Neural Networks for High-Dimensional Tabular Biomedical Datasets
Add to your list(s)
Download to your calendar using vCal
If you have a question about this talk, please contact Mateja Jamnik.
Modern machine learning algorithms frequently overfit on small-sample size and high-dimensional tabular datasets, which are common in medicine, bioinformatics and drug discovery. How can we reduce the overfitting on tabular datasets with D>>N?
This talk presents two neural methods for learning from small-sample size and high-dimensional tabular datasets. First, we present WPFS , a parameter-efficient neural architecture that performs global feature selection during training. Second, we present GCondNet, a general approach which combines Graph Neural Networks (GNNs) for incorporating the implicit relationships between samples when training standard neural networks. GCondNet exploits the high-dimensionality of the data by creating many small graphs to capture the structure between samples within a feature. We show that WPFS and GCondNet outperform both standard and more recent methods on real-world biomedical datasets.
You can also join us on Zoom
This talk is part of the Artificial Intelligence Research Group Talks (Computer Laboratory) series.
This talk is included in these lists:
Note that ex-directory lists are not shown.
|