- Professur für Angewandte Informatik, insbesondere Kognitive Systeme (2)
Schema-Guided Inductive Functional Programming through Automatic Detection of Type Morphisms
- Inductive functional programming systems can be characterised by two diametric approaches: Either they apply exhaustive program enumeration which uses input/output examples (IO) as test cases, or they perform an analytical, data-driven structural generalisation of the IO examples. Enumerative approaches ignore the structural information provided with the IO examples, but use type information to guide and restrict the search. They use higher-order functions which capture recursion schemes during their enumeration, but apply them randomly in a uninformed manner. Analytical approaches on the other side heavily exploit this structural information, but have ignored the benefits of a strong type system so far and use only recursion schemes either fixed and built in, or selected by an expert user. In category theory universal constructs, such as natural transformations or type morphisms, describe recursion schemes which can be defined on any inductively defined data type. They can be characterised by specific universal properties. Those type morphisms and related concepts provide a categorical approach to functional programming, which is often called categorical programming. This work shows how categorical programming can be applied to Inductive Programming and how universal constructs, such as catamorphisms, paramorphisms, and type functors, can be used as recursive program schemes for inductive functional programming. The use of program schemes for Inductive Programming is not new. The special appeal and novelty of this work is that, contrary to previous approaches, the program schemes are neither fixed, nor selected by an expert user: The applicability of those recursion schemes can be automatically detected in the given IO examples of a target function by checking the universal properties of the corresponding type morphisms. Applying this to the analytical system Igor2, both the capabilities and the expressiveness can be extended without a decrease in efficiency. An extension of the analytical functional inductive programming system Igor2 is proposed and its algorithms described. An empirical evaluation demonstrates the improvements with respect to efficiency and effectiveness that can be achieved by the use of type morphisms for Igor2 due to a reduction in search space complexity.
A Combined Analytical and Search-Based Approach to the Inductive Synthesis of Functional Programs
- This thesis is concerned with the inductive synthesis of recursive declarative programs and in particular with the analytical inductive synthesis of functional programs. Program synthesis addresses the problem of (semi-)automatically generating computer programs from specifications. In inductive program synthesis, recursive programs are constructed by generalizing over incomplete specifications such as finite sets of input/output examples (I/O examples). Classical methods for induction of functional programs are analytical, that is, a recursive function definition is derived by detecting and generalizing recurrent patterns between the given I/O examples. Most recent methods, on the other side, are generate-and-test based, that is, they repeatedly generate programs independently from the provided I/O examples until a program is found that correctly computes the examples. Analytical methods are much faster than generate-and-test methods, because they do not rely on search in a program space. Therefore, however, the schemas that generatable programs conform to, must be much more restricted. This thesis at first provides a comprehensive overview of current approaches and methods to inductive program synthesis. Then we present a new algorithm to the inductive synthesis of functional programs that generalizes the analytical approach and combines it with search in a program space. Thereby, the strong restrictions of analytical methods can be resolved for the most part. At the same time, applying analytical techniques allows for pruning large parts of the problem space so that solutions can often be found faster than with generate-and-test methods. By means of several experiments with an implementation of the described algorithm, we demonstrate its capabilities.