Privacy-Utility-Fairness: A Balanced Approach to Vehicular-Traffic Management System

Frank Eliassen; Poushali Sengupta; Sabita Maharjan; Yan Zhang

REVIEW 5 major objections 5 minor 26 references

Privacy-Utility-Fairness: A Balanced Approach to Vehicular-Traffic Management System

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that query-based access, dual-stage iterative shuffling, and Laplace noise together balance privacy, utility, and fairness for vehicular traffic data.

desk verdict The DP guarantee is asserted using the wrong mechanism, and the fairness theorems are tautologies; the paper's real value is the reproducible simulation and heatmap, but it needs a major rewrite before it can be trusted. read the letter →

arxiv 2507.08864 v1 pith:KQLQK47X submitted 2025-07-09 cs.CR cs.AIcs.MA

Poushali Sengupta , Sabita Maharjan , frank Eliassen , Yan Zhang This is my paper · ORCID

classification cs.CRcs.AIcs.MA

keywords DifferentialPrivacyUtilityFairnessIterativeShufflingVehicularTrafficManagementLocation-basedServicesLaplaceMechanism

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Location-based traffic management must protect drivers' precise positions while still giving planners reliable congestion data, and it must not systematically ignore rural or small regions. This paper proposes a single pipeline to achieve all three: query-based access limits data retrieval to the region and time asked for, dual-stage iterative shuffling permutes records within each regional group and then across groups so group proportions are unchanged, and Laplace noise is added to the released records. The authors claim this yields $\epsilon$-differential privacy, preserves traffic-density utility, and keeps every region proportionally represented. If the claim is right, traffic authorities can publish heatmaps and density forecasts without revealing individual trajectories and without biasing resource allocation toward densely covered urban areas.

What carries the argument

The central mechanism is the dual-stage iterative shuffling combined with the Laplace mechanism. Dual-stage shuffling works by first permuting the order of records inside each subgroup $G_j$ (local shuffling, with permutation $\sigma_j$) and then permuting the union of all shuffled subgroups (global shuffling, with permutation $\sigma$), so that the fraction $\varphi_j = |G_j|/|D|$ of each group is unchanged while inter-record linkages are scrambled. The Laplace mechanism then adds independent noise $N \sim \mathrm{Laplace}(0,b)$ to each record with $b=\Delta f/\epsilon$, which the paper claims gives $\epsilon$-differential privacy and, because the noise is independent of group membership, distributes error uniformly across groups.

What would settle it

Run the pipeline on two datasets differing by one record, repeat the release many times, and check whether every output satisfies the $\epsilon$-differential privacy bound $P(M(D) \in S) \le e^\epsilon P(M(D') \in S)$; the claim is falsified if any output violates the bound. Equivalently, compute the true maximum change ($\Delta f$) of the region-and-time query when one record is added: if it is not the value used in $b=\Delta f/\epsilon$, the guarantee collapses.

Watch

Extended reading notes

Core claim

The paper's central claim is that privacy, utility, and fairness can be balanced in a vehicular traffic-management system by combining three operations: answering only constrained queries (region and time), applying a two-stage iterative shuffling that first permutes records inside each geographical group $G_j$ and then permutes the union of shuffled groups, and finally adding Laplace noise $N\sim\mathrm{Laplace}(0,b)$ with scale $b=\Delta f/\epsilon$ to every record. The two-stage shuffling is asserted to preserve group proportions $\varphi_j = |G_j|/|D|$, hence fairness, and to break patterns vulnerable to linkage attacks; the Laplace noise is asserted to provide $\epsilon$-differential privacy and, because it is independent of group membership, to leave proportional representation intact. The authors state these as Theorem 1 and Theorem 2 and demonstrate the pipeline on a simulated traffic dataset for 50 Norwegian cities, reporting that $\epsilon = 2$ gives the best privacy-utility trade-off and that the resulting heatmaps and 24-hour predictions preserve traffic patterns.

Load-bearing premise

The whole framework stands or falls on the assumption that per-record Laplace noise with a scale derived from the query's maximum change provides $\epsilon$-differential privacy for a query that returns a set of records.

Editorial extensions

If this is right

Traffic authorities could release congestion heatmaps and density forecasts under a formal privacy budget, since the pipeline's Laplace noise is claimed to satisfy $\epsilon$-differential privacy.
Because group proportions are preserved by the two-stage shuffling, no region is over- or under-represented in the released data, supporting fairer allocation of road improvements and services.
The query-based access step keeps raw records at the edge and only transmits answers to specific region-and-time queries, reducing exposure of attributes irrelevant to traffic management.
The reported experiments indicate a tunable trade-off: at $\epsilon = 2$ the mean squared and mean absolute errors are minimized, so operators can pick a budget that keeps utility while protecting privacy.
The same pipeline is intended to transfer to other urban settings, since it operates on standard inputs (location, speed, count, time) and produces interpretable visual outputs.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural next test is to run the pipeline on real GPS traces rather than simulated city data, and to compare the privacy-utility-fairness frontier against simple baselines such as planar Laplace noise or $k$-anonymity.
The shuffling step is described as permuting whole records; an effective defense against linkage attacks would additionally need to break correlations between speed and location across timestamps, so a stress-test against a velocity-based tracker is a direct way to check the unlinkability claim.
The framework treats fairness as proportional representation of geographic groups; one could extend the same machinery to demographic groups (e.g., driver age or vehicle type) by redefining the groups $G_j$, which the formulation already allows.
Combining the decaying privacy-budget allocation with the utility-loss function suggests an optimization question the paper leaves open: how to choose the decay schedule across a day of queries to keep the loss $L(U,P)$ minimal over time.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The central privacy claim doesn't hold up: they add per-record Laplace noise to a set-valued query result without ever bounding L1 sensitivity, so the standard Laplace mechanism simply cannot be invoked. The sensitivity Δf is undefined for a query that returns a set of records, and there is no composition argument for the number of records or attributes. The result is that the advertised epsilon-DP guarantee is not established. The stress-test note is right about this, and it's the load-bearing claim of the paper.

The fairness theorems are just restatements of definitions. Theorem 1 concludes that shuffling preserves group proportions because shuffling is defined as a permutation and φj is defined as |Gj|/|D|. Theorem 2 concludes that noise preserves group membership because the noise is independent of group membership. Both are true by construction and carry no information. The paper's own description of shuffling is also inconsistent with Table I, where locations and speeds move across anonymized IDs in a way that row permutation alone cannot produce.

So what is actually here? The application to vehicular traffic management, with the dual-stage shuffling mechanism, is a domain application of the authors' earlier BUDS and Fairly Private work. That is not a new method, but the implementation effort is real: they provide the code and simulated data on GitHub, and the heatmap is a functional visualization. The utility–privacy trade-off plot (MSE vs. epsilon) is a straightforward empirical exercise, and the results are plausible. The abstract's claim that they implemented on "vehicular location-based data from Norway" is misleading—the data is synthetically generated—but the paper does say it is generated in the experiment section, so that's more a framing problem than a fabrication.

The loss function in Eq. 2 is never actually used in the experiments, and the decaying privacy budget allocation is described but not evaluated. Those are softer issues compared to the privacy gap.

Would I send this to peer review? Yes, but with the expectation of heavy revision. The authors have a concrete system, code, and a clear evaluation setup, so a reviewer can give actionable feedback. The privacy claim needs to be either fixed with a valid DP argument (e.g., treating the output as a fixed-dimensional histogram and adding calibrated noise with proper sensitivity) or removed and replaced with a weaker, clearly stated empirical privacy-utility trade-off. The theorems should be rewritten or removed.

For a reading group, this is a useful example of a common DP pitfall. I would not cite it in my own work as is, but I'd keep an eye on a revised version.

Referee Report

5 major / 5 minor

Summary. The paper proposes a framework for balancing privacy, utility, and fairness in location-based vehicular-traffic management systems. The method combines query-based data access, a two-stage iterative shuffling procedure, and Laplace noise injection, and claims to satisfy epsilon-differential privacy while preserving traffic-pattern utility and proportional representation of geographic groups. The authors evaluate the approach on a synthetic dataset mimicking traffic for 50 Norwegian cities and present heatmaps and utility-vs-privacy plots. The central advertised guarantees — epsilon-DP and fairness — are asserted in Sections II and V, with two formal theorems in Section II.

Significance. If the stated guarantees were valid, the paper would offer a practically relevant mechanism for protecting individual location data while retaining aggregate usefulness and regional fairness. The manuscript also (to its credit) states that code and data are openly available on GitHub, and it builds on prior work on shuffling-based privacy. However, the central privacy guarantee is not proven: the Laplace sensitivity for the set-valued query is never defined, no composition theorem is supplied, and the two theorems reduce to definitional tautologies rather than substantive guarantees. The experiments are purely synthetic and lack comparisons with existing methods or a fairness metric. The core claims therefore currently rest on unsupported assertions, and the paper in its present form does not meet the standards for a security or privacy venue.

major comments (5)

[Section II, noise injection paragraph] The paper defines the query output as a set of records R = {r1, ..., rk} and then states that the Laplace mechanism with scale b = Δf/ε is applied to R′ to obtain R″. The standard Laplace mechanism requires a numeric function f with bounded L1 sensitivity Δf = max_{D~D'} ||f(D)-f(D')||_1, where f(D) is a fixed-dimension vector. For a set-valued query returning records with mixed attributes, no such sensitivity is defined; even if one interprets Q(D) as a high-dimensional indicator vector, the paper gives no bound, and a single record change can alter the number of records and locations arbitrarily. Consequently Theorem 2 and the sentence 'We ensure adherence to epsilon-differential privacy standards' do not follow from the stated mechanism.
[Theorem 1 and Section II fairness definition] Since local shuffling permutes records within G_j and global shuffling applies a permutation σ to the union, the multiset of records is unchanged, so |G_j| and |D| are identical in R′; hence φ′_j = φ_j holds with probability one after a single iteration, not 'with high probability as n→∞'. Equations (3)-(4) are not a probabilistic convergence argument; Eq. (4) is the sample variance of a deterministic sequence of proportions, which is zero at every iteration. The theorem is therefore a restatement of the definition of shuffling and provides no fairness guarantee beyond preserving group sizes.
[Theorem 2] The proof establishes that Laplace noise is independent of group membership and therefore does not change the expected proportion of each group; it does not establish ε-differential privacy, nor does it establish that noise error is 'uniformly distributed across all subgroups' in any meaningful sense. In particular, the statement V ar(R″ | G_j) = V ar(R′ | G_j) + V ar(N | G_j) requires independence of R′ and N and says nothing about the DP guarantee. Since the DP claim is the paper's central privacy contribution, this theorem must be replaced with a proof that the released R″ is ε-DP for a well-defined query and sensitivity.
[Section II, privacy budget allocation] The text says total budget ε_T is distributed across queries Q1,...,Qm and subsequent queries consume progressively smaller portions, but no composition theorem is stated. For the Laplace mechanism, sequential composition would add the per-query budgets; the claim that a decaying allocation 'ensures cumulative privacy leakage remains within acceptable limits' is unproved and, as written, incompatible with standard sequential composition unless the queries operate on disjoint data partitions. This is an additional load-bearing gap in the privacy argument.
[Section III] The experiments are run on a synthetic dataset generated by the authors, not on real vehicular-location data from Norway; the abstract and conclusion describe the evaluation as 'implemented on vehicular location-based data from Norway' without this caveat. Moreover, no comparison with existing privacy-preserving traffic management methods is provided, and the only quantitative utility metrics reported are MSE/MAE as a function of ε, with no fairness metric or privacy attack evaluation. The empirical claim of a 'balanced trade-off' is therefore not demonstrated.

minor comments (5)

[Table I] The shuffled dataset changes location, speed, and count values across anonymized IDs, but the shuffling operation is defined as a row permutation σ(R), which preserves each row's joint attributes. The mechanism producing Table I is therefore ambiguous and should be clarified.
[Equation (2)] The expression Risk(U,P) = arg min_{ε,σ} L(U,P) is not well-formed: the left-hand side is a value while the right-hand side is an argmin, and U and P are not defined in the surrounding text.
[Section III.A] The text states that rainy and foggy conditions increase traffic density by 10%, 'which could add up to 100 extra vehicles per hour', and clear weather results in a 'modest rise of 20 vehicles per hour'; these are inconsistent unless the baseline counts are specified.
[Proof of Theorem 1] There is a typo 'V ar(φ'(n)_j) → 0..' with a double period, and the notation φ'(n)_j versus φ(n)_j is used inconsistently.
[Section III.C] The GitHub URL is broken across lines and ends with 'trafﬁc prediction with' followed by nothing; the repository link should be complete and verified.

Circularity Check

2 steps flagged · score 6.0 of 10

Two formal fairness theorems are definitional tautologies: shuffling-as-permutation preserves proportions by construction, and group-independent noise preserves groups by assumption.

self definitional [Section II, Theorem 1 and proof]
"After a single iteration, the shuffled dataset is, R′ = σ(⋃ m j=1 G′ j ). The proportion of group Gj in the shuffled dataset after one iteration can be expressed as: φ(1) j = |G′(1) j | / |R′| ... As the local shuffling happens within internal elements, it does not change the number of elements in the group."

The theorem concludes that iterative shuffling preserves group proportions φ′_j = φ_j, but this is immediate from the paper's own definitions: R′ is a permutation σ of the group-tagged records, and permuting records cannot change the cardinality of any group. No randomness or convergence argument is needed; the claimed variance decrease is surplus. The result is therefore an unpacking of the definition of shuffling-as-permutation, not an independently derived fairness guarantee.
self definitional [Section II, Theorem 2 and proof]
"The Laplace noise N is applied independently to each data point in R′. ... Since N is independent of group membership, the expected noise magnitude and variance are identical for all groups Gj. Since Laplace noise is unbiased and independent of group membership, it does not alter the proportional representation of groups in R′′: Pr[ri ∈ Gj | R ′′] = φj."

The theorem's conclusion that noise is uniformly distributed across subgroups is exactly the premise that N is i.i.d. Laplace noise injected independently of group membership. The proof restates this independence in the form of equal expected noise and variance, then concludes proportional representation is unchanged. That conclusion is equivalent to the assumption that noise does not depend on group labels, so the theorem proves its own input rather than establishing a property of the mechanism.

full rationale

The two formal fairness theorems reduce to their definitions. Theorem 1 defines shuffling as a permutation and then observes that permutation preserves group cardinalities, which is a tautology; Theorem 2 assumes group-independent Laplace noise and then concludes that noise is group-independent, another tautology. The central ε-differential privacy claim is not circular in the same way: it invokes the standard Laplace mechanism theorem with a sensitivity Δf, but the paper never proves a finite L1 sensitivity for its set-valued query, and adding independent Laplace noise to every record is not the textbook Laplace mechanism. That is a correctness gap rather than a definitional reduction. The self-citations to the authors' prior BUDS and Fairly Private work are not used as substitutes for these proofs, so they do not further increase the score. Because the paper's two formal fairness results are equivalent to their own assumptions by construction, while some independent empirical content remains, the appropriate circularity score is 6.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The framework relies on standard DP machinery plus hand-chosen simulation constants. The main burden is the unproven application of the Laplace mechanism to set-valued outputs, the tautological fairness theorems, and the synthetic data standing in for real Norwegian traffic.

free parameters (4)

Privacy budget epsilon = 2
Chosen after sweeping MSE/MAE on the authors' synthetic dataset to minimize error, not derived from a privacy requirement or prior theory.
Loss trade-off weights alpha and beta = not assigned
Introduced in Eq. (2) for the risk minimization but never instantiated or used in experiments.
Query sensitivity Delta_f = not specified
Assumed finite to scale Laplace noise, but Q returns a record set and no L1 sensitivity is actually defined.
Weather traffic multipliers (rain 1.1, snow 1.2, clear +20 vehicles/h) = 1.1, 1.2, +20 veh/h
Hand-set constants in the synthetic data generator, adopted from weather-traffic references [24-26]; they shape the utility evaluation.

assumptions (5)

standard math A permutation shuffling operation preserves group proportions exactly.
Used in Theorem 1; true by definition of permutation, but presented as a convergence result.
domain assumption Laplace noise applied independently to each data point is independent of group membership and preserves group proportions.
Used in Theorem 2; assumes noise is added to attributes without reassigning group labels.
ad hoc to paper Per-record Laplace noise with scale Delta_f/epsilon provides epsilon-differential privacy for the set-valued query Q.
The paper asserts epsilon-DP compliance, but does not prove that the set-valued sensitivity is bounded or that per-record perturbation is the correct DP mechanism; this is the load-bearing privacy premise.
domain assumption The synthetic traffic generator reproduces realistic Norwegian traffic patterns.
The utility evaluation assumes simulated data stands in for real Norwegian vehicular data; the abstract even says data from Norway.
standard math Sequential composition of queries with a decaying budget satisfies differential privacy.
The paper invokes a budget allocation over queries, but only cites Dwork [7] without a composition proof for this mechanism.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Privacy-Utility-Fairness: A Balanced Approach to Vehicular-Traffic Management System." pith.science (2026). https://pith.science/paper/KQLQK47X

@misc{pith2026250708864,
  author       = {Pith},
  title        = {Pith review of: Privacy-Utility-Fairness: A Balanced Approach to Vehicular-Traffic Management System},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KQLQK47X}},
  note         = {Machine review of arXiv:2507.08864}
}

read the original abstract

Location-based vehicular traffic management faces significant challenges in protecting sensitive geographical data while maintaining utility for traffic management and fairness across regions. Existing state-of-the-art solutions often fail to meet the required level of protection against linkage attacks and demographic biases, leading to privacy leakage and inequity in data analysis. In this paper, we propose a novel algorithm designed to address the challenges regarding the balance of privacy, utility, and fairness in location-based vehicular traffic management systems. In this context, utility means providing reliable and meaningful traffic information, while fairness ensures that all regions and individuals are treated equitably in data use and decision-making. Employing differential privacy techniques, we enhance data security by integrating query-based data access with iterative shuffling and calibrated noise injection, ensuring that sensitive geographical data remains protected. We ensure adherence to epsilon-differential privacy standards by implementing the Laplace mechanism. We implemented our algorithm on vehicular location-based data from Norway, demonstrating its ability to maintain data utility for traffic management and urban planning while ensuring fair representation of all geographical areas without being overrepresented or underrepresented. Additionally, we have created a heatmap of Norway based on our model, illustrating the privatized and fair representation of the traffic conditions across various cities. Our algorithm provides privacy in vehicular traffic

Figures

Figures reproduced from arXiv: 2507.08864 by the authors.

**Figure 1.** Local and Global Shuffling this batch-wise shuffling technique further reduce variance in group representation, converging to the original proportions in D, thereby achieving group fairness while anonymizing data to mitigate re-identification risks [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Privacy vs Utility 10%, reflecting slower driving speeds and occasional accidents, which could add up to 100 extra vehicles per hour [24]. Snowy conditions lead to a 20% increase, representing greater driving challenges [25]. Conversely, clear weather results in a modest rise of 20 vehicles per hour, reflecting increased travel activity under favourable conditions [26]. The dataset and code are openly available as h… view at source ↗

**Figure 3.** Uniform Application of Noise: Demonstrating unbias [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Traffic congestion heatmap for Norway, showing condi [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: Region-Wise 24-Hour Traffic Predictions: Original v [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

[1]

Zhan, Y uting, Hamed Haddadi, and Afra Mashhadi. ”Analys ing Fair- ness of Privacy-Utility Mobility Models.” Adjunct Proceed ings of the 2023 ACM International Joint Conference on Pervasive and Ub iquitous Computing & the 2023 ACM International Symposium on Wearabl e Computing. 2023

work page 2023
[2]

”Intelligent trafﬁc monitoring system s for vehicle classiﬁcation: A survey.” IEEE Access 8 (2020): 73340-7335 8

Won, Myounggyu. ”Intelligent trafﬁc monitoring system s for vehicle classiﬁcation: A survey.” IEEE Access 8 (2020): 73340-7335 8

work page 2020
[3]

”The year in risk 2017.” Risk Manageme nt 64.11 (2017): 20-25

O’Rourke, Morgan. ”The year in risk 2017.” Risk Manageme nt 64.11 (2017): 20-25. Fig. 4. Trafﬁc congestion heatmap for Norway, showing condi tions at 17:00 and 7:00 across different regions. Fig. 5. Region-Wise 24-Hour Trafﬁc Predictions: Original v s. Noisy densities across Norwegian regions

work page 2017
[4]

Mokbel, and Xuan Liu

Chow, Chi-Yin, Mohamed F. Mokbel, and Xuan Liu. ”Spatial cloaking for anonymous location-based services in mobile peer-to-peer environments.” GeoInformatica 15.2 (2011): 351-380

work page 2011
[5]

”k-anonymity: A model for protecting privacy.” Inter- national journal of uncertainty, fuzziness and knowledge- based systems 10.05 (2002): 557-570

Sweeney, Latanya. ”k-anonymity: A model for protecting privacy.” Inter- national journal of uncertainty, fuzziness and knowledge- based systems 10.05 (2002): 557-570

work page 2002
[6]

”Deep learning for person re-identiﬁcat ion: A survey and outlook.” IEEE transactions on pattern analysis and machin e intelligence 44.6 (2021): 2872-2893

Y e, Mang, et al. ”Deep learning for person re-identiﬁcat ion: A survey and outlook.” IEEE transactions on pattern analysis and machin e intelligence 44.6 (2021): 2872-2893

work page 2021
[7]

”Differential privacy.” Internationa l colloquium on au- tomata, languages, and programming

Dwork, Cynthia. ”Differential privacy.” Internationa l colloquium on au- tomata, languages, and programming. Berlin, Heidelberg: S pringer Berlin Heidelberg, 2006

work page 2006
[8]

Fairness

Francez, Nissim. Fairness. Springer Science & Business Media, 2012

work page 2012

Show all 26 references

[9]

”What’s fair about individual fairness ?.” Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society

Fleisher, Will. ”What’s fair about individual fairness ?.” Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society. 202 1

2021
[10]

”Group fairness: Independence revisited.” Proceedings of the 2021 ACM conference on fairness, accountability, and trans parency

R¨ az, Tim. ”Group fairness: Independence revisited.” Proceedings of the 2021 ACM conference on fairness, accountability, and trans parency. 2021

2021
[11]

”Demographic pari ty: Mitigating biases in real-world data.” arXiv preprint arXiv:2309.173 47 (2023)

Loukas, Orestis, and Ho-Ryun Chung. ”Demographic pari ty: Mitigating biases in real-world data.” arXiv preprint arXiv:2309.173 47 (2023)

2023
[12]

”Learning With Location-Based Fairnes s: A Statistically- Robust Framework and Acceleration.” IEEE Transactions on K nowledge and Data Engineering (2024)

He, Erhu, et al. ”Learning With Location-Based Fairnes s: A Statistically- Robust Framework and Acceleration.” IEEE Transactions on K nowledge and Data Engineering (2024)

2024
[13]

”Differential Privacy in Aggreg ated Mobility Net- works: Balancing Privacy and Utility.” arXiv preprint arXi v:2112.08487 (2021)

Haydari, Ammar, et al. ”Differential Privacy in Aggreg ated Mobility Net- works: Balancing Privacy and Utility.” arXiv preprint arXi v:2112.08487 (2021)

2021 arXiv
[14]

Wang, Han, et al. ”L-srr: Local differential privacy fo r location-based services with staircase randomized response.” Proceeding s of the 2022 ACM SIGSAC Conference on computer and communications secur ity. 2022

2022
[15]

”Classiﬁcation utility, fairn ess, and compactness via tunable information bottleneck and R´ enyi measures.” I EEE Transac- tions on Information Forensics and Security (2023)

Gronowski, Adam, et al. ”Classiﬁcation utility, fairn ess, and compactness via tunable information bottleneck and R´ enyi measures.” I EEE Transac- tions on Information Forensics and Security (2023)

2023
[16]

”Resilient pri vacy protection for location-based services through decentralization.” ACM T ransactions on Privacy and Security (TOPS) 22.4 (2019): 1-36

Jin, Hongyu, and Panos Papadimitratos. ”Resilient pri vacy protection for location-based services through decentralization.” ACM T ransactions on Privacy and Security (TOPS) 22.4 (2019): 1-36

2019
[17]

”A Framework for Tradeoff Between L ocation Privacy Preservation and Quality of Experience in Location Based Se rvices.” IEEE Open Journal of V ehicular Technology (2024)

Feng, Tianyi, et al. ”A Framework for Tradeoff Between L ocation Privacy Preservation and Quality of Experience in Location Based Se rvices.” IEEE Open Journal of V ehicular Technology (2024)

2024
[18]

”Deﬁning privacy and utility in data sets.” U

Wu, Felix T. ”Deﬁning privacy and utility in data sets.” U. Colo. L. Rev. 84 (2013): 1117

2013
[19]

”Referee-Meta-Learning for Fast Ad aptation of Locational Fairness.” Proceedings of the AAAI Conference o n Artiﬁcial Intelligence

Chen, Weiye, et al. ”Referee-Meta-Learning for Fast Ad aptation of Locational Fairness.” Proceedings of the AAAI Conference o n Artiﬁcial Intelligence. V ol. 38. No. 20. 2024

2024
[20]

”BUDS: bal- ancing utility and differential privacy by shufﬂing.” 2020 11th Inter- national Conference on Computing, Communication and Netwo rking Technologies (ICCCNT)

Sengupta, Poushali, Sudipta Paul, and Subhankar Mishr a. ”BUDS: bal- ancing utility and differential privacy by shufﬂing.” 2020 11th Inter- national Conference on Computing, Communication and Netwo rking Technologies (ICCCNT). IEEE, 2020

2020
[21]

”BUDS+: Better Privacy with Converger and Noisy Shufﬂing.” Digital Threats: Research and Practice 4.2 (2023): 1-23

Sengupta, Poushali, Sudipta Paul, and Subhankar Mishr a. ”BUDS+: Better Privacy with Converger and Noisy Shufﬂing.” Digital Threats: Research and Practice 4.2 (2023): 1-23

2023
[22]

Ghinita, Gabriel, et al. ”Preventing velocity-based l inkage attacks in location-aware applications.” Proceedings of the 17th ACM SIGSPA TIAL international conference on advances in geographic inform ation systems. 2009

2009
[23]

Sengupta, Poushali, and Subhankar Mishra. ”Fairly Pri vate Through Group Tagging and Relation Impact.” Modeling Decisions for Artiﬁcial Intelligence: 18th International Conference, MDAI 2021, U me˚ a, Sweden, September 27–30, 2021, Proceedings 18. Springer Internati onal Publi...

2021
[24]

Essien, Aniekan, et al. ”The impact of rainfall and temp erature on peak and off-peak urban trafﬁc.” Database and Expert System s Applica- tions: 29th International Conference, DEXA 2018, Regensbu rg, Germany, September 3–6, 2018, Proceedings, Part II 29. Springer Inte rnation...

2018
[25]

”Dynamic and Non-Linear Analy sis of the Impact of Diurnal Temperature Range on Road Trafﬁc Accidents.” Cli mate 11.10 (2023): 199

Shiau, Y uo-Hsien, et al. ”Dynamic and Non-Linear Analy sis of the Impact of Diurnal Temperature Range on Road Trafﬁc Accidents.” Cli mate 11.10 (2023): 199

2023
[26]

”Examining the effect of adverse w eather on road transportation using weather and trafﬁc sensors.” PLoS one 13.10 (2018): e0205409

Peng, Yichuan, et al. ”Examining the effect of adverse w eather on road transportation using weather and trafﬁc sensors.” PLoS one 13.10 (2018): e0205409

2018

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Zhan, Y uting, Hamed Haddadi, and Afra Mashhadi. ”Analys ing Fair- ness of Privacy-Utility Mobility Models.” Adjunct Proceed ings of the 2023 ACM International Joint Conference on Pervasive and Ub iquitous Computing & the 2023 ACM International Symposium on Wearabl e Computing. 2023

work page 2023

[2] [2]

”Intelligent trafﬁc monitoring system s for vehicle classiﬁcation: A survey.” IEEE Access 8 (2020): 73340-7335 8

Won, Myounggyu. ”Intelligent trafﬁc monitoring system s for vehicle classiﬁcation: A survey.” IEEE Access 8 (2020): 73340-7335 8

work page 2020

[3] [3]

”The year in risk 2017.” Risk Manageme nt 64.11 (2017): 20-25

O’Rourke, Morgan. ”The year in risk 2017.” Risk Manageme nt 64.11 (2017): 20-25. Fig. 4. Trafﬁc congestion heatmap for Norway, showing condi tions at 17:00 and 7:00 across different regions. Fig. 5. Region-Wise 24-Hour Trafﬁc Predictions: Original v s. Noisy densities across Norwegian regions

work page 2017

[4] [4]

Mokbel, and Xuan Liu

Chow, Chi-Yin, Mohamed F. Mokbel, and Xuan Liu. ”Spatial cloaking for anonymous location-based services in mobile peer-to-peer environments.” GeoInformatica 15.2 (2011): 351-380

work page 2011

[5] [5]

”k-anonymity: A model for protecting privacy.” Inter- national journal of uncertainty, fuzziness and knowledge- based systems 10.05 (2002): 557-570

Sweeney, Latanya. ”k-anonymity: A model for protecting privacy.” Inter- national journal of uncertainty, fuzziness and knowledge- based systems 10.05 (2002): 557-570

work page 2002

[6] [6]

”Deep learning for person re-identiﬁcat ion: A survey and outlook.” IEEE transactions on pattern analysis and machin e intelligence 44.6 (2021): 2872-2893

Y e, Mang, et al. ”Deep learning for person re-identiﬁcat ion: A survey and outlook.” IEEE transactions on pattern analysis and machin e intelligence 44.6 (2021): 2872-2893

work page 2021

[7] [7]

”Differential privacy.” Internationa l colloquium on au- tomata, languages, and programming

Dwork, Cynthia. ”Differential privacy.” Internationa l colloquium on au- tomata, languages, and programming. Berlin, Heidelberg: S pringer Berlin Heidelberg, 2006

work page 2006

[8] [8]

Fairness

Francez, Nissim. Fairness. Springer Science & Business Media, 2012

work page 2012