TESSERACT : Eliminating Experimental Bias in Malware Classification across Space and Time
- Submitting institution
-
King's College London
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 126612461
- Type
- E - Conference contribution
- DOI
-
-
- Title of conference / published proceedings
- Proceedings of the 28th USENIX Security Symposium : Proceedings of the 28th USENIX Security Symposium. August 14–16, 2019. Santa Clara, CA, USA
- First page
- 729
- Volume
- -
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- August
- Year of publication
- 2019
- URL
-
https://www.usenix.org/conference/usenixsecurity19/presentation/pendlebury
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
4
- Research group(s)
-
-
- Citation count
- 12
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Machine learning shows tantalizing performance in computer and network security tasks. However, the field requires reasoning on how concept drift affects such performance. This work first highlights the presence of experimental bias in several research papers published in world-leading venues. Then, based on the proposed principled reasoning, the paper defines experimental constraints promoting sound evaluations as well as the adoption of a novel metric to quantify performance decay and costs in addressing concept drift. This work led to invited talks (e.g. USENIX Enigma 2019), research (e.g. https://dl.acm.org/doi/10.1145/3372297.3417291 - CCS 2020), and institutions using our codebase (https://s2lab.kcl.ac.uk/projects/tesseract/).
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -