GIST: Generated Inputs Sets Transferability in Deep Learning (Part 2)

Part2 of the Replication Package for the paper "GIST: Generated Inputs Sets Transferability in Deep Learning"

Contains RoBERTa models and data for the KMNC property.

Github link: https://github.com/FlowSs/GIST

Part1 can be found here: https://zenodo.org/records/10028594

Abstract:

   To foster the verifiability and testability of Deep Neural Networks (DNN), an increasing number of methodsfor test case generation techniques are being developed.    When confronted with testing DNN models, the user can apply any existing test generation technique.However, it needs to do so for each technique and each DNN model under test, which can be expensive.Therefore, a paradigm shift could benefit this testing process: rather than regenerating the test set independentlyfor each DNN model under test, we could transfer from existing DNN models.     This paper introduces GIST (Generated Inputs Sets Transferability), a novel approach for the efficienttransfer of test sets. Given a property selected by a user (e.g., neurons covered, faults), GIST enables theselection of good test sets from the point of view of this property among available test sets. This allows theuser to recover similar properties on the transferred test sets as he would have obtained by generating thetest set from scratch with a test cases generation technique. Experimental results show that GIST can selecteffective test sets for the given property to transfer. Moreover, GIST scales better than reapplying test casegeneration techniques from scratch on DNN models under test.