We address the challenge of getting efficient yet accurate recognition systems with limited labels. While recognition models improve with model size and amount of data, many specialized applications of computer vision have severe resource constraints both during training and inference. Transfer learning is an effective solution for training with few labels, however often at the expense of a compu- tationally costly fine-tuning of large base models. We propose to mitigate this unpleasant trade-off between compute and accuracy via semi-supervised cross- domain distillation from a set of diverse source models. Initially, we show how to use task similarity metrics to select a single suitable source model to distill from, and that a good selection process is imperative for good downstream performance of a target model. We dub this approach DISTILLNEAREST. Though effective, DISTILLNEAREST assumes a single source model matches the target task, which is not always the case. To alleviate this, we propose a weighted multi-source distilla- tion method to distill multiple source models trained on different domains weighted by their relevance for the target task into a single efficient model (named DISTILL- WEIGHTED). Our methods need no access to source data, and merely need features and pseudo-labels of the source models. When the goal is accurate recognition under computational constraints, both DISTILLNEAREST and DISTILLWEIGHTED approaches outperform both transfer learning from strong ImageNet initializations as well as state-of-the-art semi-supervised techniques such as FixMatch. Averaged over 8 diverse target tasks our multi-source method outperforms the baselines by 5.6%-points and 4.5%-points, respectively.
more »
« less
Graph Representation Learning via Multi-task Knowledge Distillation
Machine learning on graph structured data has attracted much research interest due to its ubiquity in real world data. However, how to efficiently represent graph data in a general way is still an open problem. Traditional methods use handcraft graph features in a tabular form but suffer from the defects of domain expertise requirement and information loss. Graph representation learning overcomes these defects by automatically learning the continuous representations from graph structures, but they require abundant training labels, which are often hard to fulfill for graph-level prediction problems. In this work, we demonstrate that, if available, the domain expertise used for designing handcraft graph features can improve the graph-level representation learning when training labels are scarce. Specifically, we proposed a multi-task knowledge distillation method. By incorporating network-theory-based graph metrics as auxiliary tasks, we show on both synthetic and real datasets that the proposed multi-task learning method can improve the prediction performance of the original learning task, especially when the training data size is small.
more »
« less
- Award ID(s):
- 1633370
- NSF-PAR ID:
- 10131234
- Date Published:
- Journal Name:
- 33rd Conference on Neural Information Processing Systems (NeurIPS 2019) Graph Representation Learning Workshop
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Electronic health records (EHRs) have been heavily used in modern healthcare systems for recording patients' admission information to health facilities. Many data-driven approaches employ temporal features in EHR for predicting specific diseases, readmission times, and diagnoses of patients. However, most existing predictive models cannot fully utilize EHR data, due to an inherent lack of labels in supervised training for some temporal events. Moreover, it is hard for the existing methods to simultaneously provide generic and personalized interpretability. To address these challenges, we propose Sherbet, a self-supervised graph learning framework with hyperbolic embeddings for temporal health event prediction. We first propose a hyperbolic embedding method with information flow to pretrain medical code representations in a hierarchical structure. We incorporate these pretrained representations into a graph neural network (GNN) to detect disease complications and design a multilevel attention method to compute the contributions of particular diseases and admissions, thus enhancing personalized interpretability. We present a new hierarchy-enhanced historical prediction proxy task in our self-supervised learning framework to fully utilize EHR data and exploit medical domain knowledge. We conduct a comprehensive set of experiments on widely used publicly available EHR datasets to verify the effectiveness of our model. Our results demonstrate the proposed model's strengths in both predictive tasks and interpretable abilities.more » « less
-
Predicting the occurrence of a particular event of interest at future time points is the primary goal of survival analysis. The presence of incomplete observations due to time limitations or loss of data traces is known as censoring which brings unique challenges in this domain and differentiates survival analysis from other standard regression methods. The popularly used survival analysis methods such as Cox proportional hazard model and parametric survival regression suffer from some strict assumptions and hypotheses that are not realistic in most of the real-world applications. To overcome the weaknesses of these two types of methods, in this paper, we reformulate the survival analysis problem as a multi-task learning problem and propose a new multi-task learning based formulation to predict the survival time by estimating the survival status at each time interval during the study duration. We propose an indicator matrix to enable the multi-task learning algorithm to handle censored instances and incorporate some of the important characteristics of survival problems such as non-negative non-increasing list structure into our model through max-heap projection. We employ the L2,1-norm penalty which enables the model to learn a shared representation across related tasks and hence select important features and alleviate over-fitting in high-dimensional feature spaces; thus, reducing the prediction error of each task. To efficiently handle the two non-smooth constraints, in this paper, we propose an optimization method which employs Alternating Direction Method of Multipliers (ADMM) algorithm to solve the proposed multi-task learning problem. We demonstrate the performance of the proposed method using real-world microarray gene expression high-dimensional benchmark datasets and show that our method outperforms state-of-the-art methods.more » « less
-
A key assumption in multi-task learning is that at the inference time the multi-task model only has access to a given data point but not to the data point’s labels from other tasks. This presents an opportunity to extend multi-task learning to utilize data point’s labels from other auxiliary tasks, and this way improves performance on the new task. Here we introduce a novel relational multi-task learning setting where we leverage data point labels from auxiliary tasks to make more accurate predictions on the new task. We develop MetaLink, where our key innovation is to build a knowledge graph that connects data points and tasks and thus allows us to leverage labels from auxiliary tasks. The knowledge graph consists of two types of nodes: (1) data nodes, where node features are data embeddings computed by the neural network, and (2) task nodes, with the last layer’s weights for each task as node features. The edges in this knowledge graph capture data-task relationships, and the edge label captures the label of a data point on a particular task. Under MetaLink, we reformulate the new task as a link label prediction problem between a data node and a task node. The MetaLink framework provides flexibility to model knowledge transfer from auxiliary task labels to the task of interest. We evaluate MetaLink on 6 benchmark datasets in both biochemical and vision domains. Experiments demonstrate that MetaLink can successfully utilize the relations among different tasks, outperforming the state-of-the-art methods under the proposed relational multi-task learning setting, with up to 27% improvement in ROC AUC.more » « less
-
Multilevel modeling and multi-task learning are two widely used approaches for modeling nested (multi-level) data, which contain observations that can be clustered into groups, characterized by their group-level features. Despite the similarity of the problems they address, the explicit relationship between multilevel modeling and multi-task learning has not been carefully examined. In this paper, we present a comparative analysis between the two methods to illustrate their strengths and limitations when applied to two-level nested data. We provide a detailed analysis demonstrating the equivalence of their formulations under a mild condition from an optimization perspective. We also demonstrate their limitations in terms of their predictive performance and especially, their difficulty in identifying potential cross-scale interactions between the local and group-level features when applied to datasets with either a small number of groups or limited training examples per group. To overcome these limitations, we propose a novel method for disaggregating the coarse-scale values of the group-level features in the nested data. Experimental results on both synthetic and real-world data show that the disaggregated group-level features can help enhance the prediction accuracy of the models significantly and identify the cross-scale interactions more effectively.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
