HydaLearn: Highly Dynamic Task Weighting for Multi-task Learning with Auxiliary Tasks

Multi-task learning (MTL) can improve performance on a task by sharing representations with one or more related auxiliary-tasks. Usually, MTL-networks are trained on a composite loss function formed by a constant weighted combination of the separate task losses. In practice, constant loss weights lead to poor results for two reasons: (i) the relevance of the auxiliary tasks can gradually drift throughout the learning process; (ii) for mini-batch based optimisation, the optimal task weights vary significantly from one update to the next depending on mini-batch sample composition. We introduce HydaLearn, an intelligent weighting algorithm that connects main-task gain to the individual task gradients, in order to inform dynamic loss weighting at the mini-batch level, addressing i and ii. Using HydaLearn, we report performance increases on synthetic data, as well as on two supervised learning domains.

[1]  Joachim Bingel,et al.  Identifying beneficial task relations for multi-task learning in deep neural networks , 2017, EACL.

[2]  Rich Caruana,et al.  Multitask Learning , 1998, Encyclopedia of Machine Learning and Data Mining.

[3]  M. Saeed.  Multiparameter Intelligent Monitoring in Intensive Care II ( MIMIC-II ) : A public-access intensive care unit database , 2011 .

[4]  Massimiliano Pontil,et al.  Exploiting Unrelated Tasks in Multi-Task Learning , 2012, AISTATS.

[5]  Rich Caruana,et al.  Multitask Learning: A Knowledge-Based Source of Inductive Bias , 1993, ICML.

[6]  Jiebo Luo,et al.  End-to-end Multi-Modal Multi-Task Vehicle Control for Self-Driving Cars with Visual Perceptions , 2018, 2018 24th International Conference on Pattern Recognition (ICPR).

[7]  Jason Weston,et al.  A unified architecture for natural language processing: deep neural networks with multitask learning , 2008, ICML '08.

[8]  Jürgen Schmidhuber,et al.  Flat Minima , 1997, Neural Computation.

[9]  Razvan Pascanu,et al.  Adapting Auxiliary Losses Using Gradient Similarity , 2018, ArXiv.

[10]  Li Fei-Fei,et al.  Dynamic Task Prioritization for Multitask Learning , 2018, ECCV.

[11]  Peter Szolovits,et al.  MIMIC-III, a freely accessible critical care database , 2016, Scientific Data.

[12]  Aram Galstyan,et al.  Multitask learning and benchmarking with clinical time series data , 2017, Scientific Data.

[13]  Elad Hoffer,et al.  Train longer, generalize better: closing the generalization gap in large batch training of neural networks , 2017, NIPS.

[14]  Massimiliano Pontil,et al.  The Benefit of Multitask Representation Learning , 2015, J. Mach. Learn. Res..

[15]  Jorge Nocedal,et al.  On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima , 2016, ICLR.

[16]  Xiaoou Tang,et al.  Facial Landmark Detection by Deep Multi-task Learning , 2014, ECCV.

[17]  Karol Hausman,et al.  Gradient Surgery for Multi-Task Learning , 2020, NeurIPS.

[18]  Richard S. Sutton,et al.  Adapting Bias by Gradient Descent: An Incremental Version of Delta-Bar-Delta , 1992, AAAI.

[19]  R. Caruana.  Learning From Imbalanced Data: Rank Metrics and Extra Tasks , 2000 .

[20]  Ramakanth Pasunuru,et al.  AutoSeM: Automatic Task Selection and Mixing in Multi-Task Learning , 2019, NAACL.

[21]  Sinan Kalkan,et al.  Imbalance Problems in Object Detection: A Review , 2020, IEEE transactions on pattern analysis and machine intelligence.

[22]  Robert Steele,et al.  Predicting Hospital Length of Stay Using Neural Networks on MIMIC III Data , 2017, 2017 IEEE 15th Intl Conf on Dependable, Autonomic and Secure Computing, 15th Intl Conf on Pervasive Intelligence and Computing, 3rd Intl Conf on Big Data Intelligence and Computing and Cyber Science and Technology Congress(DASC/PiCom/DataCom/CyberSciTech).

[23]  Joachim Bingel,et al.  Latent Multi-Task Architecture Learning , 2017, AAAI.

[24]  Bart Baesens,et al.  Neural network survival analysis for personal loan data , 2005, J. Oper. Res. Soc..

[25]  Dario Amodei,et al.  An Empirical Model of Large-Batch Training , 2018, ArXiv.

[26]  Tomaso A. Poggio,et al.  Theory of Deep Learning IIb: Optimization Properties of SGD , 2018, ArXiv.

[27]  Vladlen Koltun,et al.  Multi-Task Learning as Multi-Objective Optimization , 2018, NeurIPS.

[28]  Quoc V. Le,et al.  A Bayesian Perspective on Generalization and Stochastic Gradient Descent , 2017, ICLR.

[29]  Roger G. Mark,et al.  Reproducibility in critical care: a mortality prediction case study , 2017, MLHC.

[30]  T. H. Kyaw,et al.  Multiparameter Intelligent Monitoring in Intensive Care II: A public-access intensive care unit database* , 2011, Critical care medicine.

[31]  Léon Bottou,et al.  Large-Scale Machine Learning with Stochastic Gradient Descent , 2010, COMPSTAT.

[32]  Kaiming He,et al.  Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour , 2017, ArXiv.

[33]  Yan Liu,et al.  Benchmark of Deep Learning Models on Large Healthcare MIMIC Datasets , 2017, ArXiv.

[34]  Shamim Nemati,et al.  An Interpretable Machine Learning Model for Accurate Prediction of Sepsis in the ICU , 2017, Critical care medicine.

[35]  Zhao Chen,et al.  GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks , 2017, ICML.

[36]  Joseph Gonzalez,et al.  On the Computational Inefficiency of Large Batch Sizes for Stochastic Gradient Descent , 2018, ArXiv.

[37]  Ross B. Girshick,et al.  Focal Loss for Dense Object Detection , 2017, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[38]  Sebastian Ruder,et al.  An overview of gradient descent optimization algorithms , 2016, Vestnik komp'iuternykh i informatsionnykh tekhnologii.

[39]  Véronique Van Vlasselaer,et al.  Fraud Analytics : Using Descriptive, Predictive, and Social Network Techniques:A Guide to Data Science for Fraud Detection , 2015 .

[40]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[41]  David Held,et al.  Adaptive Auxiliary Task Weighting for Reinforcement Learning , 2019, NeurIPS.

[42]  Harini Suresh,et al.  Learning Tasks for Multitask Learning: Heterogenous Patient Populations in the ICU , 2018, KDD.

[43]  L. Tarassenko,et al.  Dynamic Data During Hypotensive Episode Improves Mortality Predictions Among Patients With Sepsis and Hypotension* , 2013, Critical care medicine.

[44]  Yu Liu,et al.  Gradient Harmonized Single-stage Detector , 2018, AAAI.

[45]  Orhan Firat,et al.  Adaptive Scheduling for Multi-Task Learning , 2019, ArXiv.

[46]  Huajun Feng,et al.  Libra R-CNN: Towards Balanced Learning for Object Detection , 2019, 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[47]  Roberto Cipolla,et al.  Multi-task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics , 2017, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[48]  Tom M. Mitchell,et al.  Using the Future to Sort Out the Present: Rankprop and Multitask Learning for Medical Risk Evaluation , 1995, NIPS.

[49]  Hal Daumé,et al.  Infinite Predictor Subspace Models for Multitask Learning , 2010, AISTATS.