Time-series forecasting is an important, yet often very complex task due to inherent complex nonlinear dynamics and random fluctuations. Signal decomposition is often applied to separate deterministic from random signal parts to support the construction of forecasting models. In our new research, Temporal Dictionary Learning for Time-Series Decomposition, we demonstrate the application of Dictionary Learning (DL) for signal decomposition. DL represents complex time series data as deterministic, sparse, spatio-temporal events. This provides a theoretical foundation for further methodological development in the decomposition and forecasting of time-series data.