Please enable JavaScript.
Coggle requires JavaScript to display documents.
Automatic Patch generation by Learning Correct Code (Prophet (Patch…
Automatic Patch generation by Learning Correct Code
Prophet
Probabilistic Model
assigns a probability to each candidate patch
Maximum Likelihood Estimation
Patch Generation
Defect Localization
dynamic analysis
Search Space Generation
SPR search space
transformation schemas
staged program repair
condition synthesis
Universal Feature Extraction
extracts features
program value features
capture interactions between an occurrence of a variable
modification features
captures the kind of modification
captures relationships between the kinds of statements
Feature Extraction for C
extracted features into a binary feature vector
Rank Candidate Patches
computes a feature vector
ranks the generated patches
Validate Candidate Patches
test suite to attempt to validate the candidate patches
condition synthesis algorithm
Offline Learning
training set of successful human patches
Extract Features
analyzes a structural diff on the abstract syntax trees
Learn Model Parameters
parameterized log-linear probabilistic model
Design
Probabilistic Model
geometric distribution for the defect localization information
Defect Localization Approximation for Learning
expands the range of applications
improves the running time of the training phase
Learning Algorithm
AST Structural Difference
Initialization
Update Current θ
Measure Performance
Update Best θ and Termination
Feature Extraction
Repair Algorithm
Generate Search Space
Rank Candidate Patch
Validate Candidate Patch