Institute for Systems Studies & Analyses

This research presents a comprehensive and scalable framework for assessing the effectiveness of drone strikes in counter-terrorism operations by leveraging synthetically generated data and advanced hybrid modeling techniques. Recognizing the ethical, legal, and logistical limitations of real-world data acquisition in conflict zones, we simulate a high-fidelity synthetic dataset comprising 30,000 records grounded in the empirical distributions of verified drone strike datasets. Central to our approach is the formulation of a novel composite metric, the Drone Strike Effectiveness Index (DSEI), which holistically integrates Tactical Effectiveness, Strategic Disruption, and Collateral Impact to quantify mission outcomes beyond simplistic killcount measures. Through rigorous experimentation and feature engineering, we evaluate a spectrum of machine learning and deep learning architectures, ultimately demonstrating that a hybrid LSTM encoder coupled with a LightGBM classifier achieves exceptional performance, delivering 98.9% accuracy and F1-score across both cross-validation and unseen test sets. Our framework is robust against overfitting, scalable for large-scale defense simulations, and maintains interpretability, allowing for actionable insights that inform policy-making. Explainability is embedded through Explainable AI (SHAP-based model interpretation), enabling transparent assessment of feature contributions to each prediction. Compared to prior literature, our methodology sets a new benchmark by integrating explainability, synthetic data ethics, and mission-specific modeling for defense analytics.

Constructive simulations are the applications used by the military for the training of their commanders in planning and analysis of various threats and Courses of Action. In the ‘analysis wargames’, there are need to automate many of the tasks of the commander which are carried out by subunit commanders on the ground. Deployment of defence units is one of such important decision making by commander. Deployments of units (and sub units) is dependent on multiple factors which needs to be satisfied/optimised for meeting the given objective of the unit. In this paper we have attempted to solve the multi criterion decision problem of optimal deployment of defence units in mountainous terrain using Particle Swarm Optimization(PSO) and Adaptive Particle Swarm Optimization(APSO). The algorithm has been tested with varied number of decision parameters and their weights using digital elevation and vector data of the terrain features. The auto deployment outcomes are found satisfactory. Our solution approach has potential in automated planning in constructive simulations.
 

Machine Learning models, extensively used for various multimedia applications, are offered to users as a blackbox service on the Cloud on a pay-per-query basis. Such blackbox models are commercially valuable to adversaries, making them vulnerable to extraction attacks to reverse engineer the proprietary model thereby violating the model privacy and Intellectual Property. Here, the adversary first extracts the model architecture or hyperparameters through side channel leakage, followed by stealing the functionality of the target model by training the reconstructed architecture on a synthetic dataset. While the attacks proposed in literature are empirical, there is a need for a theoretical framework to measure the information leaked under such extraction attacks. To this extent, in this work, we propose a novel probabilistic framework, Airavata, to estimate the information leakage in such model extraction attacks. This framework captures the fact that extracting the exact target model is difficult due to experimental uncertainty while inferring model hyperparameters and stochastic nature of training to steal the target model functionality. Specifically, we use Bayesian Networks to capture uncertainty in estimating the target model under various extraction attacks based on the subjective notion of probability. We validate the proposed framework under different adversary assumptions commonly adopted in literature to reason about the attack efficacy. This provides a practical tool to infer actionable details about extracting blackbox models and help identify the best attack combination which maximises the knowledge extracted (or information leaked) from the target model.

Unmanned Aerial vehicles have developed great potential in the military and civil domains as an important platform for several mission objectives. Mission specific payloads on these unmanned platforms have facilitated a wide range of applications such as mapping, surveillance, transportation, target determination and supply of weapons. Automatic Target Recognition is the process of localizing and recognizing high-value targets from low value targets in noisy and complex backgrounds, which employ Content Based Information Retrieval techniques for object detection, identification and classification of ground systems such as aircraft. In this paper, we propose a methodology for identification and classification of ground objects such as aircraft on ground, military tanks and vehicles. The frames captured by the onboard sensors are processed for target recognition using MobileNet-SSD, a Deep Neural Network based algorithm. The change of object locations in the frames are recorded over time and these are modelled using a Finite State Machine. These state inputs form the basis of developing a Common Operational Picture over the area of Surveillance. Operations recorded and monitored by multiple distributed sensors are used to build the Geospatial intelligence over the geographical area of interest. The entire process is integrated using Apache Kafka, a distributed stream processing platform that facilitates asynchronous processing and communication. The proposed methodology is illustrated using several case studies in the paper.

Planning for the deployment of air defence (AD) assets in areas of operation to achieve maximum protection coverage against enemy air threats is an important problem in military science. A three-stage approach to address the problem is proposed: a static methodology to deploy AD resources to maximize the coverage and performance of radar systems under various terrain conditions is studied, followed by the second stage that considers the dynamics of enemy air attacks and Electronic Warfare (EW) conflict that ensues between the attacker and defender modeled using game theory. In the final stage the conflict scenarios modelled using game theory are represented as AD wargames and experimented on a battlefield simulation test-bed called Air Warfare Simulation System (AWSS) to assess the AD effectiveness in operations. The first stage uses a coverage-based optimization, and the second stage is modelled using game theory. The strategies of the attacker (enemy aircraft) and defender (sensors-radars grid and weapons-missiles grid) in the EW operations as Electronic Counter Measures and Electronic Counter-Counter Measures in a game-theoretic setting are illustrated using several scenarios.

The principle of maximum entropy (PME) is a very powerful tool for analysis in a situation, where partial information about the system in terms of simple statistical moments is available. Various useful interpretations of this principle enable researchers to apply it in problems of its own interest. In the present paper, the principle of maximum entropy has been applied to determine the probability of failure of proportions of components of a network (system), when the proportion of these components is subjected to different levels of stress and the failure probability distribution of entire network is known. It is shown that the obtained probability distribution is a continuous variate version of a popular Maxwell–Boltzmann distribution widely used in statistical mechanics. This provides the safe range of the stress to the system for its safe use. Interesting properties of the obtained probability distribution are discussed. The application of the proposed model in other areas as in Network Centric Combat Force (NCCF), social, organizational, urban and defence is also presented.

The paper is a half-way between the agent technology and the mathematical reasoning to model tactical decision making tasks. These models are applied to air defense (AD) domain for command and control (C2). It also addresses the issues related to evaluation of agents. The agents are designed and implemented using the agent-programming paradigm. The agents are deployed in an air combat simulated environment for performing the tasks of C2 like electronic counter counter measures, threat assessment, and weapon allocation. The simulated AD system runs without any human intervention, and represents state-of-the-art model for C2 autonomy. The use of agents as autonomous decision making entities is particularly useful in view of futuristic network centric warfare.

With increasing defence budget constraints and environmental safety concerns on employing live exercises for training, there has been a reinforced focus end considerable efforts on designing military training simulators using modelling, simulation, and analysis for operational analyses and training. Air Warfare Simulation System is an agent-oriented virtual warfare simulator that is designed using these concepts for operational analysis and course of action analysis for training. A critical factor that decides the next course of action and hence the results of the simulation is the skill, experience, situation awareness of the pilot in the aircraft cockpit and the pilots’ decision making ability in the cockpit. Advances in combat aircraft avionics and onboard automation, information from onboard and ground sensors and satellites poses a threat in terms of information and cognitive overload to the pilot, and triggering conditions that makes decision making a difficult task. Several mathematical models of the pilot behaviour, typically based on control theory, have been proposed in literature. In this work, we describe a novel approach based on soft computing and Computational Intelligence paradigms called ANFIS, a neuro-fuzzy hybridization technique, to model the pilot agent and its behaviour characteristics in the warfare simulator. This emerges as an interesting problem as the decisions made are dynamic and depend upon the actions taken by enemy. We also build a pilots’ database that represents the specific cognitive characteristics, skills, training experience, and as factors that affect the pilot’s decision making and study its effect on the results obtained from the warfare simulation. We illustrate the methodology with suitable examples and lessons drawn from the virtual air warfare simulator.

Abstract In this paper we propose a novel multi-agent planning framework that attempts to deal with dynamic nature of cyber physical system while planning and show usefulness of our approach in exploring different strategies and configurations of planning architecture. The proposed approach use multi-agent planners at different layers of abstraction to collaborate by using concepts of task sharing, brokering of task allocation, continuous adaptive planning and blackboard based multi-agent interaction.Keywordscyber-physical systemmulti-agent planningtask networkcontract net protocolblackboard