Selected work

Representative Papers

ChaosProbe: A Neurochaotic Lens on Frozen Transformer Input-Embedding Spaces

Aug 03, 2026

This work addresses the challenge of characterizing the input embedding space structure of frozen Transformers without relying on downstream tasks or contextual computations. It introduces ChaosProbe, a novel method that pioneers the application of neural chaos dynamics to embedding analysis: by applying deterministic chaotic trajectory transformations to input embeddings and combining neuronal firing rates with entropy responses, it generates fixed-length structural fingerprints. This approach requires no training or task-specific adaptation, yet effectively reveals macroscopic relationships among embedding spaces. Experiments across four pretrained models and 80 neutral prompts demonstrate that multiple similarity metrics consistently recover both intra-family nearest neighbors and inter-family pairings, confirming the stability and validity of the proposed fingerprints.

0 citationsRead paper

Benchmarking ConvLSTM for One-Day-Ahead IMDAA Rainfall-Field Prediction across Four Indian Cities

Jul 29, 2026

This study evaluates the capability of ConvLSTM to forecast next-day precipitation fields using low-frequency, fine-scale daily reanalysis data over the Indian monsoon region, focusing on IMDAA rainfall records from four cities during the monsoon seasons of 1998–2020. It presents the first systematic comparison of ConvLSTM against fully connected LSTM (FC-LSTM), persistence forecasting, statistical models, and tree-based methods across multiple evaluation dimensions—including full-field prediction, regional averages, spatial anomalies, and extreme rainfall events. Results indicate that ConvLSTM does not significantly outperform FC-LSTM or persistence models; neural networks consistently underestimate intense rainfall magnitudes, whereas persistence demonstrates superior skill in detecting high-rainfall days, particularly in regions with strong spatial coherence such as Mumbai. The findings highlight the critical influence of input structure and spatial continuity on model selection for precipitation forecasting.

0 citationsRead paper

Linked Data Classification using Neurochaos Learning

Feb 18, 2026

This work proposes the first application of Neurochaos Learning to node classification in knowledge graphs under conditions of scarce training samples and limited computational resources. By leveraging graph structural properties, the method designs a node aggregation strategy to extract informative features, which are then fed into a ChaosNet model for efficient classification. The approach significantly outperforms baseline models on assortative (homophilic) knowledge graphs, demonstrating its effectiveness in low-data and low-compute regimes. Although its performance is comparatively limited on disassortative (heterophilic) graphs, the study collectively highlights the promising potential and innovative value of Neurochaos Learning for relational data tasks.

0 citationsRead paper

Dependable Artificial Intelligence with Reliability and Security (DAIReS): A Unified Syndrome Decoding Approach for Hallucination and Backdoor Trigger Detection

Feb 06, 2026

This work addresses critical security and reliability threats in machine learning models—namely, backdoor data poisoning and hallucinations in large language models—that can lead to untrustworthy outputs. The authors propose a unified framework based on syndrome decoding to simultaneously detect backdoored training samples and hallucinated content within the sentence embedding space. This approach represents the first application of syndrome decoding to both security and reliability challenges, leveraging meta-explanatory task modeling and an anomaly discrimination mechanism to enable generalizable detection across distinct problem settings. Experimental results demonstrate that the method effectively distinguishes poisoned from clean data and accurately identifies hallucinations through reflexive meta-explanations, thereby significantly enhancing the trustworthiness of AI systems.

0 citationsRead paper

Deep Learning for Short-Term Precipitation Prediction in Four Major Indian Cities: A ConvLSTM Approach with Explainable AI

Nov 14, 2025

To address the limited interpretability of deep learning models for precipitation forecasting—which hinders operational deployment—this paper proposes an interpretable short-term (1–5 day) deep learning framework tailored to four major climatic zones in India (Bengaluru, Mumbai, Delhi, Kolkata). Methodologically, we design a city-adaptive Time-Distributed CNN-ConvLSTM architecture integrated with ERA5 reanalysis data and incorporate multimodal eXplainable AI (xAI) techniques: Grad-CAM for spatiotemporal saliency mapping, permutation importance for identifying dominant meteorological variables, and temporal occlusion with counterfactual perturbation to quantify temporal sensitivity. Experiments demonstrate high accuracy (RMSE: 0.21–1.80 mm/day) and robustness across diverse climatic regimes. This work constitutes the first systematic implementation of a regionally adaptive, end-to-end interpretable precipitation forecasting pipeline, significantly enhancing transparency and trustworthiness of AI-driven meteorological decision-making.

0 citationsRead paper

Recent publications

Latest Papers

ChaosProbe: A Neurochaotic Lens on Frozen Transformer Input-Embedding Spaces

Aug 03, 2026

This work addresses the challenge of characterizing the input embedding space structure of frozen Transformers without relying on downstream tasks or contextual computations. It introduces ChaosProbe, a novel method that pioneers the application of neural chaos dynamics to embedding analysis: by applying deterministic chaotic trajectory transformations to input embeddings and combining neuronal firing rates with entropy responses, it generates fixed-length structural fingerprints. This approach requires no training or task-specific adaptation, yet effectively reveals macroscopic relationships among embedding spaces. Experiments across four pretrained models and 80 neutral prompts demonstrate that multiple similarity metrics consistently recover both intra-family nearest neighbors and inter-family pairings, confirming the stability and validity of the proposed fingerprints.

0 citationsRead paper

Benchmarking ConvLSTM for One-Day-Ahead IMDAA Rainfall-Field Prediction across Four Indian Cities

Jul 29, 2026

This study evaluates the capability of ConvLSTM to forecast next-day precipitation fields using low-frequency, fine-scale daily reanalysis data over the Indian monsoon region, focusing on IMDAA rainfall records from four cities during the monsoon seasons of 1998–2020. It presents the first systematic comparison of ConvLSTM against fully connected LSTM (FC-LSTM), persistence forecasting, statistical models, and tree-based methods across multiple evaluation dimensions—including full-field prediction, regional averages, spatial anomalies, and extreme rainfall events. Results indicate that ConvLSTM does not significantly outperform FC-LSTM or persistence models; neural networks consistently underestimate intense rainfall magnitudes, whereas persistence demonstrates superior skill in detecting high-rainfall days, particularly in regions with strong spatial coherence such as Mumbai. The findings highlight the critical influence of input structure and spatial continuity on model selection for precipitation forecasting.

0 citationsRead paper

Linked Data Classification using Neurochaos Learning

Feb 18, 2026

This work proposes the first application of Neurochaos Learning to node classification in knowledge graphs under conditions of scarce training samples and limited computational resources. By leveraging graph structural properties, the method designs a node aggregation strategy to extract informative features, which are then fed into a ChaosNet model for efficient classification. The approach significantly outperforms baseline models on assortative (homophilic) knowledge graphs, demonstrating its effectiveness in low-data and low-compute regimes. Although its performance is comparatively limited on disassortative (heterophilic) graphs, the study collectively highlights the promising potential and innovative value of Neurochaos Learning for relational data tasks.

0 citationsRead paper

Dependable Artificial Intelligence with Reliability and Security (DAIReS): A Unified Syndrome Decoding Approach for Hallucination and Backdoor Trigger Detection

Feb 06, 2026

This work addresses critical security and reliability threats in machine learning models—namely, backdoor data poisoning and hallucinations in large language models—that can lead to untrustworthy outputs. The authors propose a unified framework based on syndrome decoding to simultaneously detect backdoored training samples and hallucinated content within the sentence embedding space. This approach represents the first application of syndrome decoding to both security and reliability challenges, leveraging meta-explanatory task modeling and an anomaly discrimination mechanism to enable generalizable detection across distinct problem settings. Experimental results demonstrate that the method effectively distinguishes poisoned from clean data and accurately identifies hallucinations through reflexive meta-explanations, thereby significantly enhancing the trustworthiness of AI systems.

0 citationsRead paper

Deep Learning for Short-Term Precipitation Prediction in Four Major Indian Cities: A ConvLSTM Approach with Explainable AI

Nov 14, 2025

To address the limited interpretability of deep learning models for precipitation forecasting—which hinders operational deployment—this paper proposes an interpretable short-term (1–5 day) deep learning framework tailored to four major climatic zones in India (Bengaluru, Mumbai, Delhi, Kolkata). Methodologically, we design a city-adaptive Time-Distributed CNN-ConvLSTM architecture integrated with ERA5 reanalysis data and incorporate multimodal eXplainable AI (xAI) techniques: Grad-CAM for spatiotemporal saliency mapping, permutation importance for identifying dominant meteorological variables, and temporal occlusion with counterfactual perturbation to quantify temporal sensitivity. Experiments demonstrate high accuracy (RMSE: 0.21–1.80 mm/day) and robustness across diverse climatic regimes. This work constitutes the first systematic implementation of a regionally adaptive, end-to-end interpretable precipitation forecasting pipeline, significantly enhancing transparency and trustworthiness of AI-driven meteorological decision-making.

0 citationsRead paper