Multi-task learning improves grape cold hardiness and budbreak predictions
Multi-Task Learning for Sparsely-Labeled Time Series: A Case Study on Cold-Hardiness Modeling
Machine Learning
Summary
Predicting when grapes can survive freezing temperatures is important for farmers to protect their crops. This is hard because the data about freezing damage is very limited and changes with different types of grape plants. The authors show that using multi-task learning, which trains a model to predict for multiple grape types at once, helps improve prediction accuracy. They also find that one model can learn to predict both freezing risk and when buds will start growing, making it more useful. Their approach outperforms traditional methods and single-type models.
multi-task learningrecurrent neural networkstime series datacold hardinessgrape cultivarstransfer learningbudbreak predictiontemporal sparsityagricultural modeling
Authors
Aseem Saxena, Paola Pesántez-Cabrera, Jonathan Magby, Markus Keller, Alan Fern
Abstract
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.