Assumption, how model will interpret and generalize data before learning

It is an additional assumption for contingencies to resolve errors caused by incorrect assumptions of the learning algorithm for data that has not been given. In other words, it is a bias set as a tendency for certain inference paths to be prioritized or more easily learned due to the characteristics of the data or the design of the model.

Each design choice imposes an inductive bias on the learning process. Informally, inductive biases impose a preference over 𝓕. When the inductive bias prefers a subset of 𝓕 which contains a function 𝑓 that obtains minimal generalization error, the inductive bias is useful and improves learning.

Learning Bias

model intrinsic but encodes and amplifies biases

의도되지 않은 편향으로, 학습 데이터에 있는 불필요하거나 부정적인 편향이 학습 과정에서 증폭되어 모델에 반영되는 것

Inductive Bias란 무엇일까?

[딥러닝][개념정리] Inductive Bias란?