Two continual learning methods applied to sequential early-exit training reduce interference between exits, yielding higher accuracy and faster inference at low compute budgets on standard benchmarks.
Θµ} and produces class probability vectorˆ yµ (see Figure 1), with ˆyµ(c)(forc∈ {1
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Balancing Stability and Plasticity in Sequentially Trained Early-Exiting Neural Networks
Two continual learning methods applied to sequential early-exit training reduce interference between exits, yielding higher accuracy and faster inference at low compute budgets on standard benchmarks.