Saroj Gopali

Saroj Gopali, Bipin Chhetri, Deepika Giri et al.

Cyberattacks on operational technology are increasingly causing costly downtime and physical damage, exposing the limitations of traditional rule-based monitoring in industrial IoT environments. While Large Language Models (LLMs) have strong semantic reasoning abilities to assist in decision support, their hallucinatory nature presents unacceptable safety liabilities for closed-loop control. This paper introduces a neuro-agentic control framework, a novel architecture that couples an LLM-based planner (i.e., such as Gemini 2.5 Flash-Lite) with a pre-trained Time-Series Foundation Model (TimesFM), to achieve physics-grounded autonomous defense. The paper introduces a ``Counterfactual Physics Injection'' mechanism that simulates the impact of LLM-proposed interventions within the numerical latent space of the foundation model before actuation, while allowing the system to reject hallucinatory or unsafe actions. Evaluated on an industrial dataset (e.g., the Secure Water Treatment (SWaT)) in the context of stochastic attack scenarios, the framework exhibited better performance compared to LSTM and TCN baselines. The Neuro-Agentic Loop prevented five breaches (33.3%) below the threshold versus LSTM (26.7%) and TCN (13.3%), with zero physically invalid (hallucinated) actions executed. These results demonstrate the efficacy of using foundation models as deterministic ``Sentinels'' to safeguard agentic AI in critical infrastructure.

Bipin Chhetri, Saroj Gopali, Rukayat Olapojoye et al.

Federated learning is a decentralized machine learning paradigm that allows multiple clients to collaborate by leveraging local computational power and the models transmission. This method reduces the costs and privacy concerns associated with centralized machine learning methods while ensuring data privacy by distributing training data across heterogeneous devices. On the other hand, federated learning has the drawback of data leakage due to the lack of privacy-preserving mechanisms employed during storage, transfer, and sharing, thus posing significant risks to data owners and suppliers. Blockchain technology has emerged as a promising technology for offering secure data-sharing platforms in federated learning, especially in Industrial Internet of Things (IIoT) settings. This survey aims to compare the performance and security of various data privacy mechanisms adopted in blockchain-based federated learning architectures. We conduct a systematic review of existing literature on secure data-sharing platforms for federated learning provided by blockchain technology, providing an in-depth overview of blockchain-based federated learning, its essential components, and discussing its principles, and potential applications. The primary contribution of this survey paper is to identify critical research questions and propose potential directions for future research in blockchain-based federated learning.

Saroj Gopali, Akbar S. Namin, Faranak Abri et al.

Cyber attacks continue to pose significant threats to individuals and organizations, stealing sensitive data such as personally identifiable information, financial information, and login credentials. Hence, detecting malicious websites before they cause any harm is critical to preventing fraud and monetary loss. To address the increasing number of phishing attacks, protective mechanisms must be highly responsive, adaptive, and scalable. Fortunately, advances in the field of machine learning, coupled with access to vast amounts of data, have led to the adoption of various deep learning models for timely detection of these cyber crimes. This study focuses on the detection of phishing websites using deep learning models such as Multi-Head Attention, Temporal Convolutional Network (TCN), BI-LSTM, and LSTM where URLs of the phishing websites are treated as a sequence. The results demonstrate that Multi-Head Attention and BI-LSTM model outperform some other deep learning-based algorithms such as TCN and LSTM in producing better precision, recall, and F1-scores.

Saroj Gopali, Sima Siami-Namini, Faranak Abri et al.

As an intriguing case is the goodness of the machine and deep learning models generated by these LLMs in conducting automated scientific data analysis, where a data analyst may not have enough expertise in manually coding and optimizing complex deep learning models and codes and thus may opt to leverage LLMs to generate the required models. This paper investigates and compares the performance of the mainstream LLMs, such as ChatGPT, PaLM, LLama, and Falcon, in generating deep learning models for analyzing time series data, an important and popular data type with its prevalent applications in many application domains including financial and stock market. This research conducts a set of controlled experiments where the prompts for generating deep learning-based models are controlled with respect to sensitivity levels of four criteria including 1) Clarify and Specificity, 2) Objective and Intent, 3) Contextual Information, and 4) Format and Style. While the results are relatively mix, we observe some distinct patterns. We notice that using LLMs, we are able to generate deep learning-based models with executable codes for each dataset seperatly whose performance are comparable with the manually crafted and optimized LSTM models for predicting the whole time series dataset. We also noticed that ChatGPT outperforms the other LLMs in generating more accurate models. Furthermore, we observed that the goodness of the generated models vary with respect to the ``temperature'' parameter used in configuring LLMS. The results can be beneficial for data analysts and practitioners who would like to leverage generative AIs to produce good prediction models with acceptable goodness.

Nikesh Prajapati, Bimal Karki, Saroj Gopali et al.

This paper intends to detect IoT malicious attacks through deep learning models and demonstrates a comprehensive evaluation of the deep learning and graph-based models regarding malicious network traffic detection. The models particularly are based on GraphSAGE, Bidirectional encoder representations from transformers (BERT), Temporal Convolutional Network (TCN) as well as Multi-Head Attention, together with Bidirectional Long Short-Term Memory (BI-LSTM) Multi-Head Attention and BI-LSTM and LSTM models. The chosen models demonstrated great performance to model temporal patterns and detect feature significance. The observed performance are mainly due to the fact that IoT system traffic patterns are both sequential and diverse, leaving a rich set of temporal patterns for the models to learn. Experimental results showed that BERT maintained the best performance. It achieved 99.94% accuracy rate alongside high precision and recall, F1-score and AUC-ROC score of 99.99% which demonstrates its capabilities through temporal dependency capture. The Multi-Head Attention offered promising results by providing good detection capabilities with interpretable results. On the other side, the Multi-Head Attention model required significant processing time like BI-LSTM variants. The GraphSAGE model achieved good accuracy while requiring the shortest training time but yielded the lowest accuracy, precision, and F1 score compared to the other models

Saroj Gopali, Faranak Abri, Sima Siami-Namini et al.

There exist several data-driven approaches that enable us model time series data including traditional regression-based modeling approaches (i.e., ARIMA). Recently, deep learning techniques have been introduced and explored in the context of time series analysis and prediction. A major research question to ask is the performance of these many variations of deep learning techniques in predicting time series data. This paper compares two prominent deep learning modeling techniques. The Recurrent Neural Network (RNN)-based Long Short-Term Memory (LSTM) and the convolutional Neural Network (CNN)-based Temporal Convolutional Networks (TCN) are compared and their performance and training time are reported. According to our experimental results, both modeling techniques perform comparably having TCN-based models outperform LSTM slightly. Moreover, the CNN-based TCN model builds a stable model faster than the RNN-based LSTM models.