Novel hypergraph-based ML algorithms for semi-supervised & multi-label classification — published at ICPR 2024 & CODS-COMAD 2024
Developed two novel hypergraph-based machine learning algorithms that exploit higher-order relationships in data (beyond pairwise graph edges) for semi-supervised and multi-label classification tasks.
Key Contributions
Improved Hypergraph Laplacian SVM — a semi-supervised SVM that uses an improved hypergraph Laplacian as a regularizer, enabling effective use of unlabeled data
Hypergraph LS-TSVM — a least squares twin SVM extended with hypergraph regularization for multi-label scenarios
Imbalanced Multi-label Classification — a hypergraph framework with label-specific features to handle class imbalance
Results
Delivered superior performance over state-of-the-art baselines on:
UCI ML Repository benchmark datasets
MNIST Fashion
NLP text classification datasets
Implemented high-performance retrieval pipelines using FAISS vector storage for efficient nearest-neighbour search in high-dimensional feature spaces.
This work presents a hypergraph-guided framework for feature relevance estimation in multi-label learning scenarios where labels are partially missing, enabling more robust classification under incomplete supervision.
@inproceedings{nirwal2025hypergraph,title={Leveraging Hypergraph-Guided Feature Relevance in Multi-Label Learning with Missing Labels},author={Rastogi, Reshma and Nirwal, Dev},booktitle={International Conference on Machine Learning and Data Engineering (ICMLDE)},year={2025},}
2024
ICPR
Improved Hypergraph Laplacian Based Semi-Supervised Support Vector Machine
Reshma Rastogi and Dev Nirwal
In Pattern Recognition: 27th International Conference, ICPR 2024, Kolkata, India, Dec 2024
A novel semi-supervised support vector machine using improved hypergraph Laplacian regularization, achieving superior performance on standard ML benchmarks by leveraging higher-order data relationships.
@inproceedings{nirwal2024improved,title={Improved Hypergraph {Laplacian} Based Semi-Supervised Support Vector Machine},author={Rastogi, Reshma and Nirwal, Dev},booktitle={Pattern Recognition: 27th International Conference, ICPR 2024, Kolkata, India},year={2024},month=dec,publisher={Springer},}
ICPR
Hypergraph Regularized Semi-supervised Least Squares Twin Support Vector Machine for Multilabel Classification
Reshma Rastogi and Dev Nirwal
In Pattern Recognition: 27th International Conference, ICPR 2024, Kolkata, India, Dec 2024
A hypergraph-regularized semi-supervised least squares twin SVM for multilabel classification, demonstrating superior performance on UCI ML Repository, MNIST Fashion, and NLP benchmark datasets.
@inproceedings{nirwal2024lstsvm,title={Hypergraph Regularized Semi-supervised Least Squares Twin Support Vector Machine for Multilabel Classification},author={Rastogi, Reshma and Nirwal, Dev},booktitle={Pattern Recognition: 27th International Conference, ICPR 2024, Kolkata, India},year={2024},month=dec,publisher={Springer},}
CODS-COMAD
HyperGraph Based Imbalance Multi-label Classification with Label Specific Features
Reshma Rastogi and Dev Nirwal
In Proceedings of the 7th Joint International Conference on Data Science & Management of Data (CODS-COMAD 2024), 2024
A hypergraph-based approach for imbalanced multi-label classification leveraging label-specific features to handle class imbalance and complex label correlations effectively.
@inproceedings{nirwal2024imbalance,title={{HyperGraph} Based Imbalance Multi-label Classification with Label Specific Features},author={Rastogi, Reshma and Nirwal, Dev},booktitle={Proceedings of the 7th Joint International Conference on Data Science \& Management of Data (CODS-COMAD 2024)},year={2024},publisher={ACM},}