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 ๐. When the inductive bias prefers a subset of ๐ which contains a function ๐ that obtains minimal generalization error, the inductive bias is useful and improves learning.
ย
Learning Bias
model intrinsic but encodes and amplifies biases
์๋๋์ง ์์ ํธํฅ์ผ๋ก, ํ์ต ๋ฐ์ดํฐ์ ์๋ ๋ถํ์ํ๊ฑฐ๋ ๋ถ์ ์ ์ธ ํธํฅ์ด ํ์ต ๊ณผ์ ์์ ์ฆํญ๋์ด ๋ชจ๋ธ์ ๋ฐ์๋๋ ๊ฒ
ย
ย
ย
ย
ย