Program Synthesis via Test-Time Transduction

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Lee, Kang-il, Koo, Jahyun, Yoon, Seunghyun, Kim, Minbeom, Koh, Hyukhun, Lee, Dongryeol, Jung, Kyomin
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908604093693952
author Lee, Kang-il
Koo, Jahyun
Yoon, Seunghyun
Kim, Minbeom
Koh, Hyukhun
Lee, Dongryeol
Jung, Kyomin
author_facet Lee, Kang-il
Koo, Jahyun
Yoon, Seunghyun
Kim, Minbeom
Koh, Hyukhun
Lee, Dongryeol
Jung, Kyomin
contents We introduce transductive program synthesis, a new formulation of the program synthesis task that explicitly leverages test inputs during synthesis. While prior approaches to program synthesis--whether based on natural language descriptions or input-output examples--typically aim to generalize from training examples, they often struggle with robustness, especially in real-world settings where training examples are limited and test inputs involve various edge cases. To address this, we propose a novel framework that improves robustness by treating synthesis as an active learning over a finite hypothesis class defined by programs' outputs. We use an LLM to predict outputs for selected test inputs and eliminate inconsistent hypotheses, where the inputs are chosen via a greedy maximin algorithm to minimize the number of LLM queries required. We evaluate our approach on four benchmarks: Playgol, MBPP+, 1D-ARC, and programmatic world modeling on MiniGrid. We demonstrate that our method significantly improves program synthesis in both accuracy and efficiency. We release our code at https://github.com/klee972/SYNTRA.
format Preprint
id arxiv_https___arxiv_org_abs_2509_17393
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Program Synthesis via Test-Time Transduction
Lee, Kang-il
Koo, Jahyun
Yoon, Seunghyun
Kim, Minbeom
Koh, Hyukhun
Lee, Dongryeol
Jung, Kyomin
Artificial Intelligence
Computation and Language
We introduce transductive program synthesis, a new formulation of the program synthesis task that explicitly leverages test inputs during synthesis. While prior approaches to program synthesis--whether based on natural language descriptions or input-output examples--typically aim to generalize from training examples, they often struggle with robustness, especially in real-world settings where training examples are limited and test inputs involve various edge cases. To address this, we propose a novel framework that improves robustness by treating synthesis as an active learning over a finite hypothesis class defined by programs' outputs. We use an LLM to predict outputs for selected test inputs and eliminate inconsistent hypotheses, where the inputs are chosen via a greedy maximin algorithm to minimize the number of LLM queries required. We evaluate our approach on four benchmarks: Playgol, MBPP+, 1D-ARC, and programmatic world modeling on MiniGrid. We demonstrate that our method significantly improves program synthesis in both accuracy and efficiency. We release our code at https://github.com/klee972/SYNTRA.
title Program Synthesis via Test-Time Transduction
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2509.17393