Amazon Scholar. He has authored more than 140 scholarly publications in program analysis, software specification, and automated formal methods. These research contributions have been recognized with five “test of time” (ICSE 2010, SIGSOFT 2010, FSE 2018, SIGSOFT 2021, ISSTA 2022). He was named a Fulbright Research Scholar (2011), an IEEE Fellow (2013), a Parnas Fellow (2018), and an ACM Fellow (2019), and has received the IEEE Computer Society Harlan D. Mills Award (2022).