ICLR 2024 | Class Incremental Learning via Likelihood Ratio Based Task Prediction

An emerging theoretically justified approach to solving CIL is to combine a TIL technique with an out-of-distribution (OOD) detection method, called the TIL+OOD approach(Kim et al., 2022).

It decomposes the CIL probability of a test sample $x$ of the $j$-th class $y_j^{(t)}$ in task $t$ into two probabilities

$$ \mathbf{P}\left(y_j^{(t)} \mid \boldsymbol{x}\right)=\mathbf{P}\left(y_j^{(t)} \mid \boldsymbol{x}, t\right) \mathbf{P}(t \mid \boldsymbol{x}) $$

chufanchen / read-paper-and-code

ICLR 2024 | Class Incremental Learning via Likelihood Ratio Based Task Prediction #85