Generalization Problem

The generalization problem in machine learning focuses on improving a model's ability to perform well on unseen data that differs from its training data. Current research emphasizes understanding and mitigating this challenge across various model architectures, including deep learning models (like LLMs and diffusion models), traditional machine learning algorithms, and gradient-based meta-learning, often focusing on identifying and addressing factors like spurious correlations, distribution shifts, and the impact of training data characteristics. Successfully tackling this problem is crucial for building robust and reliable AI systems applicable to diverse real-world scenarios, ranging from healthcare and robotics to natural language processing and security applications. Addressing generalization limitations is a key step towards creating more trustworthy and effective AI.

Papers

September 18, 2022

Honor of Kings Arena: an Environment for Generalization in Competitive Reinforcement Learning
Hua Wei, Jingxiao Chen, Xiyang Ji, Hongyang Qin, Minwen Deng, Siqin Li, Liang Wang, Weinan Zhang, Yong Yu, Lin Liu, Lanxiao Huang, Deheng Ye, Qiang Fu, Wei Yang
Reinforcement Learning Strong Generalization Environment Feature Generalization Problem Arena Hard Multi Agent Problem

May 31, 2022

Learning brain MRI quality control: a multi-factorial generalization problem
Ghiles Reguig, Marie Chupin, Hugo Dary, Eric Bardinet, Stéphane Lehéricy, Romain Valabregue
Conversational Data Generalization Problem MRI Data

February 9, 2022

Agree to Disagree: Diversity through Disagreement for Better Transferability
Matteo Pagliardini, Martin Jaggi, François Fleuret, Sai Praneeth Karimireddy
Genetic Algorithm Diversity Awareness Task Transferability Learning Algorithm Task Discrepancy Opinion Disagreement Generalization Problem disAgreement Score

December 10, 2021

PACMAN: PAC-style bounds accounting for the Mismatch between Accuracy and Negative log-loss
Matias Vera, Leonardo Rey Vega, Pablo Piantanida
Loss Function Generalization Error Generalization Gap Generalization Problem Model Mismatch Logarithmic Loss Negative Preference

Generalization Problem

Papers

Honor of Kings Arena: an Environment for Generalization in Competitive Reinforcement Learning

Learning brain MRI quality control: a multi-factorial generalization problem

Agree to Disagree: Diversity through Disagreement for Better Transferability

PACMAN: PAC-style bounds accounting for the Mismatch between Accuracy and Negative log-loss