Shailza Sharma
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shailza Sharma has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
Abhinav Dhall
2 shared publications- Vinay Kumar
Thapar Institute of Engineering & Technology
2 shared publications - Abhinav Dhall
Monash University, Melbourne, Australia
1 shared publication - Andrew Hooper
COMET School of Earth and Environment University of Leeds Leeds UK
1 shared publication Camila Novoa Lizamaa
1 shared publicationDavid Hogg
1 shared publication
Latest Publications
- Open Access
- 17 February 2026
- Published by Elsevier in SSRN Electronic Journal
Global Volcanic Deformation Forecasting: A ConvLSTM based Spatio-Temporal Framework with Temporal Attention
Monitoring volcanic deformation is essential for assessing eruption risk and mitigating hazards, but existing forecasting methods are typically site-specific and struggle with noisy, incomplete data. Addressing this gap requires scalable models capable of learning from diverse volcanic behaviors worldwide. We present the first global-scale volcanic deformation forecasting framework trained entirely on physics-based synthetic data and validated on real-world volcanoes. Our approach leverages a deep learning architecture based on Convolutional Long Short-Term Memory (ConvLSTM) networks, augmented with a temporal attention mechanism (TA-ConvLSTM), to forecast complex deformation signals at a global scale. Unlike previous methods tailored to individual volcanoes, our model is trained on a globally representative synthetic dataset of unwrapped Interferometric Synthetic Aperture Radar (InSAR) interferograms. These sequences capture diverse deformation scenarios including magma-driven inflation-deflation cycles, variable displacement patterns, and realistic atmospheric noise, enabling robust generalization to different volcanic settings. A temporal aggregation strategy condenses 120-frame input sequences into 25 informative frames, preserving essential dynamics while reducing computational cost. Training employs a multi-step compound loss and a curriculum-based learning schedule with scheduled sampling, while the temporal attention module adaptively emphasizes the most informative timesteps. Experiments on four geophysically and tectonically diverse volcanoes (Socompa, Domuyo, Campi Flegrei, and Wolf) using Sentinel-1 InSAR time series demonstrate that the globally trained TA-ConvLSTM achieves accurate forecasts of deformation evolution for one year without site-specific retraining. These results establish a proof of concept for operational, global volcanic deformation forecasting and represent a step toward scalable, data-driven early warning systems deployable worldwide.
- 17 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessGlobal Volcanic Deformation Forecasting: A ConvLSTM based Spatio-Temporal Framework with Temporal Attention
Monitoring volcanic deformation is essential for assessing eruption risk and mitigating hazards, but existing forecasting methods are typically site-specific and struggle with noisy, incomplete data. Addressing this gap requires scalable models capable of learning from diverse volcanic behaviors worldwide. We present the first global-scale volcanic deformation forecasting framework trained entirely on physics-based synthetic data and validated on real-world volcanoes. Our approach leverages a deep learning architecture based on Convolutional Long Short-Term Memory (ConvLSTM) networks, augmented with a temporal attention mechanism (TA-ConvLSTM), to forecast complex deformation signals at a global scale. Unlike previous methods tailored to individual volcanoes, our model is trained on a globally representative synthetic dataset of unwrapped Interferometric Synthetic Aperture Radar (InSAR) interferograms. These sequences capture diverse deformation scenarios including magma-driven inflation-deflation cycles, variable displacement patterns, and realistic atmospheric noise, enabling robust generalization to different volcanic settings. A temporal aggregation strategy condenses 120-frame input sequences into 25 informative frames, preserving essential dynamics while reducing computational cost. Training employs a multi-step compound loss and a curriculum-based learning schedule with scheduled sampling, while the temporal attention module adaptively emphasizes the most informative timesteps. Experiments on four geophysically and tectonically diverse volcanoes (Socompa, Domuyo, Campi Flegrei, and Wolf) using Sentinel-1 InSAR time series demonstrate that the globally trained TA-ConvLSTM achieves accurate forecasts of deformation evolution for one year without site-specific retraining. These results establish a proof of concept for operational, global volcanic deformation forecasting and represent a step toward scalable, data-driven early warning systems deployable worldwide.
- 11 November 2024
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 250
Dual stage semantic information based generative adversarial network for image super-resolution
- 11 November 2024
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 250
Dual stage semantic information based generative adversarial network for image super-resolution
- Open Access
- 15 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 1-9
Dual Stage Semantic Information Based Generative Adversarial Network For Image Super-Resolution✱
Deep learning methods for the super-resolution problem are showing great performance compared to other traditional techniques. However, these methods are unable to learn complex spatial structures and high frequency details; which leads to over-smooth results. In the present paper, a novel Generative Adversarial Network based architecture named as Residue and Semantic feature based Dual Subpixel Generative Adversarial Network has been proposed for generator and discriminator networks to solve super-resolution problem. The generator network is residue and semantic feature based dual subpixel generative architecture. This architecture is divided into two stages: premier residual stage and deuxieme residual stage. These two stages are concatenated together to form a two stage upsamping process, which enhances the feature learning capability of our model. Inter and intra residual connections are made within these two stages; helping us to sustain the high texture details of images. Semantic based information is implanted in generator to enhance the quality of objects in an image. For embedding semantic information in generator, feature maps extracted from pre-trained model are merged with the input image. To stabilize the training process, we introduced spectral normalization in the discriminator. Visual perception and mean opinion score shows that proposed method outperforms the other state-of-the-art methods.
- 15 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 1-9
Open AccessDual Stage Semantic Information Based Generative Adversarial Network For Image Super-Resolution✱
Deep learning methods for the super-resolution problem are showing great performance compared to other traditional techniques. However, these methods are unable to learn complex spatial structures and high frequency details; which leads to over-smooth results. In the present paper, a novel Generative Adversarial Network based architecture named as Residue and Semantic feature based Dual Subpixel Generative Adversarial Network has been proposed for generator and discriminator networks to solve super-resolution problem. The generator network is residue and semantic feature based dual subpixel generative architecture. This architecture is divided into two stages: premier residual stage and deuxieme residual stage. These two stages are concatenated together to form a two stage upsamping process, which enhances the feature learning capability of our model. Inter and intra residual connections are made within these two stages; helping us to sustain the high texture details of images. Semantic based information is implanted in generator to enhance the quality of objects in an image. For embedding semantic information in generator, feature maps extracted from pre-trained model are merged with the input image. To stabilize the training process, we introduced spectral normalization in the discriminator. Visual perception and mean opinion score shows that proposed method outperforms the other state-of-the-art methods.
- 5 September 2022
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 223
Frequency aware face hallucination generative adversarial network with semantic structural constraint
- 5 September 2022
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 223
Frequency aware face hallucination generative adversarial network with semantic structural constraint