State machine compression for scalable pattern matching

A method for processing data includes encoding a finite automaton, which includes states and transitions between the states that express a plurality of predefined patterns, by grouping the states of the automaton into sets according to a common property shared by the states in each set, and assigning codes to the states according to the grouping. The codes are stored in an electronic memory, along with rules that are associated with the patterns. The automaton is traversed in order to identify one or more of the patterns in an input sequence of data elements by iteratively reading out the codes from the memory responsively to the data elements and to the codes that have been previously read out. Upon identifying a given pattern in the input sequence, an associated action is performed.