Jump to Content
Shuang Song

Shuang Song

I work on differential privacy and machine learning. Prior to joining Google, I received my PhD from UC San Diego, advised by Prof. Kamalika Chaudhuri. My webpage.
Authored Publications
Google Publications
Other Publications
Sort By
  • Title
  • Title, desc
  • Year
  • Year, desc
    EANA: Reducing Privacy Risk on Large-scale Recommendation Models
    Devora Berlowitz
    Mei Chen
    QiQi Xue
    Steve Chien
    16th ACM Conference on Recommender Systems (2022)
    Preview abstract Embedding-based deep neural networks (DNNs) are widely used in large-scale recommendation systems. Differentially-private stochastic gradient descent (DP-SGD) provides a way to enable personalized experiences while preserving user privacy by injecting noise into every model parameter during the training process. However, it is challenging to apply DP-SGD to large-scale embedding-based DNNs due to its effect on training speed. This happens because the noise added by DP-SGD causes normally sparse gradients to become dense, introducing a large communication overhead between workers and parameter servers in a typical distributed training framework. This paper proposes embedding-aware noise addition (EANA) to mitigate the communication overhead, making training a large-scale embedding-based DNN possible. We examine the privacy benefit of EANA both analytically and empirically using secret sharer techniques. We demonstrate that training with EANA can achieve reasonable model precision while providing good practical privacy protection as measured by the secret sharer tests. Experiments on a real-world, large-scale dataset and model show that EANA is much faster than standard DP-SGD, improving the training speed by 54X and unblocking the training of a large-scale embedding-based DNN with reduced privacy risk. View details
    Preview abstract We study personalization of supervised learning with user-level differential privacy. Consider a setting with many users, each of whom has a training data set drawn from their own distribution $P_i$. Assuming some shared structure among the problems $P_i$, can users collectively learn the shared structure---and solve their tasks better than they could individually---while preserving the privacy of their data? We formulate this question using joint, \textit{user-level} differential privacy---that is, we control what is leaked about each user's entire data set. We provide algorithms that exploit popular non-private approaches in this domain like the Almost-No-Inner-Loop (ANIL) method, and give strong user-level privacy guarantees for our general approach. When the problems $P_i$ are linear regression problems with each user's regression vector lying in a common, unknown low-dimensional subspace, we show that our efficient algorithms satisfy nearly optimal estimation error guarantees. We also establish a general, information-theoretic upper bound via an exponential mechanism-based algorithm. Finally, we demonstrate empirically (through experiments on synthetic data sets) that our framework not only performs well in the studied linear regression setting, but also extends to other settings like logistic regression that are not captured by our estimation error analysis. View details
    Preview abstract We study the problem of differentially private (DP) matrix completion under user-level privacy. We design an $(\epsilon,\delta)$-joint differentially private variant of the popular Alternating-Least-Squares (ALS) method that achieves: i) (nearly) optimal sample complexity for matrix completion (in terms of number of items, users), and ii) best known privacy/utility trade-off both theoretically, as well as on benchmark data sets. In particular, despite non-convexity of low-rank matrix completion and ALS, we provide the first global convergence analysis of ALS with {\em noise} introduced to ensure DP. For $n$ being the number of users and $m$ being the number of items in the rating matrix, our analysis requires only about $\log (n+m)$ samples per user (ignoring rank, condition number factors) and obtains a sample complexity of $n=\tilde\Omega(m/(\sqrt{\zeta}\cdot \epsilon))$ to ensure relative Frobenius norm error of $\zeta$. This improves significantly on the previous state of the result of $n=\tilde\Omega\left(m^{5/4}/(\zeta^{5}\epsilon)\right)$ for the private-FW method by ~\citet{jain2018differentially}. Furthermore, we extensively validate our method on synthetic and benchmark data sets (MovieLens 10mi, MovieLens 20mi), and observe that private ALS only suffers a 6 percentage drop in accuracy when compared to the non-private baseline for $\epsilon\leq 10$. Furthermore, compared to prior work of~\cite{jain2018differentially}, it is at least better by 10 percentage for all choice of the privacy parameters. View details
    Evading the Curse of Dimensionality in Unconstrained Private Generalized Linear Problems
    Thomas Steinke
    Abhradeep Guha Thakurta
    24th International Conference on Artificial Intelligence and Statistics (AISTATS 2021) (2020)
    Preview abstract Differentially private gradient descent (DP-GD) has been extremely effective both theoretically, and in practice, for solving private empirical risk minimization (ERM) problems. In this paper, we focus on understanding the impact of the clipping norm, a critical component of DP-GD, on its convergence. We provide the first formal convergence analysis of clipped DP-GD. More generally, we show that the value which one sets for clipping really matters: done wrong, it can dramatically affect the resulting quality; done properly, it can eliminate the dependence of convergence on the model dimensionality. We do this by showing a dichotomous behavior of the clipping norm. First, we show that if the clipping norm is set smaller than the optimal, even by a constant factor, the excess empirical risk for convex ERMs can increase from $O(1/n)$ to $\Omega(1)$, where $n$ is the number of data samples. Next, we show that, regardless of the value of the clipping norm, clipped DP-GD minimizes a well-defined convex objective over an unconstrained space, as long as the underlying ERM is a generalized linear problem. Furthermore, if the clipping norm is set within at most a constant factor higher than the optimal, then one can obtain an excess empirical risk guarantee that is independent of the dimensionality of the model space. Finally, we extend our result to non-convex generalized linear problems by showing that DP-GD reaches a first-order stationary point as long as the loss is smooth, and the convergence is independent of the dimensionality of the model space. View details
    Scalable Private Learning with PATE
    Ilya Mironov
    Ananth Raghunathan
    Kunal Talwar
    Úlfar Erlingsson
    International Conference on Learning Representations (ICLR) (2018)
    Preview abstract The rapid adoption of machine learning has increased concerns about the privacy implications of machine learning models trained on sensitive data, such as medical records or other personal information. To address those concerns, one promising approach is Private Aggregation of Teacher Ensembles, or PATE, which transfers to a "student" model the knowledge of an ensemble of "teacher" models, with intuitive privacy provided by training teachers on disjoint data and strong privacy guaranteed by noisy aggregation of teachers’ answers. However, PATE has so far been evaluated only on simple classification tasks like MNIST, leaving unclear its utility when applied to larger-scale learning tasks and real-world datasets. In this work, we show how PATE can scale to learning tasks with large numbers of output classes and uncurated, imbalanced training data with errors. For this, we introduce new noisy aggregation mechanisms for teacher ensembles that are more selective and add less noise, and prove their tighter differential-privacy guarantees. Our new mechanisms build on two insights: the chance of teacher consensus is increased by using more concentrated noise and, lacking consensus, no answer need be given to a student. The consensus answers used are more likely to be correct, offer better intuitive privacy, and incur lower-differential privacy cost. Our evaluation shows our mechanisms improve on the original PATE on all measures, and scale to larger tasks with both high utility and very strong privacy (ε < 1.0). View details
    No Results Found