An Entropy Approach to Disclosure Risk Assessment: Lessons from Real Applications and Simulated Domains

A scaling law to model the effectiveness of identification techniques

AI techniques are increasingly being used to identify individuals both offline and online. However, quantifying their effectiveness at scale and, by extension, the risks they pose remains a significant challenge. Here, we propose a two-parameter Bayesian model for exact matching techniques and derive an analytical expression for correctness (κ), the fraction of people accurately identified in a population. We then generalize the model to forecast how κ scales from small-scale experiments to the real world, for exact, sparse, and machine learning-based robust identification techniques. Despite having only two degrees of freedom, our method closely fits 476 correctness curves and strongly outperforms curve-fitting methods and entropy-based rules of thumb. Our work provides a principled framework for forecasting the privacy risks posed by identification techniques, while also supporting independent accountability efforts for AI-based biometric systems.

Evaluating information loss in the National Cancer Database from cases lost to follow-up

BACKGROUND AND OBJECTIVES

Cancer registries must focus on data capture which returns value while reducing resource burden with minimal loss of data. Identifying the optimum length of follow-up data collection for patients with cancer achieves this goal.

METHODS

A two-step analysis using entropy calculations to assess information gain for each follow-up year, and second-order differences to compare survival outcomes between the defined follow-up periods and lifetime follow-up. A total of 391 567 adult cases, deidentified in the National Cancer Database and diagnosed in 1989. Comparisons examined a subset of 61 908 lung cancer cases, 48 387 colon and rectal cancer cases, and 64 134 breast cancer cases in adults. A total of 4133 pediatric cases were diagnosed in 1989 examining 1065 leukemia cases and 494 lymphoma cases.

RESULTS

Annual increases in information gain fell below 1% after 16 years of follow-up for adult cases and 9 years for pediatric cases. Comparison of second-order differences showed 62% of the comparisons were similar between 15 years and lifetime follow-up when examining restricted mean survival time. In addition, 90% of the comparisons were statistically similar when comparing hazard ratios.

CONCLUSIONS

Survival analysis using 15 years postdiagnosis follow-up showed minimal differences in information gain compared to lifetime follow-up.

An algorithmic-based multi-attribute decision making model under intuitionistic fuzzy environment

An outranking procedure for Multi-Attribute Decision-Making (MADM) problems is introduced in our work that acts as a decision-aid in recommending the products to the buyers. The buyer’s product assessment is taken as Interval-Valued Intuitionistic Fuzzy Sets (IVIFS) in each attribute. The confidence level that is implicit in the buyer’s product rating is explicated in the proposed work using fuzzy entropy. As the confidence level of the buyer on the product assessment is for both satisfaction and reluctance, it is suitably distributed in membership and non-membership parts of IVIFS. Our work generates a dominance matrix that represents partial or full dominance of one product over another after scoring the products that are unified with buyer’s confidence. The proposed work suggests the product ranking after ascertaining the buyer’s flexibility. An algorithm is written in our work to validate the procedure developed. We have compared our work with other similar works to highlight the benefits of the proposed work. A numerical example is illustrated to highlight the procedure developed.