ProTo: Program-Guided Transformer for Program-Guided Tasks
This addresses the challenge of unifying perception, reasoning, and decision-making in AI agents through program-guided tasks, representing an incremental advance in neural-symbolic approaches.
The paper tackles the problem of learning general program executors for tasks requiring execution of programs on observed specifications, proposing ProTo, which integrates semantic and structural program guidance via cross-attention and masked self-attention, and shows significant outperformance over previous state-of-the-art methods on GQA visual reasoning and 2D Minecraft policy learning datasets, with better generalization to unseen, complex, and human-written programs.
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.