Unraveling the ARC Puzzle: Mimicking Human Solutions with Object-Centric Decision Transformer

id:: 2306.08204
Authors:: Jaehyun Park, Jaegyun Im, Sanha Hwang, Mintaek Lim, Sabina Ualibekova, Sejin Kim, Sundong Kim
Published:: 2023-06-14
arXiv:: https://arxiv.org/abs/2306.08204
PDF:: https://arxiv.org/pdf/2306.08204
DOI:: N/A
Journal Reference:: N/A
Primary Category:: cs.AI
Categories:: cs.AI, cs.LG
Comment:: N/A
github_url:: _

abstract

In the pursuit of artificial general intelligence (AGI), we tackle Abstraction and Reasoning Corpus (ARC) tasks using a novel two-pronged approach. We employ the Decision Transformer in an imitation learning paradigm to model human problem-solving, and introduce an object detection algorithm, the Push and Pull clustering method. This dual strategy enhances AI’s ARC problem-solving skills and provides insights for AGI progression. Yet, our work reveals the need for advanced data collection tools, robust training datasets, and refined model structures. This study highlights potential improvements for Decision Transformers and propels future AGI research.

premise

outline

quotes

notes

summary

This paper presents a novel approach to solving Abstraction and Reasoning Corpus (ARC) puzzles by mimicking human problem-solving strategies. The researchers combine imitation learning with an object-centric decision transformer and a Push and Pull clustering algorithm to improve the accuracy of AI models in solving these complex reasoning tasks. While showing promising results on four specific ARC problem types, the study highlights the need for further advancements in data collection and model architecture to achieve human-level performance on a broader range of ARC tasks.

Brief Overview

The research tackles the challenge of imbuing AI with human-like abstraction and reasoning skills, focusing on the challenging ARC dataset. Existing solutions struggle to surpass ~30% accuracy. The proposed approach uses a Decision Transformer in an imitation learning framework, learning from human problem-solving traces captured using an Object-Oriented ARC (O2ARC) interface. Additionally, a novel object detection algorithm, Push and Pull (PnP) clustering, is introduced to enhance object understanding within the puzzle. The combined approach, termed “Object-centric Decision Transformer,” shows improved results on selected ARC problem types.

Key Points

The study employs an imitation learning paradigm using a Decision Transformer to model human problem-solving strategies on the ARC dataset.
A novel object detection algorithm, the Push and Pull (PnP) clustering method, is introduced to enhance the model’s understanding of objects within the ARC puzzles.
The combined approach, “Object-centric Decision Transformer,” yields improved results on four representative ARC problem types (diagonal flip, tetris, gravity, and stretch).
Data augmentation techniques were implemented to increase the size of the training dataset.
The study reveals a need for improved data collection methods, robust training datasets, and more sophisticated model architectures to further enhance AI’s performance on ARC tasks and progress towards artificial general intelligence (AGI).
The O2ARC tool was improved to collect more comprehensive and higher quality human traces.

Notable Quotes

None explicitly identified from the provided text.

Primary Themes

Imitation Learning: The core methodology involves learning from human problem-solving traces, a key aspect of imitation learning.
Object-Centric Approach: The integration of the PnP clustering algorithm introduces an object-centric perspective to reasoning, mirroring human cognition.
Decision Transformer Application: The Decision Transformer serves as the core model for imitation learning, leveraging its sequential modeling capabilities.
Data Augmentation and its Limitations: The challenges and limitations of data augmentation in the context of the highly specific and varied nature of ARC puzzles are discussed.
AGI Progression: The broader goal and context of the research is to contribute to the field of AGI and enhance AI’s reasoning abilities.