StreamSplit enables practical streaming contrastive learning for audio on edge devices via distribution-based framework with hybrid loss and uncertainty-guided RL splitter, achieving up to 4.7x latency reduction and 77.1% bandwidth cut with accuracy within 2.2% of server baselines.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.DC 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
StreamSplit: Continuous Audio Representation Learning via Uncertainty-Guided Adaptive Splitting
StreamSplit enables practical streaming contrastive learning for audio on edge devices via distribution-based framework with hybrid loss and uncertainty-guided RL splitter, achieving up to 4.7x latency reduction and 77.1% bandwidth cut with accuracy within 2.2% of server baselines.