Pith. sign in

REVIEW

A Critical Evaluation of Open-World Machine Learning

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2007.04391 v1 pith:HH3SM3L3 submitted 2020-07-08 cs.LG cs.CR

A Critical Evaluation of Open-World Machine Learning

classification cs.LG cs.CR
keywords open-worlddatadetectionsystemsadversarialdetectorsevaluationfalse
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Open-world machine learning (ML) combines closed-world models trained on in-distribution data with out-of-distribution (OOD) detectors, which aim to detect and reject OOD inputs. Previous works on open-world ML systems usually fail to test their reliability under diverse, and possibly adversarial conditions. Therefore, in this paper, we seek to understand how resilient are state-of-the-art open-world ML systems to changes in system components? With our evaluation across 6 OOD detectors, we find that the choice of in-distribution data, model architecture and OOD data have a strong impact on OOD detection performance, inducing false positive rates in excess of $70\%$. We further show that OOD inputs with 22 unintentional corruptions or adversarial perturbations render open-world ML systems unusable with false positive rates of up to $100\%$. To increase the resilience of open-world ML, we combine robust classifiers with OOD detection techniques and uncover a new trade-off between OOD detection and robustness.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.