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

<p>Part2 of the Replication Package for the paper "GIST: Generated Inputs Sets Transferability in Deep Learning"</p> <p>Contains RoBERTa models and data for the KMNC property.</p> <p>Github link: https://github.com/FlowSs/GIST</p> <p>Part1 can be found here: https://zenodo.org/records/10028594</p> <p>Abstract:</p> <p>&nbsp; &nbsp;To foster the verifiability and testability of Deep Neural Networks (DNN), an increasing number of methods<br>for test case generation techniques are being developed.<br>&nbsp; &nbsp; When confronted with testing DNN models, the user can apply any existing test generation technique.<br>However, it needs to do so for each technique and each DNN model under test, which can be expensive.<br>Therefore, a paradigm shift could benefit this testing process: rather than regenerating the test set independently<br>for each DNN model under test, we could transfer from existing DNN models.<br>&nbsp; &nbsp; &nbsp;This paper introduces GIST (Generated Inputs Sets Transferability), a novel approach for the efficient<br>transfer of test sets. Given a property selected by a user (e.g., neurons covered, faults), GIST enables the<br>selection of good test sets from the point of view of this property among available test sets. This allows the<br>user to recover similar properties on the transferred test sets as he would have obtained by generating the<br>test set from scratch with a test cases generation technique. Experimental results show that GIST can select<br>effective test sets for the given property to transfer. Moreover, GIST scales better than reapplying test case<br>generation techniques from scratch on DNN models under test.</p>