Problem board
Every problem below is a real gap in the literature with a documented scope, supervisor and deliverable. Pick one and the lab will help you run the study.
28 of 28 problems
Develop an early warning system that detects security anomalies in autonomous vehicle sensor networks before they propagate to decision-making layers.
Research gapExisting IDS focus on network-level threats but ignore sensor-level adversarial attacks specific to autonomous driving pipelines.
Design a federated learning framework that enables cross-hospital model training while provably preserving patient privacy through differential privacy guarantees.
Research gapMost FL frameworks either sacrifice too much utility for privacy or provide insufficient privacy guarantees against gradient inversion attacks.
Build a lightweight SLAM system that runs on resource-constrained drone hardware for GPS-denied indoor navigation.
Research gapExisting visual-inertial SLAM systems require GPU acceleration that lightweight drones cannot provide.
Investigate and develop mixed-precision quantization techniques that maintain model accuracy while enabling inference on microcontroller-class hardware.
Research gapPost-training quantization methods don't generalize well across architectures. Task-aware quantization for multi-task models is underexplored.
Study how LLM-based code assistants can be manipulated to generate vulnerable or malicious code, and develop defenses.
Research gapNo systematic study exists on adversarial attacks targeting code generation models in real-world development workflows.
Implement lightweight zero-knowledge proof protocols for authenticating IoT devices without revealing device identity or credentials.
Research gapStandard ZKP protocols are too heavy for resource-constrained IoT devices. Lightweight alternatives lack formal security proofs.
Build a semantic code search engine that understands intent rather than keywords, enabling developers to find relevant code across millions of lines.
Research gapExisting code search tools are either syntactic (AST-based) or use basic embedding similarity without understanding code semantics.
Develop automated formal verification tools that can prove safety properties of composed smart contracts across DeFi protocols.
Research gapCurrent verification tools analyze individual contracts in isolation, missing cross-contract reentrancy and state inconsistency bugs.
Investigate whether subtle behavioral changes at the operating-system level can identify a measurable pre-attack state before an actual compromise occurs, rather than detecting the attack only after it has begun.
Research gapThe key research challenge is defining and validating a measurable Pre-Attack Behavioral State rather than simply detecting an attack after it begins. Existing intrusion detection leans heavily on known signatures and established indicators of compromise; the transition region between normal operation and active compromise remains comparatively underexplored, and it is unclear which telemetry streams carry the earliest reliable signal.
Investigate an operating-system architecture that continuously learns system behavior, constructs a runtime security model, detects behavioral deviation, and adaptively changes defensive policies without human intervention.
Research gapThe research challenge is whether an OS can build and maintain a trustworthy runtime model of its own behavior across multiple domains and then act on deviations quickly enough to matter. Static policy engines and signature-based tools are mature, but a closed loop where learned behavioral risk directly drives changing defensive policy remains a largely open systems and AI problem.
Investigate whether an autonomous robot can determine when its own perception of reality has become unreliable or potentially compromised by reasoning about contradictions across its sensors and expected physical behavior.
Research gapSensor fusion for robustness has a long history, but explicitly reasoning about contradictions as a clue that perception is compromised — producing a Perception Reliability Score that feeds behavior — remains underexplored, particularly as a defense against spoofed or corrupted sensor streams.
Investigate a cyber-physical security architecture where a secure runtime — Black Wall — prevents compromised software from producing physically dangerous robot behavior, even when an attacker technically controls that software.
Research gapConventional security largely stops at software integrity and data confidentiality. Cyber-Physical Containment — enforcing bounds on the physical effects of software, so that no compromise can translate into unsafe motion or command — sits at a frontier where OS security, robotics middleware, and safety engineering must be co-designed.
Investigate whether cyber-physical systems can predict a dangerous physical consequence several seconds before it occurs, by tracking the earliest reliable cyber signals along the attack timeline.
Research gapHow much measurable warning time exists between the earliest reliable cyber signal and the eventual physical consequence — and whether that window can be formalized as a usable security metric — remains a largely open empirical and modeling question that crosses cybersecurity, robotics, and control theory.
Investigate whether an autonomous system can learn its own operational security baseline from observed normal behavior — and keep updating it — instead of relying on manually written security rules.
Research gapThe critical challenge is learning and adapting without accidentally absorbing attacker behavior into the baseline. Poisoning resistance, continual learning, and autonomous anomaly reasoning around a self-learned model remain frontier problems — and the point where learning and security quality are hardest to reconcile.
Investigate whether the operating system itself can host neural network models that watch, model and defend the machine continuously — a self-defending OS in which learned defence models are a first-class part of the kernel and the security policy.
Research gapWhere a learned defence model should live inside an operating system — user space, kernel space, or split across both — and how to run inference on kernel telemetry at syscall rates without adding unacceptable overhead remains largely open. Equally open is how such a model stays trustworthy when the very kernel it watches can be compromised.
Develop anomaly detection that keeps working when sensor streams drift, recalibrate and change noise characteristics over time, instead of assuming a fixed, stationary baseline.
Research gapMost anomaly detection methods assume a stationary distribution. How to separate genuine concept drift from gradually worsening sensor noise — and update the model accordingly without absorbing anomalies into the baseline — is still an open modelling problem.
Establish a verifiable chain of trust from UEFI firmware through the bootloader to the running kernel, including signature verification, boot measurements and recovery from verification failures.
Research areaOperating System Security
Investigate why Black Wall does not yet boot consistently across QEMU, UEFI-based physical machines and future hardware, and build stage-level diagnostics that separate firmware, bootloader, kernel and display failures.
Research areaOperating System Engineering
Restrict unauthorised memory access and contain kernel defects through stronger kernel/user boundaries, permission enforcement and component isolation.
Research areaKernel Security
Design an operating-system architecture that continuously evaluates security conditions, detects suspicious system changes and applies controlled defensive responses.
Research areaDefensive Security
Evaluate hardware trust, isolate peripherals and reduce the risks introduced by compromised firmware or unauthorised device access.
Research areaHardware and System Security
Isolate robotic workloads, constrain actuator access, authenticate component communication and keep behaviour predictable when software or the network fails.
Research areaRobotics and Embedded Systems Security
Secure communication between distributed Black Wall machines and robotic nodes while limiting the impact of compromised nodes, interrupted connectivity and unauthorised commands.
Research areaNetwork and Distributed Systems Security
Design an update architecture that verifies authenticity, resists unauthorised modification, recovers safely from interrupted installs and restricts unauthorised downgrades.
Research areaSoftware Supply Chain Security
Provide useful security monitoring while minimising the collection, exposure and retention of sensitive system information.
Research areaOperating System Privacy
Isolate applications and services with different trust levels, restrict access to sensitive resources and limit the consequences of a compromised workload.
Research areaSystem Architecture and Security
Detect unauthorised changes to protected files, executables, configuration and security policy, and distinguish them from legitimate updates.
Research areaHost Security and Integrity Monitoring
Define how Black Wall should respond when security services, authentication, storage, networking or hardware fail, including controlled degraded operation and recovery.
Research areaResilient Systems and Cyber-Physical Security
