Ai Privacy Preserving Ml

Privacy Preserving Learning Foundations

Doctoral work examines training models without exposing the underlying data. These methods permit learning from information that cannot be shared.

Privacy Definition Research

Research examines what formal privacy guarantees actually promise and exclude. Definitions differ substantially and are frequently conflated in practice.

Threat Model Research

Doctoral study examines specifying what an adversary can observe and can do. Guarantees are meaningless without a clearly stated threat model.

Adversary Capability Research

Research examines the resources and knowledge attackers realistically possess. Assumed capability determines which protections are genuinely necessary.

Honest But Curious Model

Doctoral work examines adversaries following protocols while inferring from observations. This assumption is common and frequently too optimistic in practice.

Malicious Adversary Model

Research examines adversaries willing to deviate arbitrarily from agreed protocols. Protection against active deviation costs substantially more computation.

Trust Assumption Research

Doctoral study examines who must be trusted for a stated guarantee to hold. Hidden trust assumptions undermine many claimed privacy protections.

Privacy Utility Tradeoff

Research examines the loss in model quality that protection necessarily imposes. Every formal guarantee costs some measurable predictive performance.

Privacy Accounting Research

Doctoral work examines tracking cumulative privacy expenditure across operations. Accounting determines how many analyses a dataset can support.

Composition Theory Research

Research examines how guarantees degrade when mechanisms are combined. Tight composition analysis permits far more useful analysis per dataset.

Differential Privacy Foundations

Doctoral study examines the dominant formal framework for statistical privacy. This framework bounds what any observer can learn about an individual.

Privacy Parameter Interpretation

Research examines what formal privacy parameters actually mean in practice. Deployed parameter choices are frequently far weaker than theory intends.

Local Differential Privacy

Doctoral work examines protection applied before data leaves the individual device. Local protection requires no trusted party and costs much accuracy.

Central Differential Privacy

Research examines protection applied by a trusted curator holding raw data. Central protection gives better accuracy and requires trusting the curator.

Shuffle Model Research

Doctoral study examines an intermediate model using anonymous message shuffling. Shuffling achieves accuracy between the local and central approaches.

Renyi Privacy Research

Research examines a relaxed formulation supporting tighter composition analysis. This formulation underpins most practical deep learning privacy accounting.

Concentrated Privacy Research

Doctoral work examines formulations bounding privacy loss distribution tails. These variants give sharper guarantees for repeated mechanism application.

Approximate Privacy Research

Research examines relaxations permitting a small probability of guarantee failure. Relaxation improves utility and complicates interpretation of the promise.

Privacy Amplification Research

Doctoral study examines processes strengthening guarantees without additional noise. Amplification results permit substantially more accurate private analysis.

Subsampling Amplification

Research examines random sampling strengthening the achieved privacy guarantee. Sampling amplification is central to private gradient based training.

Noise Mechanism Research

Doctoral work examines randomness added to outputs to obscure individual influence. Mechanism choice determines the accuracy achieved at a given guarantee.

Laplace Mechanism Research

Research examines a foundational mechanism adding noise scaled to sensitivity. This mechanism established the practical feasibility of formal privacy.

Gaussian Mechanism Research

Doctoral study examines the mechanism underlying most private deep learning. This mechanism composes favourably across very many training iterations.

Exponential Mechanism Research

Research examines private selection among discrete candidate output options. Selection mechanisms handle tasks that additive noise cannot address.

Sensitivity Analysis Research

Doctoral work examines how much one record can change a computed output. Sensitivity determines how much noise a guarantee actually requires.

Gradient Clipping Research

Research examines bounding individual contributions during model training. Clipping bounds sensitivity and introduces its own optimisation difficulties.

Private Stochastic Optimisation

Doctoral study examines gradient based training under formal privacy constraints. This approach is the workhorse of practical private deep learning.

Private Risk Minimisation

Research examines theoretical limits of private learning from finite samples. Theory establishes what accuracy is achievable at any given guarantee.

Private Convex Optimisation

Doctoral work examines private learning where the objective is well behaved. Convex settings admit theoretical guarantees that general models lack.

Private Deep Learning Research

Research examines applying formal privacy to large neural network training. Accuracy penalties remain substantial for the largest current models.

Hyperparameter Privacy Research

Doctoral study examines privacy cost of tuning model configuration choices. Tuning leaks information and is frequently excluded from reported accounting.

Privacy Auditing Research

Research examines empirically testing whether claimed guarantees actually hold. Auditing has repeatedly exposed implementation flaws in released systems.

Empirical Privacy Estimation

Doctoral work examines measuring actual leakage rather than theoretical bounds. Empirical leakage is frequently far below the worst case bound.

Privacy Attack Evaluation

Research examines methodology for assessing how well protections resist attack. Evaluation quality determines whether protection claims can be believed.

Membership Inference Research

Doctoral study examines determining whether a record was in the training set. This is the standard benchmark attack for evaluating model leakage.

Attribute Inference Research

Research examines inferring unknown attributes of individuals from models. Attribute leakage differs from membership and is less thoroughly studied.

Model Inversion Research

Doctoral work examines recovering representative inputs from a trained model. Inversion demonstrates that models retain traces of their training data.

Data Reconstruction Research

Research examines recovering individual training records from model artefacts. Reconstruction results motivate much of this entire research field.

Gradient Leakage Research

Doctoral study examines information revealed by shared training gradients. Shared gradients can reveal training inputs with surprising fidelity.

Property Inference Research

Research examines inferring aggregate properties of a training dataset. Aggregate leakage can be commercially sensitive without identifying anyone.

Model Extraction Research

Doctoral work examines reproducing a model through query access alone. Extraction threatens intellectual property and enables further attacks.

Memorisation Research

Research examines models retaining specific examples rather than general patterns. Memorisation appears necessary for learning rare but genuine patterns.

Training Data Extraction

Doctoral study examines recovering verbatim training content from deployed models. Large language models have been shown to emit memorised content.

Canary Insertion Research

Research examines planting distinctive records to measure subsequent leakage. Canaries provide a practical empirical measure of memorisation.

Privacy Risk Measurement

Doctoral work examines quantifying realistic privacy risk for a deployed system. Risk depends on data sensitivity as much as on formal guarantees.

Federated Learning Foundations

Research examines training shared models without centralising the training data. Federation keeps data local while still permitting collective learning.

Cross Device Federation

Doctoral study examines federation across very many unreliable personal devices. Device federation involves enormous scale and intermittent participation.

Cross Silo Federation

Research examines federation between a small number of large organisations. Silo federation involves reliable participants with legal constraints.

Federated Optimisation Research

Doctoral work examines algorithms coordinating learning across distributed participants. Optimisation must tolerate infrequent and unreliable communication.

Client Selection Research

Research examines choosing which participants contribute in each training round. Selection affects convergence speed and fairness between participants.

Aggregation Method Research

Doctoral study examines combining contributions from distributed participants. Aggregation design determines both accuracy and robustness to interference.

Secure Aggregation Research

Research examines combining contributions without any party seeing individual ones. Secure combination prevents the coordinator inspecting single contributions.

Robust Aggregation Research

Doctoral work examines combination resisting corrupted or malicious contributions. Robustness and privacy protection frequently pull against one another.

Communication Efficiency Research

Research examines reducing data exchanged during distributed model training. Communication is the dominant bottleneck in most federated deployments.

Model Compression Research

Doctoral study examines reducing the size of transmitted model information. Compression trades communication savings against convergence quality.

Quantisation Research

Research examines representing transmitted values at reduced numerical precision. Reduced precision cuts communication with modest accuracy penalty.

Sparsification Research

Doctoral work examines transmitting only the most significant model changes. Sparse transmission reduces communication volume very substantially.

Statistical Heterogeneity Research

Research examines participants holding data with differing statistical properties. Heterogeneity is the central technical difficulty in federated learning.

Non Identical Distribution Research

Doctoral study examines learning where participant data distributions differ. Standard optimisation assumptions fail badly under such conditions.

Systems Heterogeneity Research

Research examines participants differing in computation and network capability. Capability differences cause slow participants to hold back progress.

Client Divergence Research

Doctoral work examines local models moving apart during extended local training. Divergence degrades the combined model and limits local computation.

Personalisation Research

Research examines adapting shared models to individual participant needs. Personalisation addresses heterogeneity that global models cannot accommodate.

Personalisation Method Research

Doctoral study examines techniques producing participant specific model behaviour. Methods range from fine tuning to entirely separate model components.

Meta Learning In Federation

Research examines learning initialisations that adapt quickly to each participant. Meta learning frames personalisation as rapid local adaptation.

Clustered Federation Research

Doctoral work examines grouping similar participants and training group models. Clustering suits populations containing distinct underlying subgroups.

Hierarchical Federation Research

Research examines federation organised across several coordination layers. Hierarchy reduces load on any single coordinating server substantially.

Decentralised Learning Research

Doctoral study examines learning without any central coordinating server. Removing the coordinator eliminates a single point of trust and failure.

Gossip Protocol Research

Research examines participants exchanging information with random neighbours. Gossip protocols spread learning without any central coordination.

Peer To Peer Learning

Doctoral work examines direct model exchange between participating parties. Direct exchange raises distinctive trust and verification questions.

Federated Analytics Research

Research examines computing statistics across distributed data without centralising. Analytics questions differ substantially from model training questions.

Federated Evaluation Research

Doctoral study examines assessing model quality across distributed participants. Evaluation itself can leak information about participant data.

Vertical Federated Learning

Research examines parties holding differing attributes about the same individuals. Vertical settings require secure protocols for every training step.

Feature Partitioned Learning

Doctoral work examines training where features are split across organisations. Feature splitting arises whenever organisations hold complementary records.

Entity Alignment Research

Research examines matching records across parties without revealing identities. Alignment is the necessary first step in vertical federated settings.

Private Set Intersection

Doctoral study examines finding shared records without disclosing other entries. This protocol underpins privacy respecting record matching.

Federated Transfer Learning

Research examines transferring knowledge between participants with limited overlap. Transfer helps participants whose own data is genuinely insufficient.

Split Learning Research

Doctoral work examines splitting a model between participant and central server. Splitting reduces client computation and creates new leakage surfaces.

Model Partitioning Research

Research examines distributing model components across differing execution locations. Partitioning balances computation, communication and exposure.

Edge Deployment Research

Doctoral study examines running private learning on devices near the data. Edge deployment avoids transmitting sensitive information at all.

Resource Constrained Learning

Research examines private learning under severe memory and computation limits. Privacy techniques are expensive and constrained devices are common.

Energy Efficiency Research

Doctoral work examines the energy consumed by privacy preserving computation. Cryptographic protection can increase energy use by large factors.

Straggler Handling Research

Research examines managing participants that respond slowly or not at all. Waiting for slow participants dominates federated round duration.

Fault Tolerance Research

Doctoral study examines continuing training when participants fail mid protocol. Failure is entirely routine at scale and protocols must expect it.

Incentive Mechanism Research

Research examines encouraging genuine participation in collaborative learning. Participants bear costs while benefits accrue to the whole federation.

Contribution Measurement

Doctoral work examines quantifying what each participant adds to a shared model. Measurement must not itself reveal participant data characteristics.

Free Riding Research

Research examines participants benefiting without contributing genuine effort. Free riding undermines the incentive to join collaborative learning.

Federation Governance Research

Doctoral study examines who decides how a collaborative learning effort operates. Governance determines whose interests the resulting model serves.

Poisoning Attack Research

Research examines corrupted contributions degrading a collaboratively trained model. Federated settings expose training to untrusted participants directly.

Hidden Trigger Attack Research

Doctoral work examines attacks implanting behaviour activated by specific inputs. Such attacks are difficult to detect through ordinary evaluation.

Byzantine Robustness Research

Research examines protocols tolerating arbitrarily misbehaving participants. Robustness guarantees typically assume a bounded fraction misbehave.

Federated Anomaly Detection

Doctoral study examines identifying suspicious contributions within federations. Detection is difficult because contributions are deliberately concealed.

Client Authentication Research

Research examines verifying that participants are who they claim to be. Authentication must not compromise participant privacy in the process.

Sybil Resistance Research

Doctoral work examines preventing one party creating many false identities. False identities let a single attacker dominate a whole federation.

Federated Privacy Guarantee

Research examines what federation alone actually protects and what it does not. Keeping data local is not by itself a formal privacy guarantee.

Combined Technique Research

Doctoral study examines stacking several protection mechanisms together. Combination is usually necessary because no single technique suffices.

Federated Simulation Research

Research examines simulating federated settings for experimental evaluation. Most published work uses simulation rather than genuine deployment.

Benchmark Research

Doctoral work examines shared datasets and tasks for comparing methods. Benchmark realism determines whether reported gains transfer to practice.

Reproducibility Research

Research examines whether published privacy results can be independently repeated. Reproduction is complicated by unreported implementation detail.

Framework Research

Doctoral study examines software libraries implementing privacy techniques. Implementation errors in libraries silently invalidate stated guarantees.

Deployment Case Research

Research examines genuine production deployments and what they revealed. Real deployments expose difficulties that simulation entirely conceals.

Cryptographic Machine Learning

Doctoral work examines applying cryptography to protect learning computations. Cryptography offers strong guarantees at substantial computational cost.

Secure Multiparty Computation

Research examines parties jointly computing without revealing their own inputs. These protocols permit collaboration between mutually distrusting parties.

Secret Sharing Research

Doctoral study examines splitting values across parties so none learns anything. Secret sharing underpins many efficient secure computation protocols.

Garbled Circuit Research

Research examines evaluating functions on encrypted inputs between two parties. This approach suits computations expressed as boolean circuits.

Oblivious Transfer Research

Doctoral work examines a primitive underpinning most secure computation protocols. Efficient realisation of this primitive determines overall protocol speed.

Homomorphic Encryption Research

Research examines computing directly upon encrypted values without decrypting. This capability permits outsourcing computation on sensitive data.

Partially Homomorphic Methods

Doctoral study examines schemes supporting only some operations on encrypted data. Limited schemes are far faster and constrain what can be computed.

Fully Homomorphic Learning

Research examines arbitrary computation performed entirely on encrypted data. Full capability remains extremely slow for realistic model training.

Encrypted Inference Research

Doctoral work examines running trained models on encrypted user inputs. Encrypted inference protects users from the model provider entirely.

Encrypted Training Research

Research examines training models without ever decrypting the training data. Encrypted training is far harder than encrypted inference currently.

Bootstrapping Efficiency Research

Doctoral study examines refreshing ciphertexts to permit unlimited computation. This refresh operation dominates the cost of encrypted computation.

Ciphertext Packing Research

Research examines encoding many values within a single encrypted container. Packing enables parallel operations and greatly improves throughput.

Approximate Arithmetic Research

Doctoral work examines encrypted schemes supporting approximate real number arithmetic. Approximate schemes suit machine learning far better than exact ones.

Functional Encryption Research

Research examines keys revealing only a specific function of encrypted data. Functional keys permit controlled disclosure of computed results.

Zero Knowledge Proof Research

Doctoral study examines proving statements without revealing supporting information. Such proofs let parties demonstrate correctness without disclosure.

Verifiable Computation Research

Research examines proving that outsourced computation was performed correctly. Verification matters when computation happens on untrusted infrastructure.

Proof Of Training Research

Doctoral work examines demonstrating a model was trained as its owner claims. Training proofs would support both auditing and regulatory compliance.

Trusted Execution Environment

Research examines hardware isolated regions protecting computation and data. Hardware protection is fast and shifts trust onto the chip manufacturer.

Enclave Security Research

Doctoral study examines the security properties of isolated hardware regions. Repeated vulnerabilities have undermined confidence in these protections.

Side Channel Research

Research examines information leaking through timing, power or memory patterns. Side channels bypass protections that appear mathematically sound.

Hardware Acceleration Research

Doctoral work examines specialised hardware speeding privacy preserving computation. Acceleration is essential for cryptographic methods to become practical.

Cryptographic Efficiency Research

Research examines reducing the computational cost of secure protocols. Cost remains the principal barrier to widespread cryptographic adoption.

Protocol Design Research

Doctoral study examines constructing secure protocols for learning tasks. Design must balance security, efficiency and implementation practicality.

Protocol Verification Research

Research examines formally proving that protocols achieve claimed properties. Formal verification catches flaws that informal argument entirely misses.

Post Quantum Consideration

Doctoral work examines protections remaining secure against quantum adversaries. Data protected today may need to resist future decryption attempts.

Private Information Retrieval

Research examines fetching records without revealing which record was requested. Retrieval privacy matters for search and recommendation systems.

Oblivious Inference Research

Doctoral study examines prediction where neither party learns the other input. Oblivious protocols protect both the user and the model owner.

Private Nearest Neighbour

Research examines similarity search without revealing queries or stored records. Similarity search underpins many retrieval and matching applications.

Private Clustering Research

Doctoral work examines grouping records while protecting individual membership. Clustering outputs can reveal a great deal about individual records.

Private Statistics Research

Research examines releasing summary statistics with formal privacy guarantees. Statistical release is where formal privacy was first deployed at scale.

Private Query Answering

Doctoral study examines answering database queries under privacy constraints. Query budgets limit how many questions a dataset can safely answer.

Synthetic Data Generation

Research examines producing artificial data resembling a sensitive dataset. Synthetic data promises sharing without exposing any real record.

Private Generative Model Research

Doctoral work examines training generative models under formal privacy constraints. Private generation degrades quality more than discriminative tasks.

Synthetic Data Utility Research

Research examines whether synthetic data supports the analyses it is intended for. Utility must be assessed against the specific downstream task.

Synthetic Data Privacy Risk

Doctoral study examines leakage remaining within generated synthetic datasets. Synthetic data is not automatically private without formal guarantees.

Anonymisation Research

Research examines removing identifying information from released datasets. Classical anonymisation has repeatedly failed against determined attackers.

Pseudonymisation Research

Doctoral work examines replacing identifiers while retaining record linkability. Pseudonymisation reduces exposure without providing genuine anonymity.

Reidentification Risk Research

Research examines whether individuals can be recognised in released datasets. Small numbers of attributes are frequently sufficient to identify people.

Linkage Attack Research

Doctoral study examines combining datasets to identify individuals within them. Linkage attacks motivated the development of formal privacy definitions.

Data Minimisation Research

Research examines collecting and retaining only what a task genuinely requires. Minimisation reduces risk more reliably than any protection technique.

Purpose Limitation Research

Doctoral work examines restricting data use to originally stated purposes. Technical enforcement of stated purpose remains largely unsolved.

Machine Unlearning Research

Research examines removing the influence of specific records from trained models. Removal supports rights that data protection law now grants individuals.

Data Erasure Request Research

Doctoral study examines honouring individual requests to remove personal data. Requests are difficult to satisfy once models have already been trained.

Influence Estimation Research

Research examines quantifying how much one record shaped a trained model. Influence estimation underpins both unlearning and model debugging.

Certified Removal Research

Doctoral work examines provable guarantees that record influence has been removed. Certification distinguishes genuine removal from approximate forgetting.

Retraining Efficiency Research

Research examines rebuilding models cheaply after records must be removed. Full retraining is the reliable approach and is frequently infeasible.

Model Editing Research

Doctoral study examines modifying specific knowledge within an already trained model. Editing offers a route to removal without complete retraining.

Provenance Tracking Research

Research examines recording which data contributed to which model version. Provenance is essential for both auditing and honouring removal requests.

Watermarking Research

Doctoral work examines embedding detectable signals within models or datasets. Watermarks support demonstrating ownership and detecting unauthorised use.

Model Ownership Research

Research examines establishing and defending rights over trained models. Ownership questions interact closely with data rights and with privacy.

Healthcare Applications

Doctoral study examines privacy preserving learning applied to clinical data. Health data is sensitive, valuable and legally difficult to share.

Clinical Data Federation

Research examines hospitals collaborating without pooling any patient records. Clinical federation permits studies no single institution could conduct.

Genomic Privacy Research

Doctoral work examines protecting genetic information within learning systems. Genetic data identifies individuals and their relatives permanently.

Medical Imaging Privacy

Research examines privacy risks arising from medical image based learning. Images can be identifying even after conventional metadata removal.

Financial Applications

Doctoral study examines privacy preserving learning within financial services. Institutions hold sensitive records and face strict disclosure limits.

Collaborative Fraud Detection

Research examines institutions detecting fraud without sharing customer records. Collaboration substantially improves detection of coordinated activity.

Telecommunications Applications

Doctoral work examines privacy preserving learning across network operator data. Network data reveals movement and behaviour in fine detail.

Mobility Data Privacy

Research examines protecting information about how individuals move around. Movement traces are highly distinctive and very difficult to anonymise.

Location Privacy Research

Doctoral study examines protecting individual location within learning systems. A few location points are frequently sufficient to identify someone.

Smart Meter Privacy

Research examines protecting household information within energy usage data. Consumption patterns reveal occupancy and household activity in detail.

Connected Device Privacy

Doctoral work examines privacy for learning across networked household devices. These devices observe intimate settings with minimal user awareness.

Speech And Audio Privacy

Research examines protecting speaker identity within audio based learning. Speech carries identity, health and emotional information simultaneously.

Text And Language Privacy

Doctoral study examines privacy risks within models trained on written text. Writing style itself can identify authors with considerable reliability.

Large Language Model Privacy

Research examines leakage from models trained on very large text collections. These models have demonstrably emitted memorised personal information.

Retrieval System Privacy

Doctoral work examines privacy where models retrieve from external document stores. Retrieval introduces leakage paths that model training alone lacks.

Recommendation Privacy Research

Research examines privacy within systems learning individual user preferences. Preference data is revealing and commercially extremely valuable.

Advertising Data Privacy

Doctoral study examines privacy techniques within advertising measurement systems. This sector has driven substantial deployment of these methods.

Education Data Privacy

Research examines protecting learner information within educational analytics. Educational records concern minors and follow them for a long time.

Official Statistics Privacy

Doctoral work examines formal privacy applied to national statistical releases. Statistical agencies were among the earliest large scale adopters.

Government Data Research

Research examines privacy preserving analysis of administrative government records. Administrative data is comprehensive and politically very sensitive.

Cross Border Data Research

Doctoral study examines learning across differing legal jurisdictions. Legal restrictions on transfer motivate much federated learning work.

Privacy Fairness Interaction

Research examines how privacy protection affects fairness between population groups. Protection mechanisms can worsen performance for minority groups.

Disparate Privacy Impact

Doctoral work examines protection costs falling unevenly across populations. Underrepresented groups typically lose the most accuracy under protection.

Group Privacy Research

Research examines protecting information about communities rather than individuals. Formal definitions protect individuals and say little about groups.

Contextual Integrity Research

Doctoral study examines privacy as appropriate information flow within contexts. This framing captures concerns that formal definitions entirely miss.

Consent Mechanism Research

Research examines obtaining meaningful permission for data use in learning. Consent is difficult where future uses cannot be specified in advance.

Transparency Research

Doctoral work examines communicating what protection a system actually provides. Stated guarantees are routinely misunderstood by those relying on them.

Explainability Privacy Tension

Research examines explanations themselves leaking information about training data. Transparency and protection pull directly against one another here.

Auditing And Assurance

Doctoral study examines independent verification of deployed privacy claims. Assurance requires access that system operators rarely willingly grant.

Certification Research

Research examines schemes formally certifying privacy preserving systems. Certification requires standards that the field has not yet settled.

Standards Research

Doctoral work examines technical standards for privacy preserving computation. Standards would permit meaningful comparison between competing systems.

Regulatory Compliance Research

Research examines whether technical protections satisfy legal requirements. Formal guarantees do not map cleanly onto existing legal categories.

Data Protection Law Research

Doctoral study examines legal frameworks governing personal data within learning. Law shapes which technical approaches are practically usable at all.

Cross Jurisdiction Research

Research examines conflicting legal requirements across differing countries. Conflicts force architectural choices before any technical design begins.

Privacy Impact Assessment

Doctoral work examines structured evaluation of privacy risk before deployment. Assessment quality determines whether risks are identified in time.

Risk Communication Research

Research examines conveying privacy risk to people who bear the consequences. Technical parameters convey almost nothing to nonspecialist audiences.

User Perception Research

Doctoral study examines what people believe privacy technologies actually protect. Beliefs frequently diverge sharply from what is actually guaranteed.

Developer Practice Research

Research examines how practitioners actually implement privacy techniques. Implementation errors are common and silently void stated guarantees.

Privacy Tool Usability

Doctoral work examines whether practitioners can use privacy libraries correctly. Poor usability is a leading cause of deployed protection failures.

Adoption Barrier Research

Research examines obstacles preventing organisations deploying these techniques. Accuracy cost and engineering complexity dominate cited barriers.

Cost Of Privacy Research

Doctoral study examines the resources protection consumes in real deployments. Protection costs computation, accuracy and engineering effort together.

Environmental Cost Research

Research examines the energy and emissions from privacy preserving computation. Cryptographic protection substantially increases computational demand.

Open Source Ecosystem Research

Doctoral work examines the shared software underpinning this whole research field. Much critical tooling depends on very small maintainer teams.

Education And Training Research

Research examines preparing practitioners to apply these techniques correctly. Expertise combining cryptography and learning is genuinely scarce.

Benchmark Standardisation

Doctoral study examines agreed tasks for comparing privacy preserving methods. Inconsistent evaluation makes published comparisons largely meaningless.

Evaluation Methodology Research

Research examines how privacy and utility should jointly be assessed together. Reporting either measure alone gives a seriously misleading picture.

Reporting Standard Research

Doctoral work examines what must be disclosed about any stated privacy claim. Incomplete reporting prevents independent appraisal of the guarantees.

Overclaiming Research

Research examines privacy claims exceeding what techniques actually deliver. Overclaiming is widespread and erodes trust in genuine protections.

Emerging Threat Research

Doctoral study examines new attack capabilities against protected systems. Threat capability advances continuously and defences must keep pace.

Implementation And Adoption

Research examines why these techniques are or are not deployed in practice. Adoption depends on tooling maturity as much as theoretical advance.