MétaCan
Menu
Back to cohort
Record W4376626823 · doi:10.56553/popets-2023-0083

Losing Less: A Loss for Differentially Private Deep Learning

2023· article· en· W4376626823 on OpenAlex

Why this work is in the frame

A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.

affAt least one author lists a Canadian institution in the pinned OpenAlex snapshot.
fundA Canadian funder is recorded on the work.

Bibliographic record

VenueProceedings on Privacy Enhancing Technologies · 2023
Typearticle
Languageen
FieldComputer Science
TopicPrivacy-Preserving Technologies in Data
Canadian institutionsVector InstituteUniversity of Toronto
FundersNatural Sciences and Engineering Research Council of CanadaGovernment of CanadaCanadian Institute for Advanced ResearchVector InstituteAlfred P. Sloan Foundation
KeywordsStochastic gradient descentMNIST databaseComputer scienceDeep learningArtificial intelligenceCross entropyGradient descentMachine learningArtificial neural networkAlgorithmPattern recognition (psychology)

Abstract

fetched live from OpenAlex

Differentially Private Stochastic Gradient Descent, DP-SGD, is the canonical approach to training deep neural networks with guarantees of Differential Privacy (DP). However, the modifications DP-SGD introduces to vanilla gradient descent negatively impact the accuracy of deep neural networks. In this paper, we are the first to observe that some of this performance can be recovered when training with a loss tailored to DP-SGD; we challenge cross-entropy as the de facto loss for deep learning with DP. Specifically, we introduce a loss combining three terms: the summed squared error, the focal loss, and a regularization penalty. The first term encourages learning with faster convergence. The second term emphasizes hard-to-learn examples in the later stages of training. Both are beneficial because the privacy cost of learning increases with every step of DP-SGD. The third term helps control the sensitivity of learning, decreasing the bias introduced by gradient clipping in DP-SGD. Using our loss function, we achieve new state-of-the-art tradeoffs between privacy and accuracy on MNIST, FashionMNIST, and CIFAR10. Most importantly, we improve the accuracy of DP-SGD on CIFAR10 by 4% for a DP guarantee of 𝜀 = 3.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.001
metaresearch head score (Gemma)0.080
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesMetaresearch, Meta-epidemiology (narrow), Open science
Consensus categoriesOpen science
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Theoretical or conceptual · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.649
Threshold uncertainty score1.000

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0010.080
Meta-epidemiology (narrow)0.0010.001
Meta-epidemiology (broad)0.0010.000
Bibliometrics0.0010.003
Science and technology studies0.0010.000
Scholarly communication0.0010.001
Open science0.0340.079
Research integrity0.0010.001
Insufficient payload (model declined to judge)0.0000.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.031
GPT teacher head0.277
Teacher spread0.246 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it