DRIFT: Deep Reinforcement Learning for Functional Software Testing
Authors:
Luke Harries,
Rebekah Storan Clarke,
Timothy Chapman,
Swamy V. P. L. N. Nallamalli,
Levent Ozgur,
Shuktika Jain,
Alex Leung,
Steve Lim,
Aaron Dietrich,
José Miguel Hernández-Lobato,
Tom Ellis,
Cheng Zhang,
Kamil Ciosek
Abstract:
Efficient software testing is essential for productive software development and reliable user experiences. As human testing is inefficient and expensive, automated software testing is needed. In this work, we propose a Reinforcement Learning (RL) framework for functional software testing named DRIFT. DRIFT operates on the symbolic representation of the user interface. It uses Q-learning through Ba…
▽ More
Efficient software testing is essential for productive software development and reliable user experiences. As human testing is inefficient and expensive, automated software testing is needed. In this work, we propose a Reinforcement Learning (RL) framework for functional software testing named DRIFT. DRIFT operates on the symbolic representation of the user interface. It uses Q-learning through Batch-RL and models the state-action value function with a Graph Neural Network. We apply DRIFT to testing the Windows 10 operating system and show that DRIFT can robustly trigger the desired software functionality in a fully automated manner. Our experiments test the ability to perform single and combined tasks across different applications, demonstrating that our framework can efficiently test software with a large range of testing objectives.
△ Less
Submitted 16 July, 2020;
originally announced July 2020.
Interlacing Personal and Reference Genomes for Machine Learning Disease-Variant Detection
Authors:
Luke R Harries,
Suyi Zhang,
Geoffroy Dubourg-Felonneau,
James H R Farmery,
Jonathan Sinai,
Belle Taylor,
Nirmesh Patel,
John W Cassidy,
John Shawe-Taylor,
Harry W Clifford
Abstract:
DNA sequencing to identify genetic variants is becoming increasingly valuable in clinical settings. Assessment of variants in such sequencing data is commonly implemented through Bayesian heuristic algorithms. Machine learning has shown great promise in improving on these variant calls, but the input for these is still a standardized "pile-up" image, which is not always best suited. In this paper,…
▽ More
DNA sequencing to identify genetic variants is becoming increasingly valuable in clinical settings. Assessment of variants in such sequencing data is commonly implemented through Bayesian heuristic algorithms. Machine learning has shown great promise in improving on these variant calls, but the input for these is still a standardized "pile-up" image, which is not always best suited. In this paper, we present a novel method for generating images from DNA sequencing data, which interlaces the human reference genome with personalized sequencing output, to maximize usage of sequencing reads and improve machine learning algorithm performance. We demonstrate the success of this in improving standard germline variant calling. We also furthered this approach to include somatic variant calling across tumor/normal data with Siamese networks. These approaches can be used in machine learning applications on sequencing data with the hope of improving clinical outcomes, and are freely available for noncommercial use at www.ccg.ai.
△ Less
Submitted 26 November, 2018;
originally announced November 2018.
