Proposed a hypergraph neural network framework based on a tensor representation of the hypergraph structure.

 The paper is published on IEEE Transaction on Neural Networks and Learning Systems

Proved the equivalence between hypergraph neural networks and signal denoising, proposed an iterative approach to address the over-smoothing problem.

 The paper was published on the European conference on signal processing.

Addressed the challenge of learning the underlying hypergraph topology from the data by assuming the data possesses a certain regularity or smoothness. Demonstrated the effectiveness of the learned hypergraph structure in hypergraph learning-convolutional neural networks (t-HyperGLNN).

 The paper was accepted by IEEE Transactions on Network Signal Processing over Networks

Proposed a hierarchical pooing and unpooling layer that acts as a backbone of hypergraph autoencoder, performed signal reconstruciton, hypergraph classification, and anomaly detection and demonstrated competitive and robust results.

The paper was submitted after minor revision for IEEE Transaction on Pattern Analysis and Machine Intelligence