Jiwon Seo | Stanford University (original) (raw)

Papers by Jiwon Seo

We apply machine learning to thread level speculation, a future hardware framework for paralleliz... more We apply machine learning to thread level speculation, a future hardware framework for parallelizing sequential programs. By using machine learning to determine the parallel regions, the overall performance is nearly as good as the best heuristics for each application.

Proceedings of the 1st ACM Workshop on Mobile Cloud Computing & Services Social Networks and Beyond - MCS '10, 2010

This paper presents PrPl, a decentralized infrastructure that lets users participate in online so... more This paper presents PrPl, a decentralized infrastructure that lets users participate in online social networking without loss of data ownership. PrPl, short for private-public, has a person-centric architecture--each individual uses a Personal-Cloud Butler service that provides a safe haven for one's personal digital assets and supports sharing with fine-grain access control. A user can choose to run the Butler on

IEEE Transactions on Knowledge and Data Engineering, 2015

Many modern software platforms today, including browsers, middleware server architectures, cell p... more Many modern software platforms today, including browsers, middleware server architectures, cell phone operating systems, web application engines, support thirdparty software extensions. This paper proposes InvisiType, an object-oriented approach that enables platform developers to efficiently enforce fine-grained safety checks on thirdparty extensions without requiring their cooperation. This allows us to harness the true power of third-party software by giving it access to sensitive data while ensuring that it does not leak data.

ABSTRACT The convenient way for users to share data these days, be it photos, IM, or just their G... more ABSTRACT The convenient way for users to share data these days, be it photos, IM, or just their GPS locations, is for them to sign up as friends to the same social networking sites. As a result, not only are we inundated by invitations from our friends to join a large number of ...

Proceedings of the VLDB Endowment, 2013

Advertisement-supported social networking portals generally aim to lock in users' data and exploi... more Advertisement-supported social networking portals generally aim to lock in users' data and exploit personal information for ad targeting and other marketing purposes. Because of the network effect, it is not hard to envision a situation where the information of a very large population can end up in the hands of an oligopolistic group or even a sole monopolistic actor. Beyond the obvious privacy concerns, this outcome would clearly pose problems for healthy competition, ultimately harming end-users.

Proceedings of the 2014 ACM SIGMOD international conference on Management of data - SIGMOD '14, 2014

Graph algorithms are becoming increasingly important for analyzing large datasets in many fields.... more Graph algorithms are becoming increasingly important for analyzing large datasets in many fields. Real-world graph data follows a pattern of sparsity, that is not uniform but highly skewed towards a few items. Implementing graph traversal, statistics and machine learning algorithms on such data in a scalable manner is quite challenging. As a result, several graph analytics frameworks (GraphLab, CombBLAS, Giraph, SociaLite and Galois among others) have been developed, each offering a solution with different programming models and targeted at different users. Unfortunately, the "Ninja performance gap" between optimized code and most of these frameworks is very large (2-30X for most frameworks and up to 560X for Giraph) for common graph algorithms, and moreover varies widely with algorithms. This makes the end-users' choice of graph framework dependent not only on ease of use but also on performance. In this work, we offer a quantitative roadmap for improving the performance of all these frameworks and bridging the "ninja gap". We first present hand-optimized baselines that get performance close to hardware limits and higher than any published performance figure for these graph algorithms. We characterize the performance of both this native implementation as well as popular graph frameworks on a variety of algorithms. This study helps endusers delineate bottlenecks arising from the algorithms themselves vs. programming model abstractions vs. the framework implementations. Further, by analyzing the system-level behavior of these frameworks, we obtain bottlenecks that are agnostic to specific algorithms. We recommend changes to alleviate these bottlenecks (and implement some of them) and reduce the performance gap with respect to native code. These changes will enable end-users to choose frameworks based mostly on ease of use.

2013 IEEE 29th International Conference on Data Engineering (ICDE), 2013

This paper presents PrPl, a decentralized infrastructure that lets users participate in online so... more This paper presents PrPl, a decentralized infrastructure that lets users participate in online social networking without loss of data ownership. PrPl, short for private-public, has a person-centric architecture--each individual uses a Personal-Cloud Butler service that provides a safe haven for one's personal digital assets and supports sharing with fine-grain access control. A user can choose to run the Butler on

We apply machine learning to thread level speculation, a future hardware framework for paralleliz... more We apply machine learning to thread level speculation, a future hardware framework for parallelizing sequential programs. By using machine learning to determine the parallel regions, the overall performance is nearly as good as the best heuristics for each application.

Proceedings of the 1st ACM Workshop on Mobile Cloud Computing & Services Social Networks and Beyond - MCS '10, 2010

This paper presents PrPl, a decentralized infrastructure that lets users participate in online so... more This paper presents PrPl, a decentralized infrastructure that lets users participate in online social networking without loss of data ownership. PrPl, short for private-public, has a person-centric architecture--each individual uses a Personal-Cloud Butler service that provides a safe haven for one's personal digital assets and supports sharing with fine-grain access control. A user can choose to run the Butler on

IEEE Transactions on Knowledge and Data Engineering, 2015

Many modern software platforms today, including browsers, middleware server architectures, cell p... more Many modern software platforms today, including browsers, middleware server architectures, cell phone operating systems, web application engines, support thirdparty software extensions. This paper proposes InvisiType, an object-oriented approach that enables platform developers to efficiently enforce fine-grained safety checks on thirdparty extensions without requiring their cooperation. This allows us to harness the true power of third-party software by giving it access to sensitive data while ensuring that it does not leak data.

ABSTRACT The convenient way for users to share data these days, be it photos, IM, or just their G... more ABSTRACT The convenient way for users to share data these days, be it photos, IM, or just their GPS locations, is for them to sign up as friends to the same social networking sites. As a result, not only are we inundated by invitations from our friends to join a large number of ...

Proceedings of the VLDB Endowment, 2013

Advertisement-supported social networking portals generally aim to lock in users' data and exploi... more Advertisement-supported social networking portals generally aim to lock in users' data and exploit personal information for ad targeting and other marketing purposes. Because of the network effect, it is not hard to envision a situation where the information of a very large population can end up in the hands of an oligopolistic group or even a sole monopolistic actor. Beyond the obvious privacy concerns, this outcome would clearly pose problems for healthy competition, ultimately harming end-users.

Proceedings of the 2014 ACM SIGMOD international conference on Management of data - SIGMOD '14, 2014

Graph algorithms are becoming increasingly important for analyzing large datasets in many fields.... more Graph algorithms are becoming increasingly important for analyzing large datasets in many fields. Real-world graph data follows a pattern of sparsity, that is not uniform but highly skewed towards a few items. Implementing graph traversal, statistics and machine learning algorithms on such data in a scalable manner is quite challenging. As a result, several graph analytics frameworks (GraphLab, CombBLAS, Giraph, SociaLite and Galois among others) have been developed, each offering a solution with different programming models and targeted at different users. Unfortunately, the "Ninja performance gap" between optimized code and most of these frameworks is very large (2-30X for most frameworks and up to 560X for Giraph) for common graph algorithms, and moreover varies widely with algorithms. This makes the end-users' choice of graph framework dependent not only on ease of use but also on performance. In this work, we offer a quantitative roadmap for improving the performance of all these frameworks and bridging the "ninja gap". We first present hand-optimized baselines that get performance close to hardware limits and higher than any published performance figure for these graph algorithms. We characterize the performance of both this native implementation as well as popular graph frameworks on a variety of algorithms. This study helps endusers delineate bottlenecks arising from the algorithms themselves vs. programming model abstractions vs. the framework implementations. Further, by analyzing the system-level behavior of these frameworks, we obtain bottlenecks that are agnostic to specific algorithms. We recommend changes to alleviate these bottlenecks (and implement some of them) and reduce the performance gap with respect to native code. These changes will enable end-users to choose frameworks based mostly on ease of use.

2013 IEEE 29th International Conference on Data Engineering (ICDE), 2013

This paper presents PrPl, a decentralized infrastructure that lets users participate in online so... more This paper presents PrPl, a decentralized infrastructure that lets users participate in online social networking without loss of data ownership. PrPl, short for private-public, has a person-centric architecture--each individual uses a Personal-Cloud Butler service that provides a safe haven for one's personal digital assets and supports sharing with fine-grain access control. A user can choose to run the Butler on