# ProTo: Program-Guided Transformer for Program-Guided Tasks

Source: [ProTo: Program-Guided Transformer for Program-Guided Tasks](<https://www.sota2.com/research/paper/68fa6b131e209131cea2bf8c>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T03:35:51.893Z

Authors: Zelin Zhao, Karan Samel, Binghong Chen, Le Song

Published: 2021-10-02T13:46:32.000Z

## Abstract

Programs, consisting of semantic and structural information, play an important role in the communication between humans and agents. Towards learning general program executors to unify perception, reasoning, and decision making, we formulate program-guided tasks which require learning to execute a given program on the observed task specification. Furthermore, we propose the Program-guided Transformer (ProTo), which integrates both semantic and structural guidance of a program by leveraging cross-attention and masked self-attention to pass messages between the specification and routines in the program. ProTo executes a program in a learned latent space and enjoys stronger representation ability than previous neural-symbolic approaches. We demonstrate that ProTo significantly outperforms the previous state-of-the-art methods on GQA visual reasoning and 2D Minecraft policy learning datasets. Additionally, ProTo demonstrates better generalization to unseen, complex, and human-written programs.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Program-guided task execution on 2D Minecraft Standard (test)](<https://www.sota2.com/research/sota/program-guided-task-execution-on-2d-minecraft-standard-test.md>): Program-guided task execution; 2D Minecraft Standard (test); Task Completion Rate: 99.1.

- [Program-guided task execution on 2D Minecraft Longer Programs](<https://www.sota2.com/research/sota/program-guided-task-execution-on-2d-minecraft-longer-programs.md>): Program-guided task execution; 2D Minecraft Longer Programs; Task Completion Rate: 94.4.

- [Program-guided task execution on 2D Minecraft Complex Programs](<https://www.sota2.com/research/sota/program-guided-task-execution-on-2d-minecraft-complex-programs.md>): Program-guided task execution; 2D Minecraft Complex Programs; Task Completion Rate: 96.3.

- [Visual Question Answering on GQA 2019 (test)](<https://www.sota2.com/research/sota/visual-question-answering-on-gqa-2019-test.md>): Visual Question Answering; GQA 2019 (test); Binary Accuracy: 79.12.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2110.00804>)

[PDF](<https://arxiv.org/pdf/2110.00804.pdf>)