Abstract Class Incremental Learning (CIL) aims at learning a classifier in a phase-by-phase
manner, in which only data of a subset of the classes are provided at each phase. Previous …

Abstract Class Incremental Learning (CIL) aims at learning a multi-class classifier in a phase-
by-phase manner, in which only data of a subset of the classes are provided at each phase …

Class Incremental Learning (CIL) aims at learning a classifier in a phase-by-phase manner,
in which only data of a subset of the classes are provided at each phase. Previous works …

C∑ C c= 1∑ di= 1 log λ (c) i= 1 a (I (X; Z| Y)− b) as derived in Eqn.(5). Note that here Y-Axis
is negative because it is conditional mutual information minus a very large positive constant …

Abstract Class Incremental Learning (CIL) aims at learning a classifier in a phase-by-phase
manner, in which only data of a subset of the classes are provided at each phase. Previous …

Class Incremental Learning (CIL) aims at learning a multi-class classifier in a phase-by-
phase manner, in which only data of a subset of the classes are provided at each phase …

Class Incremental Learning (CIL) aims at learning a classifier in a phase-by-phase manner,
in which only data of a subset of the classes are provided at each phase. Previous works …

Next, we observe class-wise representations produced by an naively trained initial phase
model and the oracle model. We find that representations output by the naively trained …