This module addresses topics of:

Applied AI domains and tasks:
Positioning vision, language and security analytics as applied AI domains.
Task types: classification, detection, retrieval, anomaly detection, sequence labelling.
Clear separation from algorithm-centric content (covered in AI Algorithms & ML and Advanced ML & DL).

Data pipelines and engineering for applied AI:
Domain-specific data characteristics: images vs text vs logs/flows.
Data pre-processing and feature pipelines for each domain.
Handling noisy, incomplete and weakly-labelled data in operational settings.

Applied AI for Computer Vision:
Using pre-trained CNN/vision models and APIs for classification and detection.
Image augmentation and basic robustness considerations (Ex, lighting changes, occlusion).
Edge vs cloud deployment trade-offs (latency, bandwidth, privacy).

Applied AI for Natural Language Processing:
Text representation (embeddings, tokenisation) via modern libraries/APIs.
Applied tasks: sentiment analysis, intent detection, spam/phishing classification, simple NER.
Using hosted NLP/LLM services responsibly (rate limits, data handling, prompt-based workflows), without re-teaching ethics fundamentals.

Applied AI for Secure Systems:
AI in security operations: anomaly detection in logs, IDS/IPS analytics, UEBA-style behaviour profiling (conceptual and applied).
Machine-assisted phishing detection and malware triage.
Integrating AI-based analytics into SIEM/SOC workflows (alerts, dashboards, triage pipelines).
Emphasis on applied use of AI rather than teaching baseline cyber security theory (to avoid overlap with core Cyber modules).

System modelling and architecture for AI-enabled secure systems:
Architectural patterns for AI components: microservices, APIs, batch vs streaming pipelines.
Data flows, access control and secure integration of AI services.
Monitoring, logging and feedback loops; hooking AI components into existing infrastructure.

Evaluation of applied AI systems in operation:
Domain-appropriate metrics and evaluation strategies for vision, language and security tasks.
Behaviour under drift, concept change and distribution shift.
Operational trade-offs: precision vs recall, false positives vs false negatives, alert fatigue.

Robustness and basic adversarial considerations:
High-level overview of robustness issues (Ex, simple adversarial patterns, spoofing, evasion) in vision, language and security analytics.
Defensive strategies aligned with practical deployment (rate limiting, ensemble checks, enforcement via traditional controls).
Positioned at application level to avoid overlap with any specialist adversarial ML or security-theory modules.

Deployment, lifecycle and operational governance (light-touch ethics):
From prototype to pilot: packaging, configuration, basic CI/CD and MLOps concepts.
Role of human operators and escalation paths in secure deployments.
Operationalisation of constraints and principles introduced in Human-Centred AI modules (Ex, logging of decisions, user feedback mechanisms) without re-covering theory.

BCS / TechSkills / Employability elements:
System modelling: AI components are described and reasoned about as part of end-to-end systems (architectural diagrams, data flow models).
Security: Core focus in the “Secure Systems” dimension: integrating AI into security operations and considering attack surfaces introduced by AI.
Evaluation of systems: Operational evaluation and monitoring of AI components in varying conditions, not just offline metrics.
Problem solving, management and planning: Students scope, plan and execute small, applied projects with constraints on performance, latency, and security.
Legal, social, ethical and professional issues: Considered where they affect deployment decisions, logging and escalation, explicitly referencing but not re-teaching the conceptual frameworks from Human-Centred AI modules.

1. PRACTICAL – Individual Demonstration (40%)

Working individually, choose one of three domain tracks:

Vision track (AI students primarily): Ex, image classification, object detection or simple segmentation using pre-trained CNN-based models or vision APIs.

Language track (AI students primarily): Ex, text classification, sentiment analysis, intent detection or basic information extraction using transformer-based APIs or NLP libraries.

Secure-systems track (Cyber students primarily): Ex, anomaly detection in logs, intrusion detection, phishing URL/email detection, or threat-intel text analytics using ML/NLP tooling.

For your chosen track, you will:

Build an end-to-end applied AI prototype (data ingestion, pre-processing, model use, basic post-processing).

Run initial experiments to check operational functionality.

Deliver a short demonstration of the prototype, highlighting how it fits within a wider system or workflow.

2. REPORT - Individual Report (60%)

Building on (but not duplicating) the prototype work, produce an individual report that:

Defines a realistic applied AI scenario in one chosen domain (vision, language or secure systems), including operational context, data sources, constraints and stakeholders.

Designs an AI-enabled component within a larger system: data flows, interfaces, access controls, monitoring and failure modes are explicitly modelled.

Plans and reports a series of evaluation experiments focused on:
- Appropriate domain metrics (Ex, precision/recall for threat detection, IoU for detection tasks, F1 for text classification).
- Robustness tests (Ex, simple stress tests, basic adversarial/edge-case analyses appropriate to Level 6, detection of drift or degradation).
- Operational risks (Ex, false positives vs false negatives, alert fatigue, potential misuse).

Discusses deployment and governance implications for the specific context, including:
- Security considerations (attack surfaces introduced by AI components, logging, access control).
- Legal/professional constraints relevant to the domain (Ex, privacy in logs, content moderation responsibilities), drawing on but not duplicating material from Human-Centred AI modules.

Communicates design choices, trade-offs and recommendations in a style suitable for a mixed technical/non-technical audience (Ex, architecture team plus operational managers).

All teaching sessions will blend conceptual input with practical, scenario-based learning. Students from AI and Cyber Security programmes will work together in shared sessions to reflect the interdisciplinary nature of applied AI in practice. Learning activities may include:

Lectures, to introduce domain-specific applied AI concepts, patterns and case studies.
Practical classes and workshops, lab-based implementation of pipelines for vision, language and security analytics using modern libraries and APIs.
Seminars, discussion of applied case studies, operational incidents and industry reports.
Tutorials / project supervisions, structured support for prototype development and individual reports.
Groupwork and simulation activities, design reviews and architecture discussions emulating real engineering and security-operations meetings.

You will also be provided with curated independent study resources (technical documentation, example notebooks, domain datasets, industry white papers) and low-stakes formative exercises to build confidence in applying tools and interpreting results ahead of summative assessments.

1. Implement end-to-end applied AI pipelines for computer vision, natural language processing or security analytics tasks.

Application & Problem-Solving
Personal Development & Entrepreneurship

2. Evaluate the effectiveness, robustness and operational risks of applied AI solutions in vision, language and secure systems.

Knowledge & Understanding
Critical Reasoning & Collaboration

3. Model AI-enabled components within wider secure systems, including data flows, interfaces, controls and monitoring, ensuring appropriate consideration of security, resilience and system-level behaviour.

Digital Literacy
Research Skills

4. Communicate design choices and deployment recommendations for applied AI systems to technical and non-technical stakeholders.

Communication
Reflection

Students will require access to:

Networked computer labs with Python and contemporary AI libraries (Ex, PyTorch, TensorFlow/Keras, scikit-learn, Hugging Face Transformers, OpenCV) installed.
Tools and environments for data handling and experimentation (Ex, Jupyter, VS Code, containerised environments if available).
Sample datasets for images, text and security logs/flows (open data or institutionally approved case-study datasets).
Version control platforms (Ex, GitHub/GitLab) for managing code and collaboration.
University VLE (Blackboard) for learning materials, announcements and assessment submissions.

Hobson, L. and Dyshel, M. (2025) Natural Language Processing in Action, Manning Publications.

Bird, S., Klein, E. and Loper, E. (2021) Natural Language Processing with Python, 2nd edn, O’Reilly.

Kim, G. and Park, S. (2023) AI for Cybersecurity: Techniques, Tools and Case Studies, Springer.

Burkov, A. (2025) The Hundred-Page Language Models Book: hands-on with PyTorch (The Hundred-Page Books), True Positive Inc.

Hugging Face (ongoing) ‘Transformers Documentation’ [Online] Available at: https://huggingface.co/docs/transformers (accessed 15/02/26)

In this module, you will learn how to apply artificial intelligence to real-world problems in three key domains: computer vision, natural language processing and secure systems. Building on your prior machine learning and cyber security knowledge, you will design and implement end-to-end AI pipelines using contemporary tools and pre-trained models and integrate them into wider secure architectures. Through hands-on labs, case studies and a substantial applied project, you will evaluate how these systems behave in realistic conditions, explore their operational risks, and develop the skills to explain and justify your design and deployment decisions to technical and non-technical stakeholders.