Poster: Link between Bias, Node Sensitivity and Long-Tail Distribution in trained DNNs
This work addresses a challenge for DNNs in real-world applications where training data has long-tail distributions, but it is incremental as it builds on existing bias studies.
The paper tackles the problem of varying classification performance in deep neural networks (DNNs) trained on long-tail datasets by identifying node bias that leads to different node sensitivities across output classes, supported by an empirical case study on a real-world dataset.
Owing to their remarkable learning (and relearning) capabilities, deep neural networks (DNNs) find use in numerous real-world applications. However, the learning of these data-driven machine learning models is generally as good as the data available to them for training. Hence, training datasets with long-tail distribution pose a challenge for DNNs, since the DNNs trained on them may provide a varying degree of classification performance across different output classes. While the overall bias of such networks is already highlighted in existing works, this work identifies the node bias that leads to a varying sensitivity of the nodes for different output classes. To the best of our knowledge, this is the first work highlighting this unique challenge in DNNs, discussing its probable causes, and providing open challenges for this new research direction. We support our reasoning using an empirical case study of the networks trained on a real-world dataset.