Searcharxiv⌕ Search

arXiv subjects

Yuqing Zhu

Publications and source records attributed to Yuqing Zhu.

At least 37 records · Page 2Linked to original sources
arXiv2023

"Private Prediction Strikes Back!'' Private Kernelized Nearest Neighbors with Individual Renyi Filter

Most existing approaches of differentially private (DP) machine learning focus on private training. Despite its many advantages, private training lacks the flexibility in adapting to incremental changes to the training dataset such as deletion requests from exercising GDPR's right to be forgotten. We revisit a long-forgotten alternative, known as private prediction, and propose a new algorithm named Individual Kernelized Nearest Neighbor (Ind-KNN). Ind-KNN is easily updatable over dataset changes and it allows precise control of the Rényi DP at an individual user level -- a user's privacy loss is measured by the exact amount of her contribution to predictions; and a user is removed if her prescribed privacy budget runs out. Our results show that Ind-KNN consistently improves the accuracy over existing private prediction methods for a wide range of $ε$ on four vision and language tasks. We also illustrate several cases under which Ind-KNN is preferable over private training with NoisySGD.

cs.LG↗
arXiv2022

Generalized PTR: User-Friendly Recipes for Data-Adaptive Algorithms with Differential Privacy

The ''Propose-Test-Release'' (PTR) framework is a classic recipe for designing differentially private (DP) algorithms that are data-adaptive, i.e. those that add less noise when the input dataset is nice. We extend PTR to a more general setting by privately testing data-dependent privacy losses rather than local sensitivity, hence making it applicable beyond the standard noise-adding mechanisms, e.g. to queries with unbounded or undefined sensitivity. We demonstrate the versatility of generalized PTR using private linear regression as a case study. Additionally, we apply our algorithm to solve an open problem from ''Private Aggregation of Teacher Ensembles (PATE)'' -- privately releasing the entire model with a delicate data-dependent analysis.

cs.LG↗
arXiv2022

Schottky Electric Field Induced Circular Photogalvanic Effect in Cd3As2 Nanobelts

Dirac semimetals are expected to forbid the manifestation of the circular photogalvanic effect (CPGE) because of their crystal inversion symmetry. Here, we report the observation of the CPGE in Cd3As2 nanobelt field effect transistors, when the photoexcitation is focused in the vicinity of the metal contacts up to room temperature. We attribute the CPGE to the Schottky electric field induced symmetry breaking, which results in the photocurrent modulation by circularly polarized photoexcitation via spin-momentum locking. The hypothesis is supported by a suite of experiments including spatially and angularly resolved helicity dependent photocurrent, Kelvin probe force microscopy, and gate voltage dependence. First principles calculations confirmed a topological phase transition upon field induced structural distortion. This work provides key insights on the electrically controlled helicity dependent optoelectronics in Dirac materials.

cond-mat.mes-hall↗
arXiv2022

Optimal Accounting of Differential Privacy via Characteristic Function

Characterizing the privacy degradation over compositions, i.e., privacy accounting, is a fundamental topic in differential privacy (DP) with many applications to differentially private machine learning and federated learning. We propose a unification of recent advances (Renyi DP, privacy profiles, $f$-DP and the PLD formalism) via the \emph{characteristic function} ($ϕ$-function) of a certain \emph{dominating} privacy loss random variable. We show that our approach allows \emph{natural} adaptive composition like Renyi DP, provides \emph{exactly tight} privacy accounting like PLD, and can be (often \emph{losslessly}) converted to privacy profile and $f$-DP, thus providing $(ε,δ)$-DP guarantees and interpretable tradeoff functions. Algorithmically, we propose an \emph{analytical Fourier accountant} that represents the \emph{complex} logarithm of $ϕ$-functions symbolically and uses Gaussian quadrature for numerical computation. On several popular DP mechanisms and their subsampled counterparts, we demonstrate the flexibility and tightness of our approach in theory and experiments.

cs.LG↗
arXiv2022

Adaptive Private-K-Selection with Adaptive K and Application to Multi-label PATE

We provide an end-to-end Renyi DP based-framework for differentially private top-$k$ selection. Unlike previous approaches, which require a data-independent choice on $k$, we propose to privately release a data-dependent choice of $k$ such that the gap between $k$-th and the $(k+1)$st "quality" is large. This is achieved by a novel application of the Report-Noisy-Max. Not only does this eliminate one hyperparameter, the adaptive choice of $k$ also certifies the stability of the top-$k$ indices in the unordered set so we can release them using a variant of propose-test-release (PTR) without adding noise. We show that our construction improves the privacy-utility trade-offs compared to the previous top-$k$ selection algorithms theoretically and empirically. Additionally, we apply our algorithm to "Private Aggregation of Teacher Ensembles (PATE)" in multi-label classification tasks with a large number of labels and show that it leads to significant performance gains.

cs.LG↗
arXiv2022

Revisiting Model-Agnostic Private Learning: Faster Rates and Active Learning

The Private Aggregation of Teacher Ensembles (PATE) framework is one of the most promising recent approaches in differentially private learning. Existing theoretical analysis shows that PATE consistently learns any VC-classes in the realizable setting, but falls short in explaining its success in more general cases where the error rate of the optimal classifier is bounded away from zero. We fill in this gap by introducing the Tsybakov Noise Condition (TNC) and establish stronger and more interpretable learning bounds. These bounds provide new insights into when PATE works and improve over existing results even in the narrower realizable setting. We also investigate the compelling idea of using active learning for saving privacy budget, and empirical studies show the effectiveness of this new idea. The novel components in the proofs include a more refined analysis of the majority voting classifier - which could be of independent interest - and an observation that the synthetic "student" learning problem is nearly realizable by construction under the Tsybakov noise condition.

cs.LG↗
arXiv2021

IoTDataBench: Extending TPCx-IoT for Compression and Scalability

We present a record-breaking result and lessons learned in practicing TPCx-IoT benchmarking for a real-world use case. We find that more system characteristics need to be benchmarked for its application to real-world use cases. We introduce an extension to the TPCx-IoT benchmark, covering fundamental requirements of time-series data management for IoT infrastructure. We characterize them as data compression and system scalability. To evaluate these two important features of IoT databases, we propose IoTDataBench and update four aspects of TPCx-IoT, i.e., data generation, workloads, metrics and test procedures. Preliminary evaluation results show systems that fail to effectively compress data or flexibly scale can negatively affect the redesigned metrics, while systems with high compression ratios and linear scalability are rewarded in the final metrics. Such systems have the ability to scale up computing resources on demand and can thus save dollar costs.

cs.DB↗
arXiv2021

Voting-based Approaches For Differentially Private Federated Learning

Differentially Private Federated Learning (DPFL) is an emerging field with many applications. Gradient averaging based DPFL methods require costly communication rounds and hardly work with large-capacity models, due to the explicit dimension dependence in its added noise. In this work, inspired by knowledge transfer non-federated privacy learning from Papernot et al.(2017; 2018), we design two new DPFL schemes, by voting among the data labels returned from each local model, instead of averaging the gradients, which avoids the dimension dependence and significantly reduces the communication cost. Theoretically, by applying secure multi-party computation, we could exponentially amplify the (data-dependent) privacy guarantees when the margin of the voting scores are large. Extensive experiments show that our approaches significantly improve the privacy-utility trade-off over the state-of-the-arts in DPFL.

cs.LG↗
arXiv2021

JITuNE: Just-In-Time Hyperparameter Tuning for Network Embedding Algorithms

Network embedding (NE) can generate succinct node representations for massive-scale networks and enable direct applications of common machine learning methods to the network structure. Various NE algorithms have been proposed and used in a number of applications, such as node classification and link prediction. NE algorithms typically contain hyperparameters that are key to performance, but the hyperparameter tuning process can be time consuming. It is desirable to have the hyperparameters tuned within a specified length of time. Although AutoML methods have been applied to the hyperparameter tuning of NE algorithms, the problem of how to tune hyperparameters in a given period of time is not studied for NE algorithms before. In this paper, we propose JITuNE, a just-in-time hyperparameter tuning framework for NE algorithms. Our JITuNE framework enables the time-constrained hyperparameter tuning for NE algorithms by employing the tuning over hierarchical network synopses and transferring the knowledge obtained on synopses to the whole network. The hierarchical generation of synopsis and a time-constrained tuning method enable the constraining of overall tuning time. Extensive experiments demonstrate that JITuNE can significantly improve performances of NE algorithms, outperforming state-of-the-art methods within the same number of algorithm runs.

cs.LG↗
arXiv2020

When Load Rebalancing Does Not Work for Distributed Hash Table

Distributed hash table (DHT) is the foundation of many widely used storage systems, for its prominent features of high scalability and load balancing. Recently, DHT-based systems have been deployed for the Internet-of-Things (IoT) application scenarios. Unfortunately, such systems can experience a breakdown in the scale-out and load rebalancing process. This phenomenon contradicts with the common conception of DHT systems, especially about its scalability and load balancing features. In this paper, we investigate the breakdown of DHT-based systems in the scale-out process. We formulate the load rebalancing problem of DHT by considering the impacts of write workloads and data movement. We show that, the average network bandwidth of each node and the intensity of the average write workload are the two key factors that determine the feasibility of DHT load rebalancing. We theoretically prove that load rebalancing is not feasible for a large DHT system under heavy write workloads in a node-by-node scale-out process.

cs.DC↗
arXiv2019

ClassyTune: A Performance Auto-Tuner for Systems in the Cloud

Performance tuning can improve the system performance and thus enable the reduction of cloud computing resources needed to support an application. Due to the ever increasing number of parameters and complexity of systems, there is a necessity to automate performance tuning for the complicated systems in the cloud. The state-of-the-art tuning methods are adopting either the experience-driven tuning approach or the data-driven one. Data-driven tuning is attracting increasing attentions, as it has wider applicability. But existing data-driven methods cannot fully address the challenges of sample scarcity and high dimensionality simultaneously. We present ClassyTune, a data-driven automatic configuration tuning tool for cloud systems. ClassyTune exploits the machine learning model of classification for auto-tuning. This exploitation enables the induction of more training samples without increasing the input dimension. Experiments on seven popular systems in the cloud show that ClassyTune can effectively tune system performance to seven times higher for high-dimensional configuration space, outperforming expert tuning and the state-of-the-art auto-tuning solutions. We also describe a use case in which performance tuning enables the reduction of 33% computing resources needed to run an online stateless service.

cs.PF↗
arXiv2017

BigOP: Generating Comprehensive Big Data Workloads as a Benchmarking Framework

Big Data is considered proprietary asset of companies, organizations, and even nations. Turning big data into real treasure requires the support of big data systems. A variety of commercial and open source products have been unleashed for big data storage and processing. While big data users are facing the choice of which system best suits their needs, big data system developers are facing the question of how to evaluate their systems with regard to general big data processing needs. System benchmarking is the classic way of meeting the above demands. However, existent big data benchmarks either fail to represent the variety of big data processing requirements, or target only one specific platform, e.g. Hadoop. In this paper, with our industrial partners, we present BigOP, an end-to-end system benchmarking framework, featuring the abstraction of representative Operation sets, workload Patterns, and prescribed tests. BigOP is part of an open-source big data benchmarking project, BigDataBench. BigOP's abstraction model not only guides the development of BigDataBench, but also enables automatic generation of tests with comprehensive workloads. We illustrate the feasibility of BigOP by implementing an automatic test generation tool and benchmarking against three widely used big data processing systems, i.e. Hadoop, Spark and MySQL Cluster. Three tests targeting three different application scenarios are prescribed. The tests involve relational data, text data and graph data, as well as all operations and workload patterns. We report results following test specifications.

cs.DC↗
arXiv2017

BestConfig: Tapping the Performance Potential of Systems via Automatic Configuration Tuning

An ever increasing number of configuration parameters are provided to system users. But many users have used one configuration setting across different workloads, leaving untapped the performance potential of systems. A good configuration setting can greatly improve the performance of a deployed system under certain workloads. But with tens or hundreds of parameters, it becomes a highly costly task to decide which configuration setting leads to the best performance. While such task requires the strong expertise in both the system and the application, users commonly lack such expertise. To help users tap the performance potential of systems, we present BestConfig, a system for automatically finding a best configuration setting within a resource limit for a deployed system under a given application workload. BestConfig is designed with an extensible architecture to automate the configuration tuning for general systems. To tune system configurations within a resource limit, we propose the divide-and-diverge sampling method and the recursive bound-and-search algorithm. BestConfig can improve the throughput of Tomcat by 75%, that of Cassandra by 63%, that of MySQL by 430%, and reduce the running time of Hive join job by about 50% and that of Spark join job by about 80%, solely by configuration adjustment.

cs.PF↗
arXiv2017

ACTS in Need: Automatic Configuration Tuning with Scalability Guarantees

To support the variety of Big Data use cases, many Big Data related systems expose a large number of user-specifiable configuration parameters. Highlighted in our experiments, a MySQL deployment with well-tuned configuration parameters achieves a peak throughput as 12 times much as one with the default setting. However, finding the best setting for the tens or hundreds of configuration parameters is mission impossible for ordinary users. Worse still, many Big Data applications require the support of multiple systems co-deployed in the same cluster. As these co-deployed systems can interact to affect the overall performance, they must be tuned together. Automatic configuration tuning with scalability guarantees (ACTS) is in need to help system users. Solutions to ACTS must scale to various systems, workloads, deployments, parameters and resource limits. Proposing and implementing an ACTS solution, we demonstrate that ACTS can benefit users not only in improving system performance and resource utilization, but also in saving costs and enabling fairer benchmarking.

cs.DC↗
arXiv2017

ACIA, not ACID: Conditions, Properties and Challenges

Although ACID is the previous golden rule for transaction support, durability is now not a basic requirement for data storage. Rather, high availability is becoming the first-class property required by online applications. We show that high availability of data is almost surely a stronger property than durability. We thus propose ACIA (Atomicity, Consistency, Isolation, Availability) as the new standard for transaction support. Essentially, the shift from ACID to ACIA is due to the change of assumed conditions for data management. Four major condition changes exist. With ACIA transactions, more diverse requirements can be flexibly supported for applications through the specification of consistency levels, isolation levels and fault tolerance levels. Clarifying the ACIA properties enables the exploitation of techniques used for ACID transactions, as well as bringing about new challenges for research.

cs.DC↗
arXiv2017

To Vote Before Decide: A Logless One-Phase Commit Protocol for Highly-Available Datastores

Highly-available datastores are widely deployed for online applications. However, many online applications are not contented with the simple data access interface currently provided by highly-available datastores. Distributed transaction support is demanded by applications such as large-scale online payment used by Alipay or Paypal. Current solutions to distributed transaction can spend more than half of the whole transaction processing time in distributed commit. An efficient atomic commit protocol is highly desirable. This paper presents the HACommit protocol, a logless one-phase commit protocol for highly-available systems. HACommit has transaction participants vote for a commit before the client decides to commit or abort the transaction; in comparison, the state-of-the-art practice for distributed commit is to have the client decide before participants vote. The change enables the removal of both the participant logging and the coordinator logging steps in the distributed commit process; it also makes possible that, after the client initiates the transaction commit, the transaction data is visible to other transactions within one communication roundtrip time (i.e., one phase). In the evaluation with extensive experiments, HACommit outperforms recent atomic commit solutions for highly-available datastores under different workloads. In the best case, HACommit can commit in one fifth of the time 2PC does.

cs.DC↗
arXiv2015

10 Observations on Google Cluster Trace + 2 Measures for Cluster Utilization Enhancement

Utilization enhancement is a key concern to cluster owners. Google's cluster manager named Borg manages its clusters at an overall utilization higher than many others' clusters. Recently, Google has disclosed the details of its powerful cluster manager Borg. Quite a few lessons are summarized from the Borg experiences. Nevertheless, we find that more can be learned if the Borg design is correlated with the trace analysis of a Google cluster managed by Borg. There is one such trace released four years ago. In this paper, we analyze the Google cluster trace and make 10 observations not found in previous analyses. We also correlates the results of our analysis and previous analyses to the Borg design, such that we find two measures that can possibly further improve cluster utilization over Borg.

cs.DC↗