Processing of missing data by neural networks

Abstract

We propose a general, theoretically justified mechanism for processing missing data by neural networks. Our idea is to replace typical neuron's response in the first hidden layer by its expected value. This approach can be applied for various types of networks at minimal cost in their modification. Moreover, in contrast to recent approaches, it does not require complete data for training. Experimental results performed on different types of architectures show that our method gives better results than typical imputation strategies and other methods dedicated for incomplete data.

References in corpus (2)

• GAIN: Missing Data Imputation using Generative Adversarial Nets
• Classification with Low Rank and Missing Data

Cited by in corpus (13)

• MAIN: Multihead-Attention Imputation Networks
• Missing Data Imputation with Adversarially-trained Graph Convolutional Networks
• Graph Convolutional Networks for Graphs Containing Missing Features
• Handling Missing Data with Graph Representation Learning
• Decoupling Long- and Short-Term Patterns in Spatiotemporal Inference
• Why Not to Use Zero Imputation? Correcting Sparsity Bias in Training Neural Networks
• Spatial Aggregation and Temporal Convolution Networks for Real-time Kriging
• Forecasting Loss of Signal in Optical Networks with Machine Learning
• Data Imputation by Pursuing Better Classification: A Supervised Kernel-Based Method
• Towards Better Modeling with Missing Data: A Contrastive Learning-based Visual Analytics Perspective
• Robustness to Missing Features using Hierarchical Clustering with Split Neural Networks
• Split: Inferring Unobserved Event Probabilities for Disentangling Brand-Customer Interactions
• Learning from Incomplete Features by Simultaneous Training of Neural Networks and Sparse Coding