DocTer

Paper

DocTer: Documentation-Guided Fuzzing for Testing Deep Learning API Functions (ISSTA'22)

Tool

We release our tool DocTer on the Docker Hub. Please see the instruction to run DocTer in (./DocTer_instructions.md)

Details

We share the following data:

• The bug list.

• Annotation details (./annotation_details): including the annotated data of 30% of the parameters, the keywords we used for normalization (Section 2.2) and the full list of assumptions we made during annotation (Section 2.2)

• Constraints extracted (./constraints_extracted): The extracted constraints for 2,415 APIs.

• The details of generality analysis among versions (Section 4.1) (./generality_among_versions.md).

• The extended results of the impact of fuzzer’s nondeterminism (Section 4.3) (./fuzzer_nondeterminism.md).

• The break down of all analyzed APIs (Section 3) (./irrelevant_apis_break_down.md).

For further details, please check the README.md in each individual folder.