With the increasing adoption and prevalence of (LLMs), concerns regarding their inherent biases have become paramount. Existing approaches often rely on fixed datasets and metrics for bias detection and subsequent fine-tuning for mitigation. However, relying solely on such fixed datasets and metrics, akin to standardized exams, may be insufficient due to their inherent inflexibility (e.g., fixed testing content) and susceptibility to gaming. Furthermore, these benchmarks risk contamination if inadvertently included in training data. Drawing an analogy to human peer review and collaborative learning, this paper introduces a novel Cross-Model Adjudication Framework (CMAF) for detecting and mitigating biases in LLMs. We implement a distributed peer-review mechanism where four state-of-the-art models (Qwen2.5-7B, DeepSeek-7B-chat, Gemma2-9B, LLaMA3.1-8B) critically evaluate each other's responses to prompts from the HolisticBias dataset. The consensus-derived, low-bias outputs are then utilized for parameter-efficient fine-tuning. Our method achieves a bias reduction of up to 12.3%, measured by the statistical significance of token likelihood differences across demographic groups, and yields another comparable B-score metric against commercial (LLMs), while preserving core task performance and maintaining minimal inference latency overhead post-fine-tuning.

Global Navigation Satellite Systems (GNSS) are vulnerable to spoofing attacks that can compromise critical infrastructure and safety-critical applications. While Federated Learning (FL) offers a promising paradigm for collaborative GNSS spoofing detection, with privacy preservation, existing proposals remain vulnerable to Data Poisoning Attacks (DPAs), by malicious clients manipulating local training labels to degrade global model performance. This paper presents G2FL, a robust FL framework that detects and mitigates label-flipping attacks in GNSS spoofing detection. Our approach integrates two complementary mechanisms: a Gaussian Mixture Model (GMM)-based cluster analysis that distinguishes poisoned model updates from benign contributions, through validation prediction statistics, and a score-based client management system that maintains reliability assessments to exclude persistently malicious participants. Experimental evaluation using a real-world GNSS spoofing dataset demonstrates that G2FL effectively mitigates DPAs, achieving 84.29% Area Under the Curve (AUC) compared to undefended FL (56.42%) and state-of-the-art defense, FoolsGold (65.80%), while maintaining performance close to attack-free scenarios (88.77%).

Modern localization relies on satellite and terrestrial infrastructures (e.g., global navigation satellite system (GNSS) and crowd-sourced Wi-Fi), supporting critical applications ranging from autonomous navigation to life services. However, they are susceptible to multimodal location spoofing attacks, where various types of wireless signals are crafted in a coordinated manner to deceive the receiver position. Such attacks can cause severe disruptions, such as misguiding vehicles and enabling fraud. Although recent deep learning methods help detect GNSS spoofing and jamming, they require extensive labeled datasets, they are not designed for multimodal location spoofing, training is computationally intensive and centralized, thus raising privacy concerns due to the sensitive nature of location data. We propose FedRAIM, a novel federated learning framework designed for robust, efficient, and privacy-preserving multimodal location spoofing detection. The key FedRAIM elements are: training labels are autonomously generated through an extended receiver autonomous integrity monitoring (RAIM) algorithm, incorporating diverse opportunistic information, i.e., terrestrial infrastructure signals, onboard sensor data, and naturally GNSS signals. During federated training, local long short-term memory (LSTM) models are trained independently on client devices. Server then aggregates, using the FedAvg algorithm, these local model parameters, not data, into a global model, which is subsequently distributed back to the clients for iterative training. Experimental results on real-world multimodal spoofing datasets demonstrate that FedRAIM achieves an area under the curve (AUC) of up to 98.3%.

With the rise of applications that rely on terrestrial and satellite infrastructures (e.g., and crowd-sourced Wi-Fi, Bluetooth, cellular, and IP databases) for positioning, ensuring their integrity and security is paramount. However, we demonstrate that these applications are susceptible to low-cost attacks (less than 50), including Wi-Fi spoofing combined with jamming, as well as more sophisticated coordinated location spoofing. These attacks manipulate position data to control or undermine functionality, leading to user scams or service manipulation. Therefore, we propose a countermeasure to detect and thwart such attacks by utilizing readily available, redundant positioning information from off-the-shelf platforms. Our method extends the receiver autonomous integrity monitoring (RAIM) framework by incorporating opportunistic information, including data from onboard sensors and terrestrial infrastructure signals, and, naturally,. We theoretically show that the fusion of heterogeneous signals improves resilience against sophisticated adversaries on multiple fronts. Experimental evaluations show the effectiveness of the proposed scheme in improving detection accuracy by 62% at most compared to baseline schemes and restoring accurate positioning.

The limited or no protection for civilian Global Navigation Satellite System (GNSS) signals makes spoofing attacks relatively easy. With modern mobile devices often featuring network interfaces, state-of-the-art signals of opportunity (SOP) schemes can provide accurate network positions in replacement of GNSS. The use of onboard inertial sensors can also assist in the absence of GNSS, possibly in the presence of jammers. The combination of SOP and inertial sensors has received limited attention, yet it shows strong results on fully custom-built platforms. We do not seek to improve such special-purpose schemes. Rather, we focus on countering GNSS attacks, notably detecting them, with emphasis on deployment with consumer-grade platforms, notably smartphones, that provide off-the-shelf opportunistic information (i.e., network position and inertial sensor data). Our Position-based Attack Detection Scheme (PADS) is a probabilistic framework that uses regression and uncertainty analysis for positions. The regression optimization problem is a weighted mean square error of polynomial fitting, with constraints that the fitted positions satisfy the device velocity and acceleration. Then, uncertainty is modeled by a Gaussian process, which provides more flexibility to analyze how sure or unsure we are about position estimations. In the detection process, we combine all uncertainty information with the position estimations into a fused test statistic, which is the input utilized by an anomaly detector based on outlier ensembles. The evaluation shows that the PADS outperforms a set of baseline methods that rely on SOP or inertial sensor-based or statistical tests, achieving up to 3 times the true positive rate at a low false positive rate.

Location-based service (LBS) applications proliferate and support transportation, entertainment, and more. Modern mobile platforms, with smartphones being a prominent example, rely on terrestrial and satellite infrastructures (e.g., global navigation satellite system (GNSS) and crowdsourced Wi-Fi, Bluetooth, cellular, and IP databases) for correct positioning. However, they are vulnerable to attacks that manipulate positions to control and undermine LBS functionality-Thus enabling the scamming of users or services. Our work reveals that GNSS spoofing attacks succeed even though smartphones have multiple sources of positioning information. Moreover, that Wi-Fi spoofing attacks with GNSS jamming are surprisingly effective. More concerning is the evidence that sophisticated, coordinated spoofing attacks are highly effective. Attacks can target GNSS in combination with other positioning methods, thus defenses that assume that only GNSS is under attack cannot be effective. More so, resilient GNSS receivers and special-purpose antennas are not feasible on smartphones. To address this gap, we propose an extended receiver autonomous integrity monitoring (RAIM) framework that leverages the readily available, redundant, often so-called opportunistic positioning information on off-The-shelf platforms. We jointly use onboard sensors, terrestrial infrastructures, and GNSS. We show that our extended RAIM framework improves resilience against location spoofing, e.g., achieving a detection accuracy improvement of up to 24-58% compared to the state-of-The-Art algorithms and location providers; detecting attacks within 5 seconds, with a low false positive rate.

Constructing channel state information (CSI) maps is essential for advancing wireless communications and localization. However, building CSI maps in practice is hindered by the need for up-to-date, location-labeled CSI measurement data. Conventional CSI embedding methods typically project CSI into a low-dimensional latent space, which may lack clear physical interpretability for localization tasks. This paper addresses the problem of extracting user locations from CSI measurements and recovering user trajectories in indoor multiple-input multiple-output (MIMO)-Orthogonal Frequency-Division Multiplexing (OFDM) communication scenarios. We develop a hidden Markov model (HMM) framework and propose an alternating optimization algorithm to jointly learn model parameters and recover user trajectories in both line-of-sight (LOS) and non-line-of-sight (NLOS) environments. Proof-of-concept experiments are conducted using ray-tracing data from a uniform linear antenna array (ULA) antenna setup, achieving an average localization error of 0.87 meters. The results demonstrate the potential of the proposed approach for practical indoor CSI map construction.

This paper demonstrates a robust machine learning-based optical wireless positioning system that achieves high accuracy under the presence of the interference, enabling reliable position estimation and dynamic interference rejection.

Global navigation satellite system (GNSS) and other assisted positioning infrastructures provide ubiquitous, precise locations for cyber-physical system (CPS), from autonomous vehicles to location-based service (LBS) applications on mobile phones in daily lives. Combining multiple satellite constellations, network infrastructures, and onboard sensors typically makes the position solutions more accurate and robust than any single source alone. 

However, civilian GNSS signals, Wi-Fi beacons, and cellular pilot signals lack cryptographic protection and are therefore vulnerable to signal spoofing attacks. Even if they can be upgraded to support authentication, meaconing or wormhole attacks can relay and falsify the wireless signals and then manipulate the localisation. More seriously, an attacker can selectively jam the wireless signals from specific infrastructures to force CPS to downgrade to less secure signals, which are later spoofed; coordinated adversaries can also target multiple infrastructures simultaneously to manipulate the positioning result. 

This thesis is in the broad area of data trustworthiness for CPS, focusing on the security and resilience of localisation. Emphasis is given on securing the localisation based on GNSS, as they are relevant to a multiplicity of modern systems (e.g., connected vehicles, smartphones, and other Internet-of-Things (IoT) platforms). Significant efforts are dedicated to detecting attacks on position and providing secure and reliable location information, even in the presence of adversaries and benign faults (e.g., challenging propagation environments). Where perfect recovery is unlikely, the proposed methods aim for a best-effort position estimation by opportunistically fusing the remaining available benign signals. 

These efforts are concerned with designing, analysing, implementing, and evaluating diverse protocols that address GNSS-specific attacks, other positioning signal attacks, and simultaneous GNSS with other signal attacks. The approaches are theoretically rigorous, are evaluated through detailed simulations, real-world experiments, and system implementation, proposing concrete defense mechanisms.

Global navigation satellite system (GNSS) och andra infrastrukturer för assisterad positionering tillhandahåller överallt närvarande, precisa positioner för cyber-physical system (CPS) — från autonoma fordon till location-based service (LBS)-applikationer i mobiltelefoner i vardagen. Fusionen av flera satellitkonstellationer, nätverksinfrastrukturer och ombordliggande sensorer gör positionslösningarna mer precisa och robusta än vad varje enskild lokaliseringsmetod kan erbjuda.

Avsaknaden av kryptografiskt skydd för GNSS-signaler, Wi-Fi-beacons och cellulära pilotsignaler gör dem dock sårbara för signalförfalskningsattacker. Även om systemen kan uppgraderas för att stödja autentisering, kan meaconing- och maskhålsattacker vidarebefordra och förfalska trådlösa signaler och därigenom manipulera positionsbestämningen. Ännu värre är att en angripare kan selektivt störa trådlösa signaler från specifika infrastrukturer för att tvinga CPS att falla tillbaka på mindre säkra signaler, vilka därefter kan förfalskas. På liknande sätt kan angriparen samordna attacker för att förfalska samtliga infrastrukturer.

Denna avhandling rör det breda området datatillförlitlighet för CPS, med fokus på säkerhet och motståndskraft vid lokalisering (positionering). Särskild tonvikt läggs på att säkra lokalisering baserad på GNSS, eftersom dessa är relevanta för en mängd moderna system — från smarta/uppkopplade fordon till smartphones och Internet-of-Things (IoT)-plattformar. Betydande insatser ägnas åt att upptäcka attacker mot positionsinformation och att tillhandahålla säker och tillförlitlig platsinformation även i närvaro av angripare och godartade fel (t.ex. i svåra utbredningsmiljöer). I vissa fall sker detta som en best-effort-lösning genom att utnyttja alternativa lokaliseringstekniker.

Dessa insatser omfattar design, analys, implementering och utvärdering av olika protokoll som hanterar GNSS-specifika attacker, andra attacker mot positioneringssignaler samt samtidiga attacker riktade mot GNSS och andra signaler. Metoderna är teoretiskt rigorösa och utvärderas genom detaljerade simuleringar, verkliga experiment och systemimplementation, och föreslår konkreta försvarsmekanismer.

Global navigation satellite systems (GNSS) are vulnerable to spoofing attacks, with adversarial signals manipulating the location or time information of receivers, potentially causing severe disruptions. The task of discerning the spoofing signals from benign ones is naturally relevant for machine learning, thus recent interest in applying it for detection. While deep learning-based methods are promising, they require extensive labeled datasets, consume significant computational resources, and raise privacy concerns due to the sensitive nature of position data. This is why this paper proposes a self-supervised federated learning framework for GNSS spoofing detection. It consists of a cloud server and local mobile platforms. Each mobile platform employs a self-supervised anomaly detector using long short-term memory (LSTM) networks. Labels for training are generated locally through a spoofing-deviation prediction algorithm, ensuring privacy. Local models are trained independently, and only their parameters are uploaded to the cloud server, which aggregates them into a global model using FedAvg. The updated global model is then distributed back to the mobile platforms and trained iteratively. The evaluation shows that our self-supervised federated learning framework outperforms position-based and deep learning-based methods in detecting spoofing attacks while preserving data privacy.