FaCTR uses a factorization machine, instead of dense spatiotemporal attention, to model cross-channel relationships in a compact transformer for time series forecasting.
The M4 Competition: 100,000 time series and 61 methods
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
FaCTR: Factorized Channel-Temporal Representation Transformers for Efficient Time Series Forecasting
FaCTR uses a factorization machine, instead of dense spatiotemporal attention, to model cross-channel relationships in a compact transformer for time series forecasting.