Compositional Approach

The compositional approach in machine learning and related fields focuses on breaking down complex problems into smaller, manageable sub-problems, then combining their solutions to address the overall task. Current research emphasizes developing models and algorithms that can effectively learn and generalize from these component parts, including techniques like component-to-composition learning, retrieval-enhanced meta-learning, and the use of tree-based architectures to represent compositional structures. This approach is significant because it improves model interpretability, robustness to distribution shifts, and the ability to handle unseen combinations of known elements, leading to more generalizable and reliable AI systems across diverse applications.

Papers

July 14, 2022

Verification of Sigmoidal Artificial Neural Networks using iSAT
Dominik Grundt, Sorin Liviu Jurj, Willem Hagemann, Paul Kröger, Martin Fränzle
Verification Task Compositional Approach Sigmoid Function SMT Solver Interval Bound Propagation

February 20, 2022

Understanding Robust Generalization in Learning Regular Languages
Soham Dan, Osbert Bastani, Dan Roth
Recurrent Neural Network Robust Generalization Compositional Approach Regular Language

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

Compositional Approach

Papers

Verification of Sigmoidal Artificial Neural Networks using iSAT

Understanding Robust Generalization in Learning Regular Languages

Environment Generation for Zero-Shot Compositional Reinforcement Learning