AI Paper Atlas

Research map / Federated learning and privacy / Differential privacy

Two-mode Networks: Inference with as Many Parameters as Actors and Differential Privacy

Qiuping Wang, Ting Yan, Binyan Jiang, Chenlei Leng

JMLR 2022 · Federated learning and privacy › Differential privacy · approach: Privacy accounting and mechanisms

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use the denoised estimator instead of the moment-based estimatordifferentially private bipartite beta-model estimation · Simulation, Table 195% CI length: 1.21 -> 1.20Improved
increase the parameter magnitude c from 0.1 to 0.3moment estimator · Simulation (m,n)=(50,100), eps_n=log n/n^1/695% CI length: 1.21 -> 1.87Hurt
increase the parameter magnitude c from 0.1 to 0.3moment estimator · Simulation (m,n)=(50,100), eps_n=log n/n^1/6frequency estimator did not exist (x100%): 0 -> 25.54Hurt
increase the sample size from (m,n)=(50,100) to (100,200)moment estimator · Simulation pair (1,2), eps_n=log n/n^1/6, c=0.195% CI length: 1.21 -> 0.86Improved
increase the sample size from (m,n)=(50,100) to (100,200)moment estimator · Simulation pair (1,2), eps_n=log n/n^1/6, c=0.3frequency estimator did not exist (x100%): 25.54 -> 3.14Improved
use weaker privacy noise eps_n=log n/n^1/4 instead of log n/n^1/6moment estimator · Simulation (m,n)=(50,100), pair (1,2), c=0.3frequency estimator did not exist (x100%): 25.54 -> 60.43Hurt
use eps_n=log n/n^1/4 instead of log n/n^1/6moment estimator · Simulation (m,n)=(100,200), pair (1,2), c=0.395% CI coverage frequency (x100%): 94.08 -> 91.93Hurt
shrink the number of actor-side nodes to m=[n^1/2] (n=200,500,1000)moment and denoised estimators · Additional simulationestimator existence: failed to exist with probabilities larger than 96%Hurt
use m=[n^3/4] with ||theta||_inf <= 0.2 log nmoment and denoised estimators · Additional simulationestimator failure frequency: less than 4%No clear effect
apply denoising algorithm to the sparse UC Irvine forum networkdenoised estimator · UC Irvine forum network (899 students, 522 topics, 7089 edges)estimator existence: non-existent every time in 1,000 releasesHurt
increase epsilon (privacy budget) in the forum data analysismoment estimator · UC Irvine forum network95% CI lengthImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy