PMPrabhat K. Mahanti

Prabhat K. Mahanti

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Prabhat K. Mahanti has a total of 5 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Interpretable Forecasting in Multivariate Time Series Using Koopman Autoencoder and Fuzzy Transform

    by Vishwambhar Pathak, Jong-Kyou Kim, Prabhat Mahanti

    This study investigates the modeling efficacy of Koopman Autoencoder (KAE) in comparison with LSTM, GRU, and Transformer architectures for multivariate time series (MTS) prediction tasks. We additionally evaluate the effect of fuzzy logic-based feature transformations (triangular and Gaussian membership functions) on learning dynamics and predictive performance. Synthetic MTS data comprising 16 channels with varying levels of additive Gaussian noise (0.0 and 0.1) are used to simulate realistic conditions. Experiments are conducted both with fuzzy-transformed inputs and raw features to assess comparative robustness. Results indicate that Transformers with triangular fuzzy transformation achieved the best predictive accuracy. Quantitative results show that fuzzy-transformed inputs significantly reduce prediction error across all models, with Transformer achieving the best MSE (0.0516) and R² (0.043) under triangular F-transform. PCA and t-SNE visualizations affirmed the distinct latent separability of Transformer and LSTM embeddings. While KAE yielded slightly lower predictive performance (MSE > 0.06), KAE embedding demonstrated advantages in latent space linearized embedding with more compressed, interpretability via saliency maps, and robustness to structured signal evolution. The results suggest KAE with fuzzy-transformed features offers a promising trade-off between interpretability and performance for explainable AI in time series feature embedding and forecasting.

  • Open Access

    Optimizing Spectral Graph Kernels for Time-Delay Embedding of Multichannel EEG Data

    by Rashmi Nagawara Muralinath, Vishwambhar Pathak, Prabhat K. Mahanti

    The task of determining the dynamic patterns of causal connectivity among brain regions from EEG data is challenging due to inherent noise in EEG signals, sparse connections. High volume and dimensionality of data add substantial computational demands. Koopman embedding kernels offer a promising approach for capturing the low-dimensional linear embeddings of complex nonlinear system dynamics. This study focuses on leveraging the Graph Kernel Koopman Embedding (GKKE) method to extract latent factors, a novel application within the realm of EEG analysis. Specifically, the research pioneers the adaptation of the GKKE algorithm to estimate the inherent connectivity substructures, that enable investigations of the neuro-markers of cognitive load states and disorders like seizures from EEG signals. The graph for each segment was constructed by computing the correlation coefficients among the channel pairs. The algorithmic workflow was applied to a labelled cognitive load dataset from an online repository. The features were extracted in terms of Koopman-spectral decomposed clusters of eigenfunctions of the gram matrices produced by two different graph kernels applied for approximation of the Koopman operator over each EEG-segment. These features, representing the latent metastable states, were fed into machine learning classifiers, including SVM and Random Forest, and Decision Tree classifiers. Weisfeiler-Lehman Kernel (WL)-Kernel yielded an average accuracy of around 84% for all the classifiers and outperformed the Random-walk kernel. The significance of this work lies in its pioneering use of graph Koopman spectral decomposition-based low-dimensional embedding of the metastable structures of connectivity to provide an algorithm that enables neuro-marker estimation. Specific contribution is the workflow that determines the optimal combination of the graph kernel and classifier ensemble. This equips deeper understanding of the connectivity dynamics of brain regions related to cognitive states offers a new avenue for prognosis and training protocols to improve cognitive states.


  • Open Access

    Exploring the Efficacy of Explainable Deep Learning in Identifying Neuromarkers for Precise Prediction of Epilepsy and Causal Connectivity Analysis

    by Vishwambhar Pathak, Vivek Gaur, Prabhat K Mahanti, Satish Chandra

    Causal connectivity among the brain regions have been recently exploited information for discriminating epileptiforms to detect epileptic seizures. Published investigations to detect ictal, interictal, preictal EEG reported existence of long-range correlations of excitations within a functionally connected brain region and shifting of focus of excitations from one region to another region and increase or decrease of intensity in certain frequency-bands, which can be quantified using suitable measure of Granger causality (GC). Deep neural networks obviated explicit preprocessing and feature extraction. The proposed work employs temporal dilated convolutional network to estimate causal connectivity relations among brain-regions in various frequency-bands in distributed manner. It implicitly learns varying autoregressive-lag-orders using stacked layers and covers long range relationships using exponential update of layer-wise dilation-factor. Model training with several parameter-combinations were conducted over 10 subjects. The proposed model outperformed the existing approaches and baseline model in terms of accuracy, sensitivity, and false positive rate. Class-wise dominating features were obtained using statistical significance analysis followed by family wise error rate correction using Benjamini-Hochberg method.