Search

Zhilin Zhang,1,2 Aldrin D Calderon,1 Xingyu Huang,1 Axin Huang1 1School of Mechanical, Manufacturing and Energy Engineering, Mapúa University, Manila, 0900, Philippines; 2School of Physics and Telecommunications Engineering, Yulin Normal University, Yulin, 537000, People’s Republic of ChinaCorrespondence: Aldrin D Calderon, Email adcalderon@mapua.edu.phAbstract: About 80% of stroke patients have hand motor dysfunction, and wearing finger rehabilitation machinery can enable patients to carry out efficient passive rehabilitation training independently. At present, many typical finger rehabilitation machines have been developed, and clinical experiments have confirmed the effectiveness of mechanically assisted finger rehabilitation. In this paper, the finger rehabilitation machinery will be classified in the actuation mode, and the terminal traction drive/motor drive/spring drive/rope drive/memory alloy drive/electroactive material drive/hydraulic drive/pneumatic drive technology and its typical applications are analyzed. Study the structure, control methods, overlap between mechanical bending nodes and finger joints, training modes, response speed, and driving force of various types of finger rehabilitation machinery. The advantages and disadvantages of various actuation methods of finger rehabilitation machinery are summarized. Finally, the difficulties and opportunities faced by the future development of finger rehabilitation machinery are prospected. In general, with the continuous improvement of quality of life, stroke patients need flexible, segmented control, accurate bending, multi-training mode, fast response, and good driving force finger rehabilitation machinery. This will also be a future hot research direction.Keywords: fingers, machinery, rehabilitation, actuation

Online bidding and auction are crucial aspects of the online advertising industry. Conventionally, there is only one slot for ad display and most current studies focus on it. Nowadays, multi-slot display advertising is gradually becoming popular where many ads could be displayed in a list and shown as a whole to users. However, multi-slot display advertising leads to different cost-effectiveness. Advertisers have the incentive to adjust bid prices so as to win the most economical ad positions. In this study, we introduce bid shading into multi-slot display advertising for bid price adjustment with a Multi-task End-to-end Bid Shading(MEBS) method. We prove the optimality of our method theoretically and examine its performance experimentally. Through extensive offline and online experiments, we demonstrate the effectiveness and efficiency of our method, and we obtain a 7.01% lift in Gross Merchandise Volume, a 7.42% lift in Return on Investment, and a 3.26% lift in ad buy count.

In online advertising, advertisers participate in ad auctions to acquire ad opportunities, often by utilizing auto-bidding tools provided by demand-side platforms (DSPs). The current auto-bidding algorithms typically employ reinforcement learning (RL). However, due to safety concerns, most RL-based auto-bidding policies are trained in simulation, leading to a performance degradation when deployed in online environments. To narrow this gap, we can deploy multiple auto-bidding agents in parallel to collect a large interaction dataset. Offline RL algorithms can then be utilized to train a new policy. The trained policy can subsequently be deployed for further data collection, resulting in an iterative training framework, which we refer to as iterative offline RL. In this work, we identify the performance bottleneck of this iterative offline RL framework, which originates from the ineffective exploration and exploitation caused by the inherent conservatism of offline RL algorithms. To overcome this bottleneck, we propose Trajectory-wise Exploration and Exploitation (TEE), which introduces a novel data collecting and data utilization method for iterative offline RL from a trajectory perspective. Furthermore, to ensure the safety of online exploration while preserving the dataset quality for TEE, we propose Safe Exploration by Adaptive Action Selection (SEAS). Both offline experiments and real-world experiments on Alibaba display advertising platform demonstrate the effectiveness of our proposed method., Comment: Accepted by The Web Conference 2024 (WWW'24) as an oral paper

Automated auction design seeks to discover empirically high-revenue and incentive-compatible mechanisms using machine learning. Ensuring dominant strategy incentive compatibility (DSIC) is crucial, and the most effective approach is to confine the mechanism to Affine Maximizer Auctions (AMAs). Nevertheless, existing AMA-based approaches encounter challenges such as scalability issues (arising from combinatorial candidate allocations) and the non-differentiability of revenue. In this paper, to achieve a scalable AMA-based method, we further restrict the auction mechanism to Virtual Valuations Combinatorial Auctions (VVCAs), a subset of AMAs with significantly fewer parameters. Initially, we employ a parallelizable dynamic programming algorithm to compute the winning allocation of a VVCA. Subsequently, we propose a novel optimization method that combines both zeroth-order and first-order techniques to optimize the VVCA parameters. Extensive experiments demonstrate the efficacy and scalability of our proposed approach, termed Zeroth-order and First-order Optimization of VVCAs (ZFO-VVCA), particularly when applied to large-scale auctions.

Auto-bidding plays a crucial role in facilitating online advertising by automatically providing bids for advertisers. Reinforcement learning (RL) has gained popularity for auto-bidding. However, most current RL auto-bidding methods are modeled through the Markovian Decision Process (MDP), which assumes the Markovian state transition. This assumption restricts the ability to perform in long horizon scenarios and makes the model unstable when dealing with highly random online advertising environments. To tackle this issue, this paper introduces AI-Generated Bidding (AIGB), a novel paradigm for auto-bidding through generative modeling. In this paradigm, we propose DiffBid, a conditional diffusion modeling approach for bid generation. DiffBid directly models the correlation between the return and the entire trajectory, effectively avoiding error propagation across time steps in long horizons. Additionally, DiffBid offers a versatile approach for generating trajectories that maximize given targets while adhering to specific constraints. Extensive experiments conducted on the real-world dataset and online A/B test on Alibaba advertising platform demonstrate the effectiveness of DiffBid, achieving 2.81% increase in GMV and 3.36% increase in ROI.

Visual Question Answering (VQA) has emerged as a highly engaging field in recent years, attracting increasing research efforts aiming to enhance VQA accuracy through the deployment of advanced models such as Transformers. Despite this growing interest, there has been limited exploration into the comparative analysis and impact of textual modalities within VQA, particularly in terms of model complexity and its effect on performance. In this work, we conduct a comprehensive comparison between complex textual models that leverage long dependency mechanisms and simpler models focusing on local textual features within a well-established VQA framework. Our findings reveal that employing complex textual encoders is not invariably the optimal approach for the VQA-v2 dataset. Motivated by this insight, we introduce an improved model, ConvGRU, which incorporates convolutional layers to enhance the representation of question text. Tested on the VQA-v2 dataset, ConvGRU achieves better performance without substantially increasing parameter complexity.

Automated bidding, an emerging intelligent decision making paradigm powered by machine learning, has become popular in online advertising. Advertisers in automated bidding evaluate the cumulative utilities and have private financial constraints over multiple ad auctions in a long-term period. Based on these distinct features, we consider a new ad auction model for automated bidding: the values of advertisers are public while the financial constraints, such as budget and return on investment (ROI) rate, are private types. We derive the truthfulness conditions with respect to private constraints for this multi-dimensional setting, and demonstrate any feasible allocation rule could be equivalently reduced to a series of non-decreasing functions on budget. However, the resulted allocation mapped from these non-decreasing functions generally follows an irregular shape, making it difficult to obtain a closed-form expression for the auction objective. To overcome this design difficulty, we propose a family of truthful automated bidding auction with personalized rank scores, similar to the Generalized Second-Price (GSP) auction. The intuition behind our design is to leverage personalized rank scores as the criteria to allocate items, and compute a critical ROI to transform the constraints on budget to the same dimension as ROI. The experimental results demonstrate that the proposed auction mechanism outperforms the widely used ad auctions, such as first-price auction and second-price auction, in various automated bidding environments.

The standard framework of online bidding algorithm design assumes that the seller commits himself to faithfully implementing the rules of the adopted auction. However, the seller may attempt to cheat in execution to increase his revenue if the auction belongs to the class of non-credible auctions. For example, in a second-price auction, the seller could create a fake bid between the highest bid and the second highest bid. This paper focuses on one such case of online bidding in repeated second-price auctions. At each time $t$, the winner with bid $b_t$ is charged not the highest competing bid $d_t$ but a manipulated price $p_t = \alpha_0 d_t + (1-\alpha_0) b_t$, where the parameter $\alpha_0 \in [0, 1]$ in essence measures the seller's credibility. Unlike classic repeated-auction settings where the bidder has access to samples $(d_s)_{s=1}^{t-1}$, she can only receive mixed signals of $(b_s)_{s=1}^{t-1}$, $(d_s)_{s=1}^{t-1}$ and $\alpha_0$ in this problem. The task for the bidder is to learn not only the bid distributions of her competitors but also the seller's credibility. We establish regret lower bounds in various information models and provide corresponding online bidding algorithms that can achieve near-optimal performance. Specifically, we consider three cases of prior information based on whether the credibility $\alpha_0$ and the distribution of the highest competing bids are known. Our goal is to characterize the landscape of online bidding in non-credible auctions and understand the impact of the seller's credibility on online bidding algorithm design under different information structures.

The fine-grained ship image recognition task aims to identify various classes of ships. However, small inter-class, large intra-class differences between ships, and lacking of training samples are the reasons that make the task difficult. Therefore, to enhance the accuracy of the fine-grained ship image recognition, we design a fine-grained ship image recognition network based on bilinear convolutional neural network (BCNN) with Inception and additive margin Softmax (AM-Softmax). This network improves the BCNN in two aspects. Firstly, by introducing Inception branches to the BCNN network, it is helpful to enhance the ability of extracting comprehensive features from ships. Secondly, by adding margin values to the decision boundary, the AM-Softmax function can better extend the inter-class differences and reduce the intra-class differences. In addition, as there are few publicly available datasets for fine-grained ship image recognition, we construct a Ship-43 dataset containing 47,300 ship images belonging to 43 categories. Experimental results on the constructed Ship-43 dataset demonstrate that our method can effectively improve the accuracy of ship image recognition, which is 4.08% higher than the BCNN model. Moreover, comparison results on the other three public fine-grained datasets (Cub, Cars, and Aircraft) further validate the effectiveness of the proposed method.

The fine-grained ship image recognition task aims to identify various classes of ships. However, small inter-class, large intra-class differences between ships, and lacking of training samples are the reasons that make the task difficult. Therefore, to enhance the accuracy of the fine-grained ship image recognition, we design a fine-grained ship image recognition network based on bilinear convolutional neural network (BCNN) with Inception and additive margin Softmax (AM-Softmax). This network improves the BCNN in two aspects. Firstly, by introducing Inception branches to the BCNN network, it is helpful to enhance the ability of extracting comprehensive features from ships. Secondly, by adding margin values to the decision boundary, the AM-Softmax function can better extend the inter-class differences and reduce the intra-class differences. In addition, as there are few publicly available datasets for fine-grained ship image recognition, we construct a Ship-43 dataset containing 47,300 ship images belonging to 43 categories. Experimental results on the constructed Ship-43 dataset demonstrate that our method can effectively improve the accuracy of ship image recognition, which is 4.08% higher than the BCNN model. Moreover, comparison results on the other three public fine-grained datasets (Cub, Cars, and Aircraft) further validate the effectiveness of the proposed method.

The fine-grained ship image recognition task aims to identify various classes of ships. However, small inter-class, large intra-class differences between ships, and lacking of training samples are the reasons that make the task difficult. Therefore, to enhance the accuracy of the fine-grained ship image recognition, we design a fine-grained ship image recognition network based on bilinear convolutional neural network (BCNN) with Inception and additive margin Softmax (AM-Softmax). This network improves the BCNN in two aspects. Firstly, by introducing Inception branches to the BCNN network, it is helpful to enhance the ability of extracting comprehensive features from ships. Secondly, by adding margin values to the decision boundary, the AM-Softmax function can better extend the inter-class differences and reduce the intra-class differences. In addition, as there are few publicly available datasets for fine-grained ship image recognition, we construct a Ship-43 dataset containing 47,300 ship images belonging to 43 categories. Experimental results on the constructed Ship-43 dataset demonstrate that our method can effectively improve the accuracy of ship image recognition, which is 4.08% higher than the BCNN model. Moreover, comparison results on the other three public fine-grained datasets (Cub, Cars, and Aircraft) further validate the effectiveness of the proposed method.

Digital advertising constitutes one of the main revenue sources for online platforms. In recent years, some advertisers tend to adopt auto-bidding tools to facilitate advertising performance optimization, making the classical \emph{utility maximizer} model in auction theory not fit well. Some recent studies proposed a new model, called \emph{value maximizer}, for auto-bidding advertisers with return-on-investment (ROI) constraints. However, the model of either utility maximizer or value maximizer could only characterize partial advertisers in real-world advertising platforms. In a mixed environment where utility maximizers and value maximizers coexist, the truthful ad auction design would be challenging since bidders could manipulate both their values and affiliated classes, leading to a multi-parameter mechanism design problem. In this work, we address this issue by proposing a payment rule which combines the corresponding ones in classical VCG and GSP mechanisms in a novel way. Based on this payment rule, we propose a truthful auction mechanism with an approximation ratio of $2$ on social welfare, which is close to the lower bound of at least $\frac{5}{4}$ that we also prove. The designed auction mechanism is a generalization of VCG for utility maximizers and GSP for value maximizers., Comment: accepted by AAAI2023

In mechanism design, it is challenging to design the optimal auction with correlated values in general settings. Although value distribution can be further exploited to improve revenue, the complex correlation structure makes it hard to acquire in practice. Data-driven auction mechanisms, powered by machine learning, enable to design auctions directly from historical auction data, without relying on specific value distributions. In this work, we design a learning-based auction, which can encode the correlation of values into the rank score of each bidder, and further adjust the ranking rule to approach the optimal revenue. We strictly guarantee the property of strategy-proofness by encoding game theoretical conditions into the neural network structure. Furthermore, all operations in the designed auctions are differentiable to enable an end-to-end training paradigm. Experimental results demonstrate that the proposed auction mechanism can represent almost any strategy-proof auction mechanism, and outperforms the auction mechanisms wildly used in the correlated value settings., Comment: The proof of the epxressiveness of CAN is wrong. We made some unnecessary assumptions. We need to correct this idea and resubmit it later

Searchable symmetric encryption enables a cloud server to answer queries directly over encrypted data. Two key requirements are a strong security guarantee and a sub-linear search performance. The bucketization approach in the literature addresses these requirements at the expense of downloading false positives and requiring the local search at the client side. In this article, we propose a novel approach to meet these requirements while minimizing the clients work and communication cost. First, a relaxed notion of ciphertext indistinguishability on partitioned data is formalized, called class indistinguishability, which provides a level of ciphertext indistinguishability similar to that of bucketization but allows the server to perform search of relevant data and filter false positives. We present a construction for achieving these goals through a two-phase search algorithm. The first phase finds a candidate set through a sub-linear search. The second phase finds the exact query result using a linear search applied to the candidate set. The experiment results on large real-world data-sets show that our approach outperforms the state-of-the-art. This article focuses on the class of equality conjunction search, but it applies to the general class of Boolean queries of equalities because the latter can be reduced to several equality conjunction queries. © 2013 IEEE.

For the scalability of industrial online advertising systems, a two-stage auction architecture is widely used to enable efficient ad allocation on a large set of corpus within a limited response time. The current deployed two-stage ad auction usually retrieves an ad subset by a coarse ad quality metric in a pre-auction stage, and then determines the auction outcome by a refined metric in the subsequent stage. However, this simple and greedy solution suffers from serious performance degradation, as it regards the decision in each stage separately, leading to an improper ad selection metric for the pre-auction stage. In this work, we explicitly investigate the relation between the coarse and refined ad quality metrics, and design a two-stage ad auction by taking the decision interaction between the two stages into account. We decouple the design of the two-stage auction by solving a stochastic subset selection problem in the pre-auction stage and conducting a general second price (GSP) auction in the second stage. We demonstrate that this decouple still preserves the incentive compatibility of the auction mechanism. As the proposed formulation of the pre-auction stage is an NP-hard problem, we propose a scalable approximation solution by defining a new subset selection metric, namely Pre-Auction Score (PAS). Experiment results on both public and industrial dataset demonstrate the significant improvement on social welfare and revenue of the proposed two-stage ad auction, than the intuitive greedy two-stage auction and other baselines.

In online advertising, auto-bidding has become an essential tool for advertisers to optimize their preferred ad performance metrics by simply expressing high-level campaign objectives and constraints. Previous works designed auto-bidding tools from the view of single-agent, without modeling the mutual influence between agents. In this paper, we instead consider this problem from a distributed multi-agent perspective, and propose a general $\underline{M}$ulti-$\underline{A}$gent reinforcement learning framework for $\underline{A}$uto-$\underline{B}$idding, namely MAAB, to learn the auto-bidding strategies. First, we investigate the competition and cooperation relation among auto-bidding agents, and propose a temperature-regularized credit assignment to establish a mixed cooperative-competitive paradigm. By carefully making a competition and cooperation trade-off among agents, we can reach an equilibrium state that guarantees not only individual advertiser's utility but also the system performance (i.e., social welfare). Second, to avoid the potential collusion behaviors of bidding low prices underlying the cooperation, we further propose bar agents to set a personalized bidding bar for each agent, and then alleviate the revenue degradation due to the cooperation. Third, to deploy MAAB in the large-scale advertising system with millions of advertisers, we propose a mean-field approach. By grouping advertisers with the same objective as a mean auto-bidding agent, the interactions among the large-scale advertisers are greatly simplified, making it practical to train MAAB efficiently. Extensive experiments on the offline industrial dataset and Alibaba advertising platform demonstrate that our approach outperforms several baseline methods in terms of social welfare and revenue., Comment: Accepted by WSDM 2022

In e-commerce advertising, it is crucial to jointly consider various performance metrics, e.g., user experience, advertiser utility, and platform revenue. Traditional auction mechanisms, such as GSP and VCG auctions, can be suboptimal due to their fixed allocation rules to optimize a single performance metric (e.g., revenue or social welfare). Recently, data-driven auctions, learned directly from auction outcomes to optimize multiple performance metrics, have attracted increasing research interests. However, the procedure of auction mechanisms involves various discrete calculation operations, making it challenging to be compatible with continuous optimization pipelines in machine learning. In this paper, we design \underline{D}eep \underline{N}eural \underline{A}uctions (DNAs) to enable end-to-end auction learning by proposing a differentiable model to relax the discrete sorting operation, a key component in auctions. We optimize the performance metrics by developing deep models to efficiently extract contexts from auctions, providing rich features for auction design. We further integrate the game theoretical conditions within the model design, to guarantee the stability of the auctions. DNAs have been successfully deployed in the e-commerce advertising system at Taobao. Experimental evaluation results on both large-scale data set as well as online A/B test demonstrated that DNAs significantly outperformed other mechanisms widely adopted in industry., Comment: To appear in the Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2021

In e-commerce advertising, the ad platform usually relies on auction mechanisms to optimize different performance metrics, such as user experience, advertiser utility, and platform revenue. However, most of the state-of-the-art auction mechanisms only focus on optimizing a single performance metric, e.g., either social welfare or revenue, and are not suitable for e-commerce advertising with various, dynamic, difficult to estimate, and even conflicting performance metrics. In this paper, we propose a new mechanism called Deep GSP auction, which leverages deep learning to design new rank score functions within the celebrated GSP auction framework. These new rank score functions are implemented via deep neural network models under the constraints of monotone allocation and smooth transition. The requirement of monotone allocation ensures Deep GSP auction nice game theoretical properties, while the requirement of smooth transition guarantees the advertiser utilities would not fluctuate too much when the auction mechanism switches among candidate mechanisms to achieve different optimization objectives. We deployed the proposed mechanisms in a leading e-commerce ad platform and conducted comprehensive experimental evaluations with both offline simulations and online A/B tests. The results demonstrated the effectiveness of the Deep GSP auction compared to the state-of-the-art auction mechanisms., Comment: To appear in the Proceedings of the 14th ACM International Conference on Web Search and Data Mining (WSDM), 2021

In the field of fixed wing aircraft, many morphing technologies have been applied to the wing, such as adaptive airfoil, variable span aircraft, variable swept angle aircraft, etc., but few are aimed at the tail. The traditional fixed wing tail includes horizontal and vertical tail. Inspired by the bird tail, this paper will introduce a new bionic tail. The tail has a novel control mode, which has multiple control variables. Compared with the traditional fixed wing tail, it adds the area control and rotation control around the longitudinal symmetry axis, so it can control the pitch and yaw of the aircraft at the same time. When the area of the tail changes, the maneuverability and stability of the aircraft can be changed, and the aerodynamic efficiency of the aircraft can also be improved. The aircraft with morphing ability is often difficult to establish accurate mathematical model, because the model has a strong nonlinear, model-based control method is difficult to deal with the strong nonlinear aircraft. In recent years, with the rapid development of artificial intelligence technology, learning based control methods are also brilliant, in which the deep reinforcement learning algorithm can be a good solution to the control object which is difficult to establish model. In this paper, the model-free control algorithm PPO is used to control the tail, and the traditional PID is used to control the aileron and throttle. After training in simulation, the tail shows excellent attitude control ability., Comment: PDF, 8 pages with 10 figures, IEEE Robotics and Automation Letters and ICRA (under review)

As data outsourcing becomes popular, oblivious algorithms have raised extensive attentions. Their control flow and data access pattern appear to be independent of the input data they compute on. Oblivious algorithms, therefore, are especially suitable for secure processing in outsourced environments. In this work, we focus on oblivious shuffling algorithms that aim to shuffle encrypted data blocks outsourced to a cloud server without disclosing the actual permutation of blocks to the server. Existing oblivious shuffling algorithms suffer from issues of heavy communication cost and client computation cost for shuffling large-sized blocks because all outsourced blocks must be downloaded to the client for shuffling or peeling off extra encryption layers. To help eliminate this void, we introduce the "repeatable oblivious shuffling" notation that avoids moving blocks to the client and thus restricts the communication and client computation costs to be independent of the block size. For the first time, we present a concrete construction of repeatable oblivious shuffling using additively homomorphic encryption. The comprehensive evaluation of our construction shows its effective usability in practice for shuffling large-sized blocks.

We consider the following secure data retrieval problem: a client outsources encrypted data blocks to a semi-trusted cloud server and later retrieves blocks without disclosing access patterns. Existing PIR and ORAM solutions suffer from serious performance bottlenecks in terms of communication or computation costs. To help eliminate this void, we introduce “access pattern unlinkability” that separates access pattern privacy into short-term privacy at individual query level and long-term privacy at query distribution level. This new security definition provides tunable trade-offs between privacy and query performance. We present an efficient construction, called SBR protocol, using PIR and Oblivious Shuffling to enable secure data retrieval while satisfying access pattern unlinkability. Both analytical and empirical analysis show that SBR exhibits flexibility and usability in practice. © 2019 Copyright held by the owner/author(s). Publication rights licensed to ACM.

Searchable symmetric encryption (SSE) enables a remote cloud server to answer queries directly over encrypted data on a client’s behalf, therefore, relieves the resource limited client from complicated data management tasks. Two key requirements are a strong security guarantee and a sub-linear search performance. The bucketization approach in the literature addresses these requirements at the expense of downloading many false positives or requiring the client to search relevant bucket ids locally, which limits the applicability of the method. In this paper, we propose a novel approach CLASS to meet these requirements for equality conjunction search while minimizing the client work and communication cost. First, we generalize the standard ciphertext indistinguishability to partitioned data, called class indistinguishability, which provides a level of ciphertext indistinguishability similar to that of bucketization but allows the cloud server to perform search of relevant data and filtering of false positives. We present a construction achieving these goals through a two-phase search algorithm for a query. The first phase finds a candidate set through a sub-linear search. The second phase finds the exact query result using a linear search applied to the candidate set. Both phases are performed by the server and are implemented by plugging in existing search methods. The experiment results on large real-world data sets show that our approach outperforms the state-of-the-art. © Springer Nature Switzerland AG 2019.

As data outsourcing becomes popular, oblivious algorithms have raised extensive attentions. Their control flow and data access pattern appear to be independent of the input data they compute on. Oblivious algorithms, therefore, are especially suitable for secure processing in outsourced environments. In this work, we focus on oblivious shuffling algorithms that aim to shuffle encrypted data blocks outsourced to a cloud server without disclosing the actual permutation of blocks to the server. Existing oblivious shuffling algorithms suffer from issues of heavy communication cost and client computation cost for shuffling large-sized blocks because all outsourced blocks must be downloaded to the client for shuffling or peeling off extra encryption layers. To help eliminate this void, we introduce the "repeatable oblivious shuffling" notation that avoids moving blocks to the client and thus restricts the communication and client computation costs to be independent of the block size. For the first time, we present a concrete construction of repeatable oblivious shuffling using additively homomorphic encryption. The comprehensive evaluation of our construction shows its effective usability in practice for shuffling large-sized blocks.

We consider the following secure data retrieval problem: a client outsources encrypted data blocks to a semi-trusted cloud server and later retrieves blocks without disclosing access patterns. Existing PIR and ORAM solutions suffer from serious performance bottlenecks in terms of communication or computation costs. To help eliminate this void, we introduce “access pattern unlinkability” that separates access pattern privacy into short-term privacy at individual query level and long-term privacy at query distribution level. This new security definition provides tunable trade-offs between privacy and query performance. We present an efficient construction, called SBR protocol, using PIR and Oblivious Shuffling to enable secure data retrieval while satisfying access pattern unlinkability. Both analytical and empirical analysis show that SBR exhibits flexibility and usability in practice. © 2019 Copyright held by the owner/author(s). Publication rights licensed to ACM.

Searchable symmetric encryption (SSE) enables a remote cloud server to answer queries directly over encrypted data on a client’s behalf, therefore, relieves the resource limited client from complicated data management tasks. Two key requirements are a strong security guarantee and a sub-linear search performance. The bucketization approach in the literature addresses these requirements at the expense of downloading many false positives or requiring the client to search relevant bucket ids locally, which limits the applicability of the method. In this paper, we propose a novel approach CLASS to meet these requirements for equality conjunction search while minimizing the client work and communication cost. First, we generalize the standard ciphertext indistinguishability to partitioned data, called class indistinguishability, which provides a level of ciphertext indistinguishability similar to that of bucketization but allows the cloud server to perform search of relevant data and filtering of false positives. We present a construction achieving these goals through a two-phase search algorithm for a query. The first phase finds a candidate set through a sub-linear search. The second phase finds the exact query result using a linear search applied to the candidate set. Both phases are performed by the server and are implemented by plugging in existing search methods. The experiment results on large real-world data sets show that our approach outperforms the state-of-the-art. © Springer Nature Switzerland AG 2019.

As data outsourcing becomes popular, oblivious algorithms have raised extensive attentions. Their control flow and data access pattern appear to be independent of the input data they compute on. Oblivious algorithms, therefore, are especially suitable for secure processing in outsourced environments. In this work, we focus on oblivious shuffling algorithms that aim to shuffle encrypted data blocks outsourced to a cloud server without disclosing the actual permutation of blocks to the server. Existing oblivious shuffling algorithms suffer from issues of heavy communication cost and client computation cost for shuffling large-sized blocks because all outsourced blocks must be downloaded to the client for shuffling or peeling off extra encryption layers. To help eliminate this void, we introduce the "repeatable oblivious shuffling" notation that avoids moving blocks to the client and thus restricts the communication and client computation costs to be independent of the block size. For the first time, we present a concrete construction of repeatable oblivious shuffling using additively homomorphic encryption. The comprehensive evaluation of our construction shows its effective usability in practice for shuffling large-sized blocks.

We consider the following secure data retrieval problem: a client outsources encrypted data blocks to a semi-trusted cloud server and later retrieves blocks without disclosing access patterns. Existing PIR and ORAM solutions suffer from serious performance bottlenecks in terms of communication or computation costs. To help eliminate this void, we introduce “access pattern unlinkability” that separates access pattern privacy into short-term privacy at individual query level and long-term privacy at query distribution level. This new security definition provides tunable trade-offs between privacy and query performance. We present an efficient construction, called SBR protocol, using PIR and Oblivious Shuffling to enable secure data retrieval while satisfying access pattern unlinkability. Both analytical and empirical analysis show that SBR exhibits flexibility and usability in practice. © 2019 Copyright held by the owner/author(s). Publication rights licensed to ACM.

Searchable symmetric encryption (SSE) enables a remote cloud server to answer queries directly over encrypted data on a client’s behalf, therefore, relieves the resource limited client from complicated data management tasks. Two key requirements are a strong security guarantee and a sub-linear search performance. The bucketization approach in the literature addresses these requirements at the expense of downloading many false positives or requiring the client to search relevant bucket ids locally, which limits the applicability of the method. In this paper, we propose a novel approach CLASS to meet these requirements for equality conjunction search while minimizing the client work and communication cost. First, we generalize the standard ciphertext indistinguishability to partitioned data, called class indistinguishability, which provides a level of ciphertext indistinguishability similar to that of bucketization but allows the cloud server to perform search of relevant data and filtering of false positives. We present a construction achieving these goals through a two-phase search algorithm for a query. The first phase finds a candidate set through a sub-linear search. The second phase finds the exact query result using a linear search applied to the candidate set. Both phases are performed by the server and are implemented by plugging in existing search methods. The experiment results on large real-world data sets show that our approach outperforms the state-of-the-art. © Springer Nature Switzerland AG 2019.

This submission has been removed by arXiv administrators due to copyright infringement., Comment: This submission has been removed by arXiv administrators due to copyright infringement

Energy consumption is an important issue in continuous wireless telemonitoring of physiological signals. Compressed sensing (CS) is a promising framework to address it, due to its energy-efficient data compression procedure. However, most CS algorithms have difficulty in data recovery due to nonsparsity characteristic of many physiological signals. Block sparse Bayesian learning (BSBL) is an effective approach to recover such signals with satisfactory recovery quality. However, it is time-consuming in recovering multichannel signals, since its computational load almost linearly increases with the number of channels. This work proposes a spatiotemporal sparse Bayesian learning algorithm to recover multichannel signals simultaneously. It not only exploits temporal correlation within each channel signal, but also exploits inter-channel correlation among different channel signals. Furthermore, its computational load is not significantly affected by the number of channels. The proposed algorithm was applied to brain computer interface (BCI) and EEG-based driver's drowsiness estimation. Results showed that the algorithm had both better recovery performance and much higher speed than BSBL. Particularly, the proposed algorithm ensured that the BCI classification and the drowsiness estimation had little degradation even when data were compressed by 80%, making it very suitable for continuous wireless telemonitoring of multichannel signals.

Energy consumption is an important issue in continuous wireless telemonitoring of physiological signals. Compressed sensing (CS) is a promising framework to address it, due to its energy-efficient data compression procedure. However, most CS algorithms have difficulty in data recovery due to nonsparsity characteristic of many physiological signals. Block sparse Bayesian learning (BSBL) is an effective approach to recover such signals with satisfactory recovery quality. However, it is time-consuming in recovering multichannel signals, since its computational load almost linearly increases with the number of channels. This work proposes a spatiotemporal sparse Bayesian learning algorithm to recover multichannel signals simultaneously. It not only exploits temporal correlation within each channel signal, but also exploits inter-channel correlation among different channel signals. Furthermore, its computational load is not significantly affected by the number of channels. The proposed algorithm was applied to brain computer interface (BCI) and EEG-based driver's drowsiness estimation. Results showed that the algorithm had both better recovery performance and much higher speed than BSBL. Particularly, the proposed algorithm ensured that the BCI classification and the drowsiness estimation had little degradation even when data were compressed by 80%, making it very suitable for continuous wireless telemonitoring of multichannel signals.

Sparse signal recovery algorithms have significant impact on many fields. The core of these algorithms is to find a solution to an underdetermined inverse system of equations, where the solution is expected to be sparse or approximately sparse. However, most algorithms ignored correlation among nonzero entries of a solution, which is often encountered in a practical problem. Thus, it is unclear what role the correlation plays in signal recovery. This work aims to design algorithms which can exploit a variety of correlation structures in solutions and reveal the impact of these correlation structures on algorithms' recovery performance. First, a block sparse Bayesian learning (BSBL) framework is proposed. Based on it, a number of sparse Bayesian learning (SBL) algorithms are derived to exploit intra-block correlation in a block sparse model, temporal correlation in a multiple measurement vector model, spatiotemporal correlation in a spatiotemporal sparse model, and local temporal correlation in a time-varying sparse model. Several optimization approaches are employed in the algorithm development, such as the expectation-maximization method, the bound-optimization method, and a fixed-point method. Experimental results show that these algorithms have superior performance. With these algorithms, we find that different correlation structures affect the quality of estimated solutions to different degrees. However, if these correlation structures are present and exploited, algorithms' performance can be largely improved. Inspired by this, we connect these algorithms to Group-Lasso type algorithms and iterative reweighted [ell]₁ and [ell]₂ algorithms, and suggest strategies to modify them to exploit the correlation structures for better performance. The derived algorithms have been used with considerable success in various challenging applications such as wireless telemonitoring of raw physiological signals and prediction of patients' cognitive levels from their neuroimagin

Sparse signal recovery algorithms have significant impact on many fields. The core of these algorithms is to find a solution to an underdetermined inverse system of equations, where the solution is expected to be sparse or approximately sparse. However, most algorithms ignored correlation among nonzero entries of a solution, which is often encountered in a practical problem. Thus, it is unclear what role the correlation plays in signal recovery. This work aims to design algorithms which can exploit a variety of correlation structures in solutions and reveal the impact of these correlation structures on algorithms' recovery performance. First, a block sparse Bayesian learning (BSBL) framework is proposed. Based on it, a number of sparse Bayesian learning (SBL) algorithms are derived to exploit intra-block correlation in a block sparse model, temporal correlation in a multiple measurement vector model, spatiotemporal correlation in a spatiotemporal sparse model, and local temporal correlation in a time-varying sparse model. Several optimization approaches are employed in the algorithm development, such as the expectation-maximization method, the bound-optimization method, and a fixed-point method. Experimental results show that these algorithms have superior performance. With these algorithms, we find that different correlation structures affect the quality of estimated solutions to different degrees. However, if these correlation structures are present and exploited, algorithms' performance can be largely improved. Inspired by this, we connect these algorithms to Group-Lasso type algorithms and iterative reweighted [ell]₁ and [ell]₂ algorithms, and suggest strategies to modify them to exploit the correlation structures for better performance. The derived algorithms have been used with considerable success in various challenging applications such as wireless telemonitoring of raw physiological signals and prediction of patients' cognitive levels from their neuroimagin

The measurement problem of quantum mechanics has been there for almost a century. Academia tends to regard this problem as a combination of definite outcomes and preferred basis. At the start of this century, John Conway and Simon Kochen proposed the strong free will theorem. This theorem clearly proved that free will exist within any given particle, which implies that particles have perceptions and freedom of choices. A particle’s perception and its free will are then linked with energy, and two principles of energy are reached: the perception-energy principle and the will-energy principle. With these two principles, the measurement problem can be solved, and a new interpretation of quantum mechanics can be revealed.

We address the issue of applying quantized compressed sensing (CS) on low-energy telemonitoring. So far, few works studied this problem in applications where signals were only approximately sparse. We propose a two-stage data compressor based on quantized CS, where signals are compressed by compressed sensing and then the compressed measurements are quantized with only 2 bits per measurement. This compressor can greatly reduce the transmission bit-budget. To recover signals from underdetermined, quantized measurements, we develop a Bayesian De-quantization algorithm. It can exploit both the model of quantization errors and the correlated structure of physiological signals to improve the quality of recovery. The proposed data compressor and the recovery algorithm are validated on a dataset recorded on 12 subjects during fast running. Experiment results showed that an averaged 2.596 beat per minute (BPM) estimation error was achieved by jointly using compressed sensing with 50% compression ratio and a 2-bit quantizer. The results imply that we can effectively transmit n bits instead of n samples, which is a substantial improvement for low-energy wireless telemonitoring.

Goal: A new method for heart rate monitoring using photoplethysmography (PPG) during physical activities is proposed. Methods: It jointly estimates spectra of PPG signals and simultaneous acceleration signals, utilizing the multiple measurement vector model in sparse signal recovery. Due to a common sparsity constraint on spectral coefficients, the method can easily identify and remove spectral peaks of motion artifact (MA) in PPG spectra. Thus, it does not need any extra signal processing modular to remove MA as in some other algorithms. Furthermore, seeking spectral peaks associated with heart rate is simplified. Results: Experimental results on 12 PPG datasets sampled at 25 Hz and recorded during subjects' fast running showed that it had high performance. The average absolute estimation error was 1.28 beat per minute and the standard deviation was 2.61 beat per minute. Conclusion and Significance: These results show that the method has great potential to be used for PPG-based heart rate monitoring in wearable devices for fitness tracking and health monitoring., Comment: Published in IEEE Transactions on Biomedical Engineering, Vol. 62, No. 8, PP. 1902-1910, August 2015

Predicting cognitive performance of subjects from their magnetic resonance imaging (MRI) measures and identifying relevant imaging biomarkers are important research topics in the study of Alzheimer's disease. Traditionally, this task is performed by formulating a linear regression problem. Recently, it is found that using a linear sparse regression model can achieve better prediction accuracy. However, most existing studies only focus on the exploitation of sparsity of regression coefficients, ignoring useful structure information in regression coefficients. Also, these linear sparse models may not capture more complicated and possibly nonlinear relationships between cognitive performance and MRI measures. Motivated by these observations, in this work we build a sparse multivariate regression model for this task and propose an empirical sparse Bayesian learning algorithm. Different from existing sparse algorithms, the proposed algorithm models the response as a nonlinear function of the predictors by extending the predictor matrix with block structures. Further, it exploits not only inter-vector correlation among regression coefficient vectors, but also intra-block correlation in each regression coefficient vector. Experiments on the Alzheimer's Disease Neuroimaging Initiative database showed that the proposed algorithm not only achieved better prediction performance than state-of-the-art competitive methods, but also effectively identified biologically meaningful patterns.

Energy consumption is an important issue in continuous wireless telemonitoring of physiological signals. Compressed sensing (CS) is a promising framework to address it, due to its energy-efficient data compression procedure. However, most CS algorithms have difficulty in data recovery due to non-sparsity characteristic of many physiological signals. Block sparse Bayesian learning (BSBL) is an effective approach to recover such signals with satisfactory recovery quality. However, it is time-consuming in recovering multichannel signals, since its computational load almost linearly increases with the number of channels. This work proposes a spatiotemporal sparse Bayesian learning algorithm to recover multichannel signals simultaneously. It not only exploits temporal correlation within each channel signal, but also exploits inter-channel correlation among different channel signals. Furthermore, its computational load is not significantly affected by the number of channels. The proposed algorithm was applied to brain computer interface (BCI) and EEG-based driver's drowsiness estimation. Results showed that the algorithm had both better recovery performance and much higher speed than BSBL. Particularly, the proposed algorithm ensured that the BCI classification and the drowsiness estimation had little degradation even when data were compressed by 80%, making it very suitable for continuous wireless telemonitoring of multichannel signals., Comment: Codes are available at: https://sites.google.com/site/researchbyzhang/stsbl

Non-intrusive load monitoring (NILM) is an important topic in smart-grid and smart-home. Many energy disaggregation algorithms have been proposed to detect various individual appliances from one aggregated signal observation. However, few works studied the energy disaggregation of plug-in electric vehicle (EV) charging in the residential environment since EVs charging at home has emerged only recently. Recent studies showed that EV charging has a large impact on smart-grid especially in summer. Therefore, EV charging monitoring has become a more important and urgent missing piece in energy disaggregation. In this paper, we present a novel method to disaggregate EV charging signals from aggregated real power signals. The proposed method can effectively mitigate interference coming from air-conditioner (AC), enabling accurate EV charging detection and energy estimation under the presence of AC power signals. Besides, the proposed algorithm requires no training, demands a light computational load, delivers high estimation accuracy, and works well for data recorded at the low sampling rate 1/60 Hz. When the algorithm is tested on real-world data recorded from 11 houses over about a whole year (total 125 months worth of data), the averaged error in estimating energy consumption of EV charging is 15.7 kwh/month (while the true averaged energy consumption of EV charging is 208.5 kwh/month), and the averaged normalized mean square error in disaggregating EV charging load signals is 0.19., Comment: Accepted by The 40th Annual Conference of the IEEE Industrial Electronics Society (IECON 2014)

Heart rate monitoring using wrist-type photoplethysmographic (PPG) signals during subjects' intensive exercise is a difficult problem, since the signals are contaminated by extremely strong motion artifacts caused by subjects' hand movements. So far few works have studied this problem. In this work, a general framework, termed TROIKA, is proposed, which consists of signal decomposiTion for denoising, sparse signal RecOnstructIon for high-resolution spectrum estimation, and spectral peaK trAcking with verification. The TROIKA framework has high estimation accuracy and is robust to strong motion artifacts. Many variants can be straightforwardly derived from this framework. Experimental results on datasets recorded from 12 subjects during fast running at the peak speed of 15 km/hour showed that the average absolute error of heart rate estimation was 2.34 beat per minute (BPM), and the Pearson correlation between the estimates and the ground-truth of heart rate was 0.992. This framework is of great values to wearable devices such as smart-watches which use PPG signals to monitor heart rate for fitness., Comment: Matlab codes and data are available at: https://sites.google.com/site/researchbyzhang

As a lossy compression framework, compressed sensing has drawn much attention in wireless telemonitoring of biosignals due to its ability to reduce energy consumption and make possible the design of low-power devices. However, the non-sparseness of biosignals presents a major challenge to compressed sensing. This study proposes and evaluates a spatio-temporal sparse Bayesian learning algorithm, which has the desired ability to recover such non-sparse biosignals. It exploits both temporal correlation in each individual biosignal and inter-channel correlation among biosignals from different channels. The proposed algorithm was used for compressed sensing of multichannel electroencephalographic (EEG) signals for estimating vehicle drivers' drowsiness. Results showed that the drowsiness estimation was almost unaffected even if raw EEG signals (containing various artifacts) were compressed by 90%., Comment: Invited paper for 2013 Asilomar Conference on Signals, Systems & Computers (Asilomar 2013)

Wireless telemonitoring of physiological signals is an important topic in eHealth. In order to reduce on-chip energy consumption and extend sensor life, recorded signals are usually compressed before transmission. In this paper, we adopt compressed sensing (CS) as a low-power compression framework, and propose a fast block sparse Bayesian learning (BSBL) algorithm to reconstruct original signals. Experiments on real-world fetal ECG signals and epilepsy EEG signals showed that the proposed algorithm has good balance between speed and data reconstruction fidelity when compared to state-of-the-art CS algorithms. Further, we implemented the CS-based compression procedure and a low-power compression procedure based on a wavelet transform in Filed Programmable Gate Array (FPGA), showing that the CS-based compression can largely save energy and other on-chip computing resources.

Telehealth and wearable equipment can deliver personal healthcare and necessary treatment remotely. One major challenge is transmitting large amount of biosignals through wireless networks. The limited battery life calls for low-power data compressors. Compressive Sensing (CS) has proved to be a low-power compressor. In this study, we apply CS on the compression of multichannel biosignals. We firstly develop an efficient CS algorithm from the Block Sparse Bayesian Learning (BSBL) framework. It is based on a combination of the block sparse model and multiple measurement vector model. Experiments on real-life Fetal ECGs showed that the proposed algorithm has high fidelity and efficiency. Implemented in hardware, the proposed algorithm was compared to a Discrete Wavelet Transform (DWT) based algorithm, verifying the proposed one has low power consumption and occupies less computational resources., Comment: 2013 International Workshop on Biomedical and Health Informatics

Face recognition (FR) is an important task in pattern recognition and computer vision. Sparse representation (SR) has been demonstrated to be a powerful framework for FR. In general, an SR algorithm treats each face in a training dataset as a basis function, and tries to find a sparse representation of a test face under these basis functions. The sparse representation coefficients then provide a recognition hint. Early SR algorithms are based on a basic sparse model. Recently, it has been found that algorithms based on a block sparse model can achieve better recognition rates. Based on this model, in this study we use block sparse Bayesian learning (BSBL) to find a sparse representation of a test face for recognition. BSBL is a recently proposed framework, which has many advantages over existing block-sparse-model based algorithms. Experimental results on the Extended Yale B, the AR and the CMU PIE face databases show that using BSBL can achieve better recognition rates and higher robustness than state-of-the-art algorithms in most cases., Comment: Accepted by Mathematical Problems in Engineering in 2013

The performance of sparse signal recovery from noise corrupted, underdetermined measurements can be improved if both sparsity and correlation structure of signals are exploited. One typical correlation structure is the intra-block correlation in block sparse signals. To exploit this structure, a framework, called block sparse Bayesian learning (BSBL), has been proposed recently. Algorithms derived from this framework showed superior performance but they are not very fast, which limits their applications. This work derives an efficient algorithm from this framework, using a marginalized likelihood maximization method. Compared to existing BSBL algorithms, it has close recovery performance but is much faster. Therefore, it is more suitable for large scale datasets and applications requiring real-time implementation.

This work discusses the problem of sparse signal recovery when there is correlation among the values of non-zero entries. We examine intra-vector correlation in the context of the block sparse model and inter-vector correlation in the context of the multiple measurement vector model, as well as their combination. Algorithms based on the sparse Bayesian learning are presented and the benefits of incorporating correlation at the algorithm level are discussed. The impact of correlation on the limits of support recovery is also discussed highlighting the different impact intra-vector and inter-vector correlations have on such limits., Comment: Invited review paper of 2012 International Conference on Signal Processing and Communications (SPCOM 2012)

Fetal ECG (FECG) telemonitoring is an important branch in telemedicine. The design of a telemonitoring system via a wireless body-area network with low energy consumption for ambulatory use is highly desirable. As an emerging technique, compressed sensing (CS) shows great promise in compressing/reconstructing data with low energy consumption. However, due to some specific characteristics of raw FECG recordings such as non-sparsity and strong noise contamination, current CS algorithms generally fail in this application. This work proposes to use the block sparse Bayesian learning (BSBL) framework to compress/reconstruct non-sparse raw FECG recordings. Experimental results show that the framework can reconstruct the raw recordings with high quality. Especially, the reconstruction does not destroy the interdependence relation among the multichannel recordings. This ensures that the independent component analysis decomposition of the reconstructed recordings has high fidelity. Furthermore, the framework allows the use of a sparse binary sensing matrix with much fewer nonzero entries to compress recordings. Particularly, each column of the matrix can contain only two nonzero entries. This shows the framework, compared to other algorithms such as current CS algorithms and wavelet algorithms, can greatly reduce code execution in CPU in the data compression stage., Comment: The code and the data can be downloaded from the first author's homepage: http://sites.google.com/site/researchbyzhang/bsbl, or http://dsp.ucsd.edu/~zhilin/BSBL.html

Telemonitoring of electroencephalogram (EEG) through wireless body-area networks is an evolving direction in personalized medicine. Among various constraints in designing such a system, three important constraints are energy consumption, data compression, and device cost. Conventional data compression methodologies, although effective in data compression, consumes significant energy and cannot reduce device cost. Compressed sensing (CS), as an emerging data compression methodology, is promising in catering to these constraints. However, EEG is non-sparse in the time domain and also non-sparse in transformed domains (such as the wavelet domain). Therefore, it is extremely difficult for current CS algorithms to recover EEG with the quality that satisfies the requirements of clinical diagnosis and engineering applications. Recently, Block Sparse Bayesian Learning (BSBL) was proposed as a new method to the CS problem. This study introduces the technique to the telemonitoring of EEG. Experimental results show that its recovery quality is better than state-of-the-art CS algorithms, and sufficient for practical use. These results suggest that BSBL is very promising for telemonitoring of EEG and other non-sparse physiological signals., Comment: Matlab codes can be downloaded at: http://dsp.ucsd.edu/~zhilin/BSBL.html, or http://sites.google.com/site/researchbyzhang/bsbl

Iterative reweighted algorithms, as a class of algorithms for sparse signal recovery, have been found to have better performance than their non-reweighted counterparts. However, for solving the problem of multiple measurement vectors (MMVs), all the existing reweighted algorithms do not account for temporal correlation among source vectors and thus their performance degrades significantly in the presence of correlation. In this work we propose an iterative reweighted sparse Bayesian learning (SBL) algorithm exploiting the temporal correlation, and motivated by it, we propose a strategy to improve existing reweighted $\ell_2$ algorithms for the MMV problem, i.e. replacing their row norms with Mahalanobis distance measure. Simulations show that the proposed reweighted SBL algorithm has superior performance, and the proposed improvement strategy is effective for existing reweighted $\ell_2$ algorithms., Comment: Accepted by ICASSP 2011

A trend in compressed sensing (CS) is to exploit structure for improved reconstruction performance. In the basic CS model, exploiting the clustering structure among nonzero elements in the solution vector has drawn much attention, and many algorithms have been proposed. However, few algorithms explicitly consider correlation within a cluster. Meanwhile, in the multiple measurement vector (MMV) model correlation among multiple solution vectors is largely ignored. Although several recently developed algorithms consider the exploitation of the correlation, these algorithms need to know a priori the correlation structure, thus limiting their effectiveness in practical problems. Recently, we developed a sparse Bayesian learning (SBL) algorithm, namely T-SBL, and its variants, which adaptively learn the correlation structure and exploit such correlation information to significantly improve reconstruction performance. Here we establish their connections to other popular algorithms, such as the group Lasso, iterative reweighted $\ell_1$ and $\ell_2$ algorithms, and algorithms for time-varying sparsity. We also provide strategies to improve these existing algorithms., Comment: Extended abstract for ICML 2011 Structured Sparsity: Learning and Inference Workshop. Experiment codes can be downloaded from: http://dsp.ucsd.edu/~zhilin/papers/ICMLworkshop_code.zip