stheid/mcts-fuzzer

MCTS-Fuzzer

This project implements a novel mcts-based fuzzing algorithm. The Monte Carlo tree search algorithm uses a grammar model to progressively build words from the startword of the grammar.

The implementation here is just the plain algorithm. To execute this algorithm have a look at the pylibfuzzer repo

mcts-fuzzer's Issues

Memory optimization

This cache variable consumes about 17% of memory. I also tried to convert it into a pair of lists with no measurable success. This hashmap basically expresses the tree structure

MCTS-Fuzzer/src/main/kotlin/isml/aidev/starlib/PCFGSuccGen.kt

Lines 9 to 11 in 78ee822

	class PCFGSuccGen(private val prodRules: ProdRules) : ISuccessorGenerator<SymbolsNode, RuleEdge> {
	private val cache = HashMap<SymbolsNode, List<INewNodeDescription<SymbolsNode, RuleEdge>>>()

This class consumes about 15% +20% for lazy/cached property. I think there is nothing i can strip from it. This class is basically the nodes that make up the tree.

MCTS-Fuzzer/src/main/kotlin/isml/aidev/SymbolsNode.kt

Lines 5 to 10 in 182db30

	class SymbolsNode(
	val currNT: Unique<Symbol.NonTerminal>?,
	val substitutionNTs: List<Unique<Symbol.NonTerminal>> = emptyList(),
	private val parent: SymbolsNode? = null,
	val depth: Int = 0,
	) {