Compositional Task

Compositional tasks challenge artificial intelligence systems to solve problems by combining simpler sub-tasks, mirroring human cognitive abilities. Current research focuses on improving large language models (LLMs) and reinforcement learning (RL) agents' performance on these tasks, exploring techniques like recursive tuning, knowledge distillation, and specialized prompting strategies (e.g., skills-in-context prompting) to enhance their compositional reasoning capabilities. These efforts aim to overcome limitations in current architectures, such as Transformers, which struggle with efficient function composition and generalization to unseen combinations of sub-tasks. Success in this area would significantly advance AI's ability to handle complex, real-world problems requiring multi-step reasoning and flexible adaptation.

Papers

April 25, 2023

Exploring Compositional Visual Generation with Latent Classifier Guidance
Changhao Shi, Haomiao Ni, Kai Li, Shaobo Han, Mingfu Liang, Martin Renqiang Min
Latent Diffusion Model Pre Trained Generative Model Compositional Task Compositional Image Generation Classifier Guidance

December 17, 2022

Improving Cross-task Generalization of Unified Table-to-text Models with Compositional Task Configurations
Jifan Chen, Yuhao Zhang, Lan Liu, Rui Dong, Xinchi Chen, Patrick Ng, William Yang Wang, Zhiheng Huang
Multi Task Learning Cross Task Generalization Compositional Task Table to Text Encode Task

November 18, 2022

Visual Programming: Compositional visual reasoning without training
Tanmay Gupta, Aniruddha Kembhavi
Visual Question Answering Image Editing Visual Programming Neuro Symbolic Approach Compositional Task Task Specific Training Compositional Visual Reasoning

November 3, 2022

FactorMatte: Redefining Video Matting for Re-Composition Tasks
Zeqi Gu, Wenqi Xian, Noah Snavely, Abe Davis
Image Matting Counterfactual Image Compositional Task Background Subtraction

October 23, 2022

Learning to Perform Complex Tasks through Compositional Fine-Tuning of Language Models
Victor S. Bursztyn, David Demeter, Doug Downey, Larry Birnbaum
Language Model LeArning Abstract Fine Tuning Explanatory Inference Complex Task Compositional Task

October 1, 2022

Zero-Shot Policy Transfer with Disentangled Task Representation of Meta-Reinforcement Learning
Zheng Wu, Yichen Xie, Wenzhao Lian, Changhao Wang, Yanjiang Guo, Jianyu Chen, Stefan Schaal, Masayoshi Tomizuka
Reinforcement Learning Zero Shot Meta Reinforcement Learning Compositional Task Policy Generalization

March 28, 2022

Learning Parameterized Task Structure for Generalization to Unseen Entities
Anthony Z. Liu, Sungryull Sohn, Mahdi Qazwini, Honglak Lee
Strong Generalization Unseen Task Unseen Entity Compositional Task Task Graph

January 21, 2022

Environment Generation for Zero-Shot Compositional Reinforcement Learning
Izzeddin Gur, Natasha Jaques, Yingjie Miao, Jongwook Choi, Manoj Tiwari, Honglak Lee, Aleksandra Faust
Deep Reinforcement Learning Compositional Zero Shot Learning Compositional Task Compositional Approach MiniGrid Environment Environment Generation