Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
To tackle extreme label sparsity in temporal biological modeling, this work evaluates multi-task learning (MTL) recurrent neural network architectures trained on daily meteorological time series across distinct plant cultivars. Predicting cold hardiness and budbreak is vital for agricultural frost protection, yet acquiring dense ground-truth measurements across heterogeneous cultivars is practically infeasible. The authors find that specific MTL configurations consistently outperform standard single-task baselines and domain-specific scientific models, with joint MTL across distinct phenological targets (budbreak and cold hardiness) yielding mutually reinforcing performance gains.
Jointly modeling distinct phenological tasks in a shared multi-task RNN overcomes severe temporal label scarcity, decisively outperforming established domain-specific scientific process models.
We present a real-world case study of multi-task learning (MTL) for temporal process modeling from limited data with temporally sparse labels. Specifically, we investigate multi-task learning for the important agricultural problem of predicting grape cold hardiness, which is the temperature at which lethal freezing occurs. Cold hardiness changes in response to weather and is difficult to measure directly in the field. Thus, growers rely on predictions to decide when to apply costly frost mitigation measures. We apply recurrent neural networks (RNNs) for daily cold-hardiness prediction from time series weather data. A major challenge is that the cold hardiness response varies across plant cultivars and ground-truth data for each cultivar is temporally sparse and limited. To address this challenge, we investigate multi-task learning (MTL) approaches for combining data, where different tasks correspond to different cultivars. We develop a variety of MTL architectures and evaluate them in both MTL and transfer learning settings. Our results show significant differences between architectures and that certain architectures are able to consistently outperform single-task learning and state-of-the-art scientific models. Additionally, we show similar results for the qualitatively different, but related, task of budbreak prediction. Further, improved accuracy for budbreak and cold hardiness is achieved by a single MTL model that simultaneously learns both tasks.