Whether approving loans, supporting hiring decisions, prioritizing healthcare interventions, or assisting criminal justice systems, machine learning models are now operating in domains where fairness is not optional — it is essential.

Yet a persistent challenge remains:

How do we build models that are both accurate and fair?

For years, this has often been framed as a trade-off: improve fairness and risk losing predictive performance, or maximize accuracy and accept possible bias.

But what if that trade-off can be navigated more intelligently?

Our recent research, “Balancing Fairness and Accuracy Using Grammatical Evolution,” explores precisely that question through an unusual but powerful lens: evolutionary computation combined with causal reasoning.

Why Fairness in AI Is Still Hard

Modern machine learning systems often inherit historical and structural biases embedded in data. Even highly accurate models can generate systematically unequal outcomes for different groups.

This is particularly problematic in high-stakes settings.

Traditional fairness interventions often happen after a model is built:

• Pre-process the data
• Adjust the training objective
• Post-process predictions to satisfy fairness constraints

While valuable, these approaches often treat fairness as a corrective patch.

We wanted to explore a different idea:

What if fairness could be part of the search process itself?

Enter Grammatical Evolution

At the center of our work is Grammatical Evolution (GE) — a form of evolutionary computation that evolves structured solutions using grammar-based representations.

Rather than simply optimizing parameters, GE can evolve causal graph structures, enabling models that are:

• Predictive
• Fairness-aware
• Interpretable

That last property matters enormously.

In many fairness discussions, interpretability is often overlooked. But understanding why a model behaves fairly (or unfairly) is just as important as measuring the fairness outcome.

From Optimization to Fairness-Aware Evolution

Our work introduces three complementary approaches.

GE-α: Constrained Fairness Optimization

The first approach embeds fairness directly as a constraint during evolution.

Instead of evolving models solely for predictive accuracy, we ask:

Can evolution search for models that satisfy fairness criteria while remaining accurate?

Using Equalized Odds Difference as the fairness objective, GE-α explicitly searches for balanced solutions rather than treating fairness as an afterthought.

This turns fairness into part of the fitness landscape.

And that changes everything.

GE-β: A Minimax Perspective on Fairness

The second approach pushes further.

Rather than optimizing average behavior, GE-β uses a minimax formulation that focuses on worst-case trade-offs between error and unfairness.

This is important because responsible AI should not only perform well on average — it should be robust under difficult conditions.

The minimax perspective introduces a stronger notion of balance:

Not just “good enough” fairness.

But fairness that holds under pressure.

GE-γ: Causal Interventions for Explainable Fairness

This is perhaps the most exciting part of the work.

With GE-γ, we integrate causal reasoning through Average Causal Effect (ACE) and intervention-based node pruning.

The goal is not only to improve fairness metrics but to simplify and clarify the learned causal structures themselves.

Low-impact nodes can be removed.

Causal roles like confounders, mediators, and colliders are preserved.

The resulting models become more interpretable without sacrificing performance.

This opens an intriguing possibility:

Using evolution not only to optimize models, but to evolve explanations.

That is a very different vision of machine learning.

Why This Matters

What excites me most about this work is not just the fairness results.

It is the broader idea that evolutionary computation may be an underexplored engine for Responsible AI.

We often associate evolutionary algorithms with optimization benchmarks.

But they may also help us tackle problems involving:

• Fairness
• Transparency
• Robustness
• Causal structure discovery
• Multi-objective ethical trade-offs

That is a much richer role.

And perhaps a much more important one.

Beyond Accuracy-Centric AI

There is a quiet shift happening in AI research.

We are moving beyond asking:

“How accurate is the model?”

Toward asking:

• Is it fair?
• Is it explainable?
• Is it trustworthy?
• Can we intervene when harms emerge?

Those questions require new computational ideas.

We believe evolutionary methods have something valuable to contribute.

A Broader Vision

This work also points toward something larger:

A future where machine learning systems are not merely optimized for performance, but evolved under principles of responsibility.

Imagine search processes where objectives include:

• Accuracy
• Fairness
• Interpretability
• Causal validity
• Societal constraints

That begins to look less like conventional model training—

and more like engineering trustworthy intelligence.

Final Thoughts

Fairness and accuracy do not have to be opposing goals.

Sometimes the real opportunity lies in changing how we search for solutions altogether.

That is what makes evolutionary approaches so compelling.

They do not simply tune models.

They explore possibility spaces.

And in those spaces, we may discover better ways to build AI systems worthy of trust.

Our work on Balancing Fairness and Accuracy Using Grammatical Evolution is one small step in that direction.

I believe there is much more ahead.

If you work in fairness, causal machine learning, evolutionary computation, or Responsible AI, I’d love to hear your thoughts.

How should we evolve the next generation of trustworthy AI?

—
#ArtificialIntelligence #ResponsibleAI #FairnessInML #EvolutionaryComputation #CausalInference #ExplainableAI #MachineLearning #Research

The future of autonomous aerial systems is not arriving suddenly—it is being carefully engineered, tested, and refined in simulation environments that mirror the complexity of the real world.

Our latest IEEE research explores this future through the development of a distributed reinforcement learning testbed for UAV target tracking, where multiple autonomous drones learn to coordinate in real time to follow a dynamic airborne target.

At its core, this work investigates a simple but powerful question:

How can UAV swarms learn to track moving targets more efficiently in unpredictable environments?

The answer lies in combining realistic flight simulation, distributed networking, and modern reinforcement learning algorithms.

Why UAV Swarm Target Tracking Matters

Target tracking is one of the most important capabilities in autonomous drone systems.

Whether the mission involves:

• search and rescue
• disaster monitoring
• perimeter surveillance
• defense simulation
• intelligent logistics
• environmental observation

…the ability for multiple UAVs to collaboratively maintain awareness of a moving target is essential.

Traditional rule-based control methods often struggle when the target behaves unpredictably or when the environment becomes dynamic.

This is where reinforcement learning (RL) becomes transformative.

Instead of following hard-coded instructions, UAVs learn through interaction with the environment, continuously improving their decision-making policies based on experience.

A Realistic Testbed Built on FlightGear and JSBSim

To study this problem, we developed a distributed UAV simulation testbed using:

• FlightGear for high-fidelity 3D flight simulation
• JSBSim for realistic flight dynamics modeling
• UDP-based distributed communication
• real-time reinforcement learning control loops

The architecture allows multiple UAVs to operate as independent learning agents while exchanging state information such as:

• positional coordinates
• orientation
• velocity
• control signals

This creates a highly scalable framework for testing swarm intelligence strategies under near-realistic conditions.

In our experimental setup:

• one UAV acts as the autonomous target
• multiple UAVs act as tracking agents
• distributed reinforcement learning coordinates the swarm in real time

Comparing Modern Reinforcement Learning Models

The study compares three influential RL methods:

• A2C (Advantage Actor-Critic)
• A3C (Asynchronous Advantage Actor-Critic)
• PPO (Proximal Policy Optimization)

Each algorithm contributes different strengths.

A2C for the Target UAV

A2C was used to control the target UAV, generating complex motion patterns that make the tracking task challenging and realistic.

A3C for Distributed Swarm Coordination

A3C enables multiple worker agents to learn asynchronously, making it highly suitable for swarm UAV coordination where multiple trackers operate in parallel.

PPO for Stable Policy Learning

PPO was used to provide robust and stable policy optimization, particularly useful in dynamic environments where abrupt policy updates can destabilize learning.

The Role of Intelligent Exploration

One of the biggest challenges in reinforcement learning is the sparse reward problem.

In target tracking, useful feedback may not arrive frequently enough for agents to learn efficiently.

This means UAVs may spend too much time exploring ineffective strategies before discovering successful behaviours.

To address this, our work integrates an Intrinsic Curiosity Module (ICM), which generates internal rewards whenever the agent encounters novel or difficult-to-predict states.

This mechanism encourages:

• better exploration
• faster discovery of useful strategies
• improved adaptation to unfamiliar target behaviour
• more efficient learning in dynamic environments

Rather than waiting for explicit environmental rewards, the swarm develops an internal motivation to learn.

This significantly improves learning speed and robustness.

What the Results Showed

The results were highly encouraging.

Across multiple simulation runs, the UAV swarm agents enhanced with curiosity-driven exploration demonstrated:

• faster convergence
• higher cumulative rewards
• smoother actor-critic losses
• stronger policy stability
• improved entropy-driven exploration
• better generalisation to dynamic target motion

Among all tested models, A3C integrated with curiosity mechanisms showed the strongest overall performance, delivering the most stable and effective swarm target tracking.

This is particularly significant because asynchronous distributed learning closely mirrors how real swarm systems may operate across multiple compute nodes or edge devices.

Why This Matters Beyond Simulation

The importance of this research extends far beyond virtual flight environments.

The same principles can directly influence real-world systems in:

• disaster response drones
• persistent surveillance
• maritime monitoring
• intelligent border systems
• military training simulation
• autonomous delivery fleets
• environmental hazard assessment

The ability of UAVs to learn collaboratively, adapt to novelty, and coordinate under uncertainty is central to the next generation of autonomous aerospace systems.

Simulation-first research provides a safe, cost-effective pathway to develop these capabilities before real deployment.

Looking Ahead

This work represents an important step toward truly intelligent UAV swarms.

As reinforcement learning continues to mature, the combination of:

• distributed simulation
• curiosity-driven exploration
• asynchronous swarm learning
• realistic flight dynamics
• scalable communication architectures

…will become increasingly important for building resilient autonomous systems.

The sky is no longer the limit.

It is the next intelligent frontier.

The post Building Smarter UAV Swarms: How Reinforcement Learning is Transforming Autonomous Target Tracking first appeared on Psyops Prime.]]> https://psyopsprime.com/ideas/building-smarter-uav-swarms-how-reinforcement-learning-is-transforming-autonomous-target-tracking/feed/ 0 https://creativecommons.org/licenses/by-nc-nd/4.0/ 2730 https://psyopsprime.com/ideas/teaching-machines-to-be-curious-a-step-toward-intelligent-uav-swarms/?utm_source=rss&utm_medium=rss&utm_campaign=teaching-machines-to-be-curious-a-step-toward-intelligent-uav-swarms https://psyopsprime.com/ideas/teaching-machines-to-be-curious-a-step-toward-intelligent-uav-swarms/#respond Wed, 25 Mar 2026 17:29:08 +0000 https://psyopsprime.com/?p=2723 Autonomous systems are often described as the future—but in many ways, they are still struggling with a very human problem: learning from delayed consequences. In

Autonomous systems are often described as the future—but in many ways, they are still struggling with a very human problem: learning from delayed consequences.

In reinforcement learning, this challenge is known as the delayed reward problem. An agent performs a sequence of actions, but the reward—or feedback—arrives much later. By then, it becomes difficult to determine which action actually led to success or failure. For systems operating in complex, dynamic environments—like unmanned aerial vehicles (UAVs)—this problem becomes even more pronounced.

In this post, I want to share insights from a research project focused on addressing this challenge in the context of multi-UAV systems, and how introducing a concept as simple—and as powerful—as curiosity can significantly improve learning.

---

The Problem with Learning Too Late

Imagine trying to learn how to fly a drone, but you only receive feedback minutes after making a mistake. You wouldn’t know what exactly went wrong. Reinforcement learning agents face a similar issue.

In UAV tracking tasks, for example:

• A drone may take dozens of actions before receiving a reward
• The learning signal becomes weak and noisy
• Training becomes unstable and slow

This is particularly problematic in real-time systems, where decisions must be made continuously and reliably.

---

Building a Realistic UAV Testbed

To study this problem, we developed a multi-UAV testbed that combines:

• A high-fidelity flight simulator (FlightGear)
• A Flight Dynamics Model (JSBSim)
• A real-time communication layer using UDP
• Reinforcement learning models integrated directly into the control loop

This setup allows UAVs to:

• Interact with a realistic environment
• Learn from continuous feedback
• Be evaluated under dynamic flight conditions

The goal was not just to simulate intelligence—but to create a platform where intelligent behavior can emerge.

---

A Hybrid Learning Approach

One of the key design decisions was to use different reinforcement learning strategies for different roles:

• The target UAV is controlled using Advantage Actor-Critic (A2C)
  → This ensures stable and predictable flight behavior
• The tracking UAV is controlled using Asynchronous Advantage Actor-Critic (A3C)
  → This enables parallel exploration and faster learning

This separation is important. In multi-agent systems, if all agents behave unpredictably, the environment becomes chaotic. By keeping one agent stable and allowing the other to explore, we create a balanced learning ecosystem.

---

Introducing Curiosity into Machines

The real breakthrough comes from integrating an Intrinsic Curiosity Module (ICM) into the learning process.

Instead of relying only on external rewards (e.g., “you successfully tracked the target”), the UAV also receives intrinsic rewards based on how surprised it is by new experiences.

In simple terms:

• If the UAV encounters something unexpected → it gets rewarded
• If it explores new states → it gets encouraged
• If it keeps doing the same thing → rewards diminish

This transforms learning in a fundamental way.

---

From Sparse Rewards to Continuous Learning

By combining external and intrinsic rewards, we effectively turn:

Sparse, delayed feedback
into
Continuous, meaningful learning signals

This allows the UAV to:

• Keep learning even when external rewards are absent
• Explore more effectively
• Adapt to changing environments in real time

Curiosity acts as a bridge over the gap created by delayed rewards.

---

What We Observed

The results were both encouraging and insightful:

• Traditional methods showed initial learning followed by instability
• The curiosity-driven approach demonstrated:
  • Smoother learning curves
  • Better exploration
  • More reliable tracking behavior

In practical terms, the tracking UAV was able to:

• Maintain pursuit more effectively
• Adapt to variations in the target’s movement
• Continue learning even in uncertain conditions

---

Why This Matters

Most UAV research focuses on:

• Flight control
• Navigation
• Multi-agent coordination

But relatively little attention is given to how these systems actually learn over time, especially under imperfect conditions.

This work highlights an important shift:

Instead of designing systems that rely solely on external feedback, we can build systems that motivate themselves to learn.

This idea has implications far beyond UAVs:

• Autonomous vehicles
• Robotics
• Smart surveillance systems
• Distributed AI systems

---

Looking Ahead

There is still much to explore.

Future directions include:

• Expanding to multi-UAV swarm coordination
• Incorporating vision-based perception
• Exploring advanced algorithms like Proximal Policy Optimization (PPO)
• Moving toward real-world deployment and digital twins

Each of these steps brings us closer to systems that are not just automated—but truly autonomous.

---

Final Thoughts

Curiosity is often seen as a uniquely human trait—the drive to explore, to learn, to understand the unknown.

But what happens when machines begin to exhibit the same behavior?

This research suggests that by embedding curiosity into artificial systems, we can overcome some of the most persistent challenges in learning—transforming hesitation into exploration, and delay into discovery.

And perhaps, in doing so, we move one step closer to building machines that don’t just follow instructions—but learn how to think for themselves.

The post Teaching Machines to Be Curious: A Step Toward Intelligent UAV Swarms first appeared on Psyops Prime.]]>

Artificial Intelligence systems are no longer confined to research labs. They influence decisions about loans, employment, healthcare, and criminal justice—domains where fairness is not optional. Yet, much of modern machine learning still treats fairness as a secondary concern: something to be fixed after a model has already learned its patterns.

One of the most persistent assumptions in this space is that fairness and accuracy are inherently at odds. Improve one, and the other must suffer. But is this trade-off inevitable—or is it simply a limitation of how we frame the problem?

In our recent work, A Multi-Objective Approach to Balance Fairness and Accuracy, we argue for a different perspective: fairness should be treated not as a constraint or post-processing correction, but as a first-class optimisation objective, explored alongside accuracy rather than subordinated to it. I would like to congratulate my co-authors about this who are: 1. the doctoral candidate Zahid Irfan, Dr. Roisin Loughran, and Professor Fergal Mc Caffery. Basically this is the work was done by Zahid, who is a colleague as well as a very close friend of mine.

Why Bias Persists in Machine Learning

Bias in AI systems often reflects deeper structural issues: biased data collection, historical inequalities, and spurious correlations that models eagerly exploit. When these correlations involve protected attributes—such as sex, age, or race—the resulting systems may achieve impressive accuracy while still producing unfair outcomes.

Traditional bias-mitigation approaches typically fall into three categories:

• Pre-processing, where the data is modified before training
• In-processing, where fairness is incorporated into the learning algorithm
• Post-processing, where predictions are adjusted after training

While all three have their place, many approaches operate largely as black-box fixes. They may improve a fairness metric, but often at the cost of interpretability and deeper understanding.

This is where causal modelling becomes essential.

Bringing Causality into the Picture

Correlation alone cannot tell us why a model behaves unfairly. Causal models, on the other hand, explicitly represent cause–effect relationships between variables.

We use Causal Bayesian Networks (CBNs)—directed acyclic graphs where nodes represent variables and edges encode causal influence. These structures allow us to reason about dependencies, confounders, and interventions, rather than relying solely on statistical association.

However, learning causal structures from data is a notoriously difficult problem. The search space of possible graphs grows exponentially, making exhaustive search infeasible.

To address this, we turned to Evolutionary Computation.

Evolving Causal Graphs with Grammatical Evolution

Our approach uses Grammatical Evolution (GE) to automatically generate and evolve causal graph structures. A context-free grammar constrains the search space to valid causal graphs, while still allowing a rich variety of structures to emerge.

Each individual in the evolutionary population represents a candidate causal graph. From this graph, we build a CBN, train it on data, and evaluate its performance.

Crucially, we do not evaluate performance using a single objective.

Fairness and Accuracy as Joint Objectives

Instead of collapsing everything into one score, we adopt a multi-objective optimisation framework using NSGA-II, a well-established evolutionary algorithm.

We optimise two objectives simultaneously:

1. Accuracy, measuring predictive performance
2. Fairness, measured using Equal Opportunity Difference (EOD), which captures disparities in true positive rates between protected groups

This produces not a single “best” model, but a Pareto front—a set of non-dominated solutions representing different fairness–accuracy trade-offs.

This is a powerful shift in mindset. Rather than asking “What is the best model?”, we ask:
“Which trade-off best fits the ethical and operational requirements of this domain?”

What We Observed

Using the German Credit dataset as a case study, our experiments showed that:

• It is possible to achieve very low fairness disparity while maintaining competitive accuracy
• Multiple causal graphs can yield similar performance, offering flexibility and interpretability
• The evolved graphs are non-trivial, capturing meaningful dependencies among features
• Practitioners can choose models that slightly sacrifice accuracy for substantial gains in fairness—or vice versa

Importantly, the causal graphs themselves provide insight. They allow us to inspect how features influence outcomes, opening the door to causal reasoning, domain validation, and future intervention analysis.

Why This Matters

Fair AI is not just about metrics—it’s about understanding.

By combining causality with multi-objective evolutionary optimisation, this work demonstrates that:

• Fairness does not have to be an afterthought
• Accuracy does not have to be blindly maximised
• Interpretability and performance can coexist

Most importantly, it reframes fairness as an optimisation problem, not a moral constraint imposed from outside the model.

Looking Ahead

Future directions include:

• Exploring additional fairness metrics to capture different notions of equity
• Extending experiments to larger and more diverse datasets
• Incorporating causal interventions and counterfactual analysis
• Further strengthening the link between ethical requirements and model design

As AI systems continue to shape society, approaches that integrate ethics, causality, and optimisation will be essential—not optional.

Fairness is not something we bolt onto AI.
It is something we design for

I am delighted to share a significant milestone in my research journey: the acceptance of our latest paper, “A Multi-Objective Scheme for Collision Avoidance, Swarm Cohesion, and Target Tracking for Smart UAVs,” for publication in the Journal of Information and Intelligence.

This work represents several years of development, collaboration, reflection, refinement — and most importantly, a deep fascination with how artificial intelligence can push intelligent aerial systems into entirely new territory.

In this blog post, I want to take the opportunity to describe not just the technical details, but the intellectual narrative behind the research, the people and organisations who made it possible, and how this work fits into a much larger continuum of ideas.

Why UAV Swarm Intelligence Matters

Unmanned Aerial Vehicles are no longer just flying sensors or remote-controlled devices. Increasingly, they are becoming autonomously intelligent systems capable of:

• sensing
• decision-making
• coordination
• adaptation
• collective behaviour

When multiple UAVs work together cooperatively, they can accomplish feats that a single drone never could:

• searching complex environments efficiently
• forming dynamic formations
• collectively tracking moving targets
• supporting search-and-rescue missions
• surveying hazardous or inaccessible regions

But making such behaviours stable, safe, and reliable is enormously challenging — especially when seven UAVs are learning simultaneously, as in our study.

Swarm intelligence is delicate. If drones fly too close, they risk collision. If they spread too far apart, the swarm loses coherence. If they track the target too aggressively, they destabilise; if too passively, they fall behind.

Our goal was to build a learning-based testbed in which UAVs discover behaviours that naturally balance all three objectives:

1. Collision avoidance
2. Swarm cohesion
3. Target tracking

This required innovation across simulation engineering, artificial intelligence, control theory, and mathematical modelling.

Reinforcement Learning at the Core

The heart of our system is Reinforcement Learning (RL) — a type of AI inspired by how organisms learn through trial and error. Instead of being explicitly programmed, UAVs:

• observe their environment
• choose actions
• receive rewards or penalties
• update their behaviour
• gradually become more skilled

We designed a dual-model structure:

A2C

Controls the target UAV, which performs random but physically realistic manoeuvres.

A3C

Controls seven tracking UAVs, each governed by a separate asynchronous worker, enabling parallel learning and higher exploration diversity.

To make learning more effective, we included an Intrinsic Curiosity Module (ICM), which allows drones to reward themselves for exploring unfamiliar states. This is essential in environments where external rewards are sparse or delayed — a frequent challenge in multi-agent flight scenarios.

The Ellipsoid: A New Way to Think About Space and Safety

One of the key innovations in this research is our use of 3D ellipsoids to define “safety spaces” around each UAV.

A simple sphere could work, but real aircraft dynamics aren’t symmetric:

• they extend more along particular axes
• orientation matters
• distance alone is not enough

By using ellipsoids aligned with each UAV’s orientation, we created a geometrically meaningful safety envelope. This allowed us to mathematically express:

• how close two UAVs are
• whether that distance is safe
• whether they are aligned with each other
• how far they should remain from the target for optimal tracking

To build intelligence around this, we wrapped a Gaussian reward function around the ellipsoidal boundary.
This means:

• maximum reward = exactly on the boundary
• penalties = too close or too far
• smooth gradient = stable learning

This mathematical framework is one of the strongest contributions of the paper — and integral to the elegant behaviour shown in the trajectories.

Real-Time Simulation with FlightGear and JSBSim

Our testbed is fully integrated with:

• FlightGear for 3D simulation
• JSBSim for realistic flight dynamics
• UDP networking for high-speed communication

All seven UAVs plus the target operate simultaneously in real time. This is not a simplified physics environment — it is grounded in real flight dynamics, giving the results credibility and transfer potential.

Intellectual Roots: The NUAV Testbed and the Namal Education Foundation

Every research project stands on the contributions of earlier work.
In our case, one of the most important inspirations was the NUAV Testbed, whose development was originally funded by the Namal Education Foundation.

The NUAV Testbed was one of the early attempts to create:

• an accessible UAV simulation environment
• a modular architecture
• a cost-effective flight testing system
• infrastructure for experimentation in autonomy

Its philosophy of openness, affordability, and rigorous experimentation helped inspire key architectural decisions in our current system. While our work moves significantly beyond the original design — adding multi-agent RL, curiosity-driven learning, and ellipsoidal safety geometry — the intellectual DNA of NUAV remains present.

It is important to recognise this evolution. Research is a continuum, and we are proud to build upon a foundation that was shaped years earlier through the support of the Namal Education Foundation.

A Special Acknowledgment: Dr. Junaid Akhtar

A project of this scale requires not only technical effort but also the conceptual clarity needed to lay out a compelling research proposal.
For that, I want to express my deep gratitude to Dr. Junaid Akhtar.

Dr. Akhtar holds a PhD in non-Darwinian schemes for evolutionary computation — a highly specialised and intellectually demanding field. His expertise in alternative evolutionary paradigms, theoretical modelling, and computational intelligence is remarkable.

During the proposal development stage, his insights:

• sharpened the conceptual direction,
• strengthened the problem formulation,
• deepened the evolutionary computation perspective,
• and helped shape a proposal that was both technically ambitious and academically solid.

His support was instrumental, and I am grateful for his contributions.
It is a privilege to receive guidance from a scientist of his calibre.

Celebrating Collaboration

No research endeavour is done alone. I am fortunate to have worked with:

• Jawad Mahmood
• Dr. John Loane
• Professor Fergal McCaffery

Their expertise, commitment, and collaborative energy powered every stage of this project — from initial conceptualisation to simulation to manuscript preparation.

I am honoured to share authorship with them.

Funding That Made This Possible

This research was funded by the
Technological University Transformation Fund (TUTF)
of the
Higher Education Authority (HEA) of Ireland.

Their support for innovative, forward-looking research in AI and autonomy has created a thriving environment for ambitious projects such as this one. We are sincerely grateful for this backing.

Looking Toward the Future

The development of this testbed opens exciting new possibilities:

• deploying UAV swarms in real-world experiments
• integrating explainable AI for safer autonomous behaviour
• studying adversarial or cooperative swarm strategies
• expanding multi-objective learning frameworks
• applying swarm AI to environmental monitoring and disaster response

This is only the beginning.

The future of intelligent UAV swarms — dynamic, adaptive, curiosity-driven, and cooperative — holds immense promise. I am excited to continue pushing the boundaries of what is possible.

Thank you for reading, and thank you to everyone who supported this journey.
If you have questions, ideas, or interest in collaboration, I would be delighted to connect.

The post Advancing Intelligent UAV Swarms — A Journey of Research, Collaboration, and Discovery first appeared on Psyops Prime.]]> https://psyopsprime.com/ideas/advancing-intelligent-uav-swarms-a-journey-of-research-collaboration-and-discovery/feed/ 0 https://creativecommons.org/licenses/by-nc-nd/4.0/ 2626 https://psyopsprime.com/machine-learning/balancing-accuracy-and-fairness-in-ai-a-multi-objective-approach-with-causal-bayesian-networks/?utm_source=rss&utm_medium=rss&utm_campaign=balancing-accuracy-and-fairness-in-ai-a-multi-objective-approach-with-causal-bayesian-networks https://psyopsprime.com/machine-learning/balancing-accuracy-and-fairness-in-ai-a-multi-objective-approach-with-causal-bayesian-networks/#respond Tue, 12 Aug 2025 10:58:37 +0000 https://psyopsprime.com/?p=2570 Artificial Intelligence is changing the way we make decisions — from who gets a loan, to who gets a job interview, to who is flagged

Artificial Intelligence is changing the way we make decisions — from who gets a loan, to who gets a job interview, to who is flagged for additional security checks. But there’s a challenge that goes beyond predictive performance: bias.

Bias in AI isn’t always intentional. Sometimes it’s the byproduct of skewed datasets, hidden correlations, or unexamined assumptions in algorithms. The result? Models that are accurate on paper but systematically disadvantage certain groups. In sensitive domains like finance, healthcare, or justice, this isn’t just a bug — it’s a social risk.

The Research Question

Our team at Dundalk Institute of Technology — Zahid Irfan, Róisín Loughran, Muhammad Adil Raja, and Fergal McCaffery — asked a simple but important question:

Can we design AI systems that are both accurate and fair — without sacrificing too much of either?

Our Approach: Cause Meets Evolution

To tackle this, we brought together two powerful ideas:

1. Causal Bayesian Networks (CBNs)
   These are graphical models that capture cause-and-effect relationships, not just correlations. By structuring knowledge in a causal way, we can better understand why predictions are made and detect pathways that lead to unfair outcomes.
2. Grammatical Evolution (GE)
   An evolutionary algorithm that “evolves” solutions over generations, guided by a grammar that ensures valid models. Think of it as natural selection for algorithms — only the fittest survive.

We combined these in a multi-objective optimisation setting, using NSGA-II, to evolve CBNs that balanced two fitness measures:

• Accuracy — the percentage of correct predictions.
• Fairness — measured by Equal Opportunity Difference (EOD), which compares true positive rates across protected groups (in our case, male vs. female applicants).

Testing Ground: The German Credit Dataset

We tested our approach on the German Credit dataset, which contains credit application data with a notable gender imbalance (70% male, 30% female).

We ran two sets of experiments:

• Single-objective: optimise for fairness alone or accuracy alone.
• Multi-objective: optimise both together.

The Results

The results were striking:

• Single-objective fairness: high fairness, low accuracy.
• Single-objective accuracy: high accuracy, poor fairness.
• Multi-objective optimisation:
  • Fairness improved by 32% compared to accuracy-only optimisation.
  • Accuracy dropped by just 2.85% — a small price for a large fairness gain.

In other words, it is possible to have AI that is both fairer and still highly accurate.

Why This Matters

This work shows that fairness in AI isn’t just a theoretical ideal — it’s an achievable design goal. By combining causal reasoning with evolutionary search, we can navigate the trade-offs between accuracy and fairness more intelligently.

For industries deploying AI in sensitive decision-making, this means:

• More equitable outcomes.
• Greater transparency in why decisions are made.
• Less risk of unintentionally embedding social biases into automated systems.

Looking Ahead

We’ll be presenting this research at GECCO 2025 in Malaga, Spain. As AI becomes more embedded in critical infrastructure, balancing accuracy with fairness will be essential for maintaining public trust.

If we want AI to truly serve everyone, we need to design it with fairness as a first-class objective — not an afterthought.

---

Keywords: Fair AI, Causal Models, Bayesian Networks, Evolutionary Computation, Multi-objective Optimisation, Machine Learning Ethics, GECCO 2025.

---

 

The post Balancing Accuracy and Fairness in AI: A Multi-Objective Approach with Causal Bayesian Networks first appeared on Psyops Prime.]]>

Hello, and welcome to my blog! Today, I want to talk about something truly thrilling and transformative that I’ve been a part of: a groundbreaking new approach to Unmanned Aerial Vehicles (UAVs) that promises to be a game-changer for the future of autonomous flight.

We’re all familiar with drones, but imagine a future where these devices aren’t just remote-controlled tools—they’re intelligent, adaptive, and highly coordinated partners capable of learning on their own. This is the vision driving some cutting-edge research that addresses a fundamental challenge in artificial intelligence: the “delayed reward problem” in Reinforcement Learning (RL).

What is the “Delayed Reward Problem”?

In simple terms, RL works by teaching an AI agent to perform a task by giving it rewards. If a drone needs to track a moving target, it should get a reward for staying close. But what happens if the reward is only given after a long period, or is sparse and infrequent? The agent struggles to learn what it did right, and its training becomes inefficient. This has been a major hurdle for developing truly autonomous UAVs, especially when they need to operate in dynamic, real-time environments.

A Novel Solution: The Intrinsic Curiosity Module

This new research introduces a truly novel solution by integrating an Intrinsic Curiosity Module (ICM) with the powerful Asynchronous Advantage Actor-Critic (A3C) algorithm. This isn’t just about giving the drones external rewards; the ICM gives them an internal sense of curiosity. It encourages them to explore their environment and learn new behaviors even when an external reward isn’t immediately available. This makes the learning process much more robust and efficient.

To make it even smarter, a Self-Reflective Curiosity-Weighted (SRCW) hyperparameter tuning mechanism was developed. This ingenious system allows the agents to adjust their own learning parameters in real-time based on their performance. Think of it as a swarm of drones that can learn how to learn better, all on their own. The result? Unprecedented efficiency in training and a dramatic improvement in the agents’ ability to adapt to complex and evasive scenarios.

From Simulation to Reality

This technology was developed and tested within a high-fidelity simulation environment that interfaces with the FlightGear flight simulator and the JSBSim Flight Dynamics Model (FDM). This allows for a realistic and scalable testbed where multiple UAVs can operate and learn simultaneously. This work builds upon the foundational NUAV testbed, which was originally funded by the Namal Education Foundation, showcasing a fantastic evolution of capabilities.

This research was passionately supported by the Technological University Transformation Fund (TUTF) of the Higher Education Authority (HEA) of Ireland, a testament to the country’s commitment to pushing the boundaries of innovation in technology.

Game-Changing Applications for the Future

So, what does this mean for the future of aerial navigation? The implications are truly immense and span multiple domains:

• Search and Rescue: Swarms of autonomous UAVs could rapidly and efficiently search vast, complex terrains for missing persons, adapting their search patterns in real-time without constant human input.
• Precision Agriculture: Drones could dynamically monitor crop health and autonomously target specific areas for watering or pest control, leading to more sustainable and efficient farming practices.
• Infrastructure Inspection: Imagine a fleet of drones inspecting bridges, power lines, or pipelines, not just flying along a pre-programmed path but intelligently adapting to find and assess potential issues faster and more safely than ever before.
• Environmental Monitoring: From tracking endangered wildlife to monitoring air quality or assessing the damage after a natural disaster, these intelligent swarms could collect critical data with greater agility and resilience.
• Dynamic Delivery Systems: In the future, fleets of delivery drones could navigate complex urban environments, reacting to unforeseen obstacles and optimizing routes on the fly, fundamentally transforming logistics.

This work marks a significant step towards a future where autonomous aerial systems are not just tools, but truly intelligent, adaptive partners in a multitude of critical domains. If you find it interesting, you can read our complete research article that was published recently on Elsevier’s Machine Learning With Applications. It’s an exciting time to be involved in this field, and I can’t wait to see what comes next!

Revolutionizing the Operating Room: How AI is Reshaping Surgical Tool Detection

The future of surgery is evolving at an unprecedented pace, driven by groundbreaking advancements in artificial intelligence. A recent paper, “A Review of Performance of Recent YOLO Models on Cholecystectomy Tool Detection,” by Muhammad Adil Raja, Róisín Loughran, and Fergal Mc Caffery from the Regulated Software Research Center (RSRC) at Dundalk Institute of Technology (DkIT), sheds light on how cutting-edge AI models are set to enhance precision and safety in the operating room.

The Challenge: Precision in Computer-Aided Laparoscopy

In the complex environment of laparoscopic surgery, particularly during procedures like cholecystectomy, the accurate and real-time identification of surgical instruments is paramount. This capability is crucial for everything from precise surgical navigation and assessing surgeon performance to estimating the complexity of a procedure. Traditional methods can be prone to human error and limitations, highlighting the need for advanced automated solutions.

The AI Solution: Leveraging YOLO Models

The research by Raja, Loughran, and Mc Caffery dives deep into the performance of various state-of-the-art You Only Look Once (YOLO) object detection algorithms. These AI models are renowned for their efficiency and accuracy in identifying objects within images and video streams. The study systematically evaluated recent YOLO variants, including YOLOv7, all versions of YOLOv8, v9, v10, v11, v12, and three variants of YOLO-Neural Architecture Search (NAS).

Rigorous Testing and Key Findings

To ensure comprehensive evaluation, the researchers trained and tested these models using the m2cai16-tool-locations benchmark dataset, which comprises 2,811 frames and over 3,000 annotations across seven distinct classes of surgical instruments (grasper, bipolar, hook, scissors, clipper, irrigator, and specimen bag). The models were trained on high-performance supercomputers, demonstrating the computational power required for such sophisticated AI applications.

The findings from this extensive analysis are particularly insightful:

• Accuracy Leaders: Variants of YOLOv12 generally showed superior performance in terms of overall accuracy, with YOLOv12x specifically excelling in Precision.
• Optimal Detection: YOLOv9t emerged as the top performer for mean Average Precision (mAP50), indicating its robust ability to correctly identify and locate tools.
• Efficiency Champions: For real-time applications where speed is critical, YOLOv11n demonstrated the fastest inference speed, making it highly suitable for integration into live surgical environments. Conversely, while powerful, YOLOv9e was found to be the slowest.
• NAS Performance: Interestingly, the YOLO-NAS variants exhibited lower detection accuracy compared to other YOLO versions in this specific context.

The Impact on Future Surgical Practices

This research makes a significant contribution to both algorithmic development in object detection and the broader field of medical imaging. By providing a thorough comparison of the accuracy and computational efficiency of leading YOLO models, the paper offers invaluable insights for developers and medical professionals looking to integrate AI into surgical workflows.

The implications are vast: from enhancing surgical training simulations and developing more precise robotic surgery systems to providing real-time decision support for surgeons, the advancements in AI-powered tool detection promise a future of safer, more efficient, and ultimately, more successful surgical outcomes.

This pioneering work underscores the exciting potential of AI to transform healthcare, bringing us closer to a new era of intelligent, computer-aided medicine.

In the rapidly advancing field of healthcare, technology plays a pivotal role in enhancing surgical practices. One of the most exciting developments in recent years is the integration of advanced object detection algorithms in Computer Aided Laparoscopy (CAL). Our latest research, titled “Cholecystectomy Surgical Instrument Detection Using Variants of YOLOv8,” explores how these innovations can significantly improve surgical outcomes and redefine the operating room experience.

The Importance of Object Detection in Surgery

Surgery is an intricate art that demands precision, skill, and the ability to navigate complex anatomical structures. As the demand for surgical procedures continues to rise globally, the need for efficient and accurate surgical techniques has never been more critical. This is where object detection technologies come into play. By enabling real-time localization and tracking of surgical instruments, these technologies empower surgeons to perform with enhanced accuracy and confidence.

Enter YOLOv8: A Game Changer in Object Detection

The You Only Look Once (YOLO) algorithm has long been a leader in the field of object detection, and its latest iteration, YOLOv8, promises even greater advancements. Our research focuses on leveraging all variants of the YOLOv8 model to achieve superior performance in detecting surgical instruments during cholecystectomy procedures.

Key Features of YOLOv8:

• Improved Detection Accuracy: YOLOv8 has been designed to enhance prediction accuracy, allowing for more reliable identification of surgical tools.
• Faster Inference Speed: The algorithm’s efficiency means that surgeons can receive real-time feedback, crucial for maintaining the flow of surgery.
• Robust Performance: Our experiments demonstrate that YOLOv8 can effectively handle the complexities of laparoscopic video feeds, ensuring that instruments are accurately detected even in challenging conditions.

Research Insights and Findings

In our study, we utilized the well-known m2cai16-tool-locations dataset, which comprises 2,811 frames from 10 videos, annotated with 3,141 instances of seven different surgical instruments. By training various YOLOv8 models on this dataset, we achieved remarkable results that not only highlight the algorithm’s capabilities but also set a new benchmark for surgical instrument detection.

Benefits of Our Findings:

• Enhanced Surgical Workflow: The integration of YOLOv8 into CAL systems allows for automated tool recognition, reducing the cognitive load on surgeons and enabling them to focus on the procedure at hand.
• Improved Patient Safety: By minimizing the risk of surgical errors through accurate instrument tracking, we can enhance overall patient safety and outcomes.
• Contribution to the Surgical Community: Our research not only benefits surgeons but also contributes to the ongoing development of the YOLO algorithm, paving the way for future advancements in object detection.

Looking Ahead: The Future of Surgery

As we continue to explore the potential of advanced technologies in healthcare, the implications of our findings are profound. The future of surgery is not just about human skill; it’s about harnessing the power of innovation to create safer, more efficient, and data-driven surgical environments.

In conclusion, the integration of YOLOv8 in cholecystectomy instrument detection represents a significant leap forward in surgical technology. By embracing these advancements, we can redefine the standards of surgical precision and ultimately improve patient care.

Join me on this exciting journey as we continue to explore the intersection of technology and healthcare, and work towards a future where surgical excellence is within reach for all.

In our increasingly digital world, web browsers have become essential tools for daily activities, from communication to online shopping. However, with this convenience comes the risk of cybercrime, making it crucial to understand how to investigate and analyze browser activities effectively. A recent study published in the Journal of Information Security and Applications sheds light on this important topic, focusing on the forensic analysis of popular web browsers—Firefox, Chrome, and Edge—on the latest Windows 11 operating system.

The Importance of Browser Forensics

As cybercriminals develop more sophisticated methods to exploit vulnerabilities, the need for robust forensic analysis of web browsers has never been more critical. Browsers store a wealth of information, including browsing history, cookies, and cached data, which can provide valuable insights into user behavior and potential illicit activities. Understanding how to collect and analyze this data is essential for law enforcement and cybersecurity professionals.

Key Findings from the Study

The research conducted by a team of experts highlights several important aspects of browser forensics:

1. Artifact Collection: The study emphasizes the significance of identifying and collecting artifacts from browser usage. These artifacts can reveal crucial information about user activities, including access to prohibited sites and suspicious communications.

2. Methodology Development: By simulating cyber-criminal activities, the researchers developed a comprehensive methodology for analyzing browser usage. This approach covers all stages of browser activity, from installation to uninstallation, ensuring a thorough examination of potential evidence.

3. Comparative Analysis: The study also compares the artifact generation of different browsers, revealing that Firefox produces fewer artifacts than Chrome and Edge. This finding raises important questions about data recovery and the effectiveness of forensic investigations across different platforms.

Implications for Cybersecurity

The insights gained from this research are invaluable for enhancing cybersecurity measures. As organizations and individuals continue to rely on web browsers for various activities, understanding the implications of online behavior is essential. This study not only provides a foundation for future research in browser forensics but also highlights the need for continuous improvement in forensic methodologies to keep pace with evolving cyber threats.

Conclusion

As we navigate the complexities of the digital landscape, the importance of browser forensics cannot be overstated. The recent study published in the *Journal of Information Security and Applications* serves as a crucial step forward in understanding how to investigate and analyze web browser activities effectively. By staying informed about the latest research and developments in this field, we can better prepare ourselves to combat cybercrime and protect our digital lives.

For those interested in delving deeper into the world of browser forensics, I encourage you to read the full study and consider the implications for your own cybersecurity practices.

The post Unraveling the Mysteries of Browser Forensics: Insights from Recent Research first appeared on Psyops Prime.]]> https://psyopsprime.com/ideas/unraveling-the-mysteries-of-browser-forensics-insights-from-recent-research/feed/ 0 https://creativecommons.org/licenses/by-nc-nd/4.0/ 2476