Doctoral Student Seminar: ​Kyle Singer, Justin Deters Student Seminar: ​Kyle Singer, Justin Deters2019-11-22T06:00:00Z12:30pm1:30pmLopata 101<p><strong rtenodeid="13" style="text-decoration: underline;">​Kyle Singer</strong><br rtenodeid="15"/></p><p><strong>Title: </strong>Scheduling I/O Latency-Hiding Futures in Task-Parallel Platforms</p><p><strong>Abstract: </strong>Task parallelism research has traditionally focused on optimizing computation-intensive applications. Due to the proliferation of commodity parallel processors, there has been recent interest in supporting interactive applications. Such interactive applications frequently rely on I/O operations that require few processing cycles but may incur significant latency to complete. In order to increase performance, when a particular thread of control is blocked on an I/O operation, ideally we would like to hide this latency by using the processing resources to do other ready work instead of blocking or spin waiting on this I/O. There has been limited prior work on hiding this latency and only one result that provides a theoretical bound for interactive applications that use I/O operations.</p><p>In this work, we propose a task parallel platform that supports I/O operations using the futures abstraction and a corresponding scheduler that schedules the I/O operations while hiding their latency. We provide a theoretical analysis of our scheduling algorithm that shows our algorithm provides better execution time guarantees than prior work. We also implemented the algorithm in a practically efficient prototype library that runs on top of the Cilk-F runtime, a runtime system that supports futures within the context of the Cilk Plus language, and performed experiments that demonstrate the efficiency of our implementation.<br/></p><p><strong style="text-decoration-line: underline;">Justin Deters</strong><br/></p><p><strong>Title: </strong>Nobody Gets a Free Lunch: The Challenges of Measuring Work Inflation on Parallel Platforms<br/></p><p><strong>Abstract:  </strong>Writing parallel programs is hard. What’s more, writing parallel programs that perform well is even harder. One of the primary reasons parallel programs are not preformant is work inflation. This is the effect where the total time that the computation spends doing useful work “inflates” due to non-optimal interactions with the underlying hardware. We seek to create a parallel tool that can diagnose and suggest solutions for the work inflation effects in parallel programs. However, in order to properly diagnose work inflation, one must be able to accurately measure it. This is challenging because the very act of measuring work inflation can negatively impact performance as the measurement itself perturbs the system. We present the difficulties faced in accurately measuring work inflation on parallel platforms as motivation for the continuing research into designing and implementing a parallel work inflation tool.<br rtenodeid="8"/></p>
CSE Colloquia Series-Phebe Vayanos Colloquia Series-Phebe Vayanos2019-11-22T06:00:00Z11:00 a.m.Lopata Hall, Room 101<p style="text-align: center;"><strong>Robust Active Preference Elicitation to Learn the Priorities of Policy-Makers at the Los Angeles Homeless Services Authority</strong></p><p><strong>Abstract</strong></p><p>Motivated by our collaboration with the Los Angeles Homeless Services Authority, the authority in charge of allocating housing resources and services to those experiencing homeless in LA, we consider the problem faced by a recommender system which seeks to offer a user with unknown preferences their favorite item among a potentially infinite collection. Before making a recommendation, the system has the opportunity to elicit the user's preferences by making a moderate number of queries. We take the point of view of a risk-averse recommendation system which only possesses limited, set-based, information on the user utility function and investigate two complementary settings. In the first setting, each query corresponds to a pairwise comparison, in the spirit of choice-based conjoint analysis. In the second setting, each query asks the user to rate an item on a scale from 0 to 1. We show that these problems can be formulated as multi-stage robust optimization problems with decision-dependent information discovery and propose reformulations in the form of mixed-binary linear problems that, combined with decomposition techniques, can be solved efficiently with off-the-shelf solvers. We evaluate the performance of our approaches on both synthetic and real-world data from the Homeless Management Information System where we learn the preferences of policy-makers in terms of characteristics (fairness-efficiency-interpretability trade-offs) of a policy for allocating housing resources. Our results illustrate that our framework outperforms state-of-the-art techniques from the literature.</p><p>This work is based on two papers, one joint with Duncan McElfresh, Yingxiao Ye, John Dickerson, and Eric Rice and the other with Angelos Georghiou and Han Yu.</p><p><strong>Biography</strong></p><p>Phebe Vayanos is an Assistant Professor of Industrial & Systems Engineering and Computer Science at the University of Southern California. She is also an Associate Director of the CAIS Center for Artificial Intelligence in Society at USC. Her research aims to address fundamental questions arising in data-driven optimization (a.k.a. prescriptive analytics) with aim to tackle real-world decision- and policy-making problems in uncertain and adversarial environments. Her work is motivated by resource allocation problems that are important for social good, such as those arising in public health, public safety and security, public housing, biodiversity preservation, and education. She is also interested in issues surrounding fairness, efficiency, and interpretability in resource allocation. Prior to joining USC, she was lecturer in the Operations Research and Statistics Group at the MIT Sloan School of Management, and a postdoctoral research associate in the Operations Research Center at MIT. She holds a PhD degree in Operations Research and an MEng degree in Electrical & Electronic Engineering, both from Imperial College London.<br/></p>Sanmay Das
No Classes: Thanksgiving Break Classes: Thanksgiving Break2019-11-27T06:00:00Z
No Classes: Thanksgiving Break Classes: Thanksgiving Break2019-11-28T06:00:00Z
No Classes: Thanksgiving Break Classes: Thanksgiving Break2019-11-29T06:00:00Z
No Classes: Thanksgiving Break Classes: Thanksgiving Break2019-11-30T06:00:00Z
No Classes: Thanksgiving Break Classes: Thanksgiving Break2019-12-01T06:00:00Z
Doctoral Student Seminar: ​Funda Atik, Zihao Deng Student Seminar: ​Funda Atik, Zihao Deng2019-12-06T06:00:00Z12:30pm1:30pmLopata 101<p><strong rtenodeid="13" style="text-decoration: underline;">​Zihao Deng</strong><strong style="text-decoration: underline;"></strong><br rtenodeid="12"/></p><p><strong>Title:</strong> Syntax-Guided Semantic Role Labeling<br/></p><p><strong>Abstract: </strong>Semantic role labeling (SRL) is a fundamental task in natural language understanding. The task is to extract the predicate-argument structure of a sentence, determining "who did what to whom", "when", "where", etc. Both capabilities are useful in several downstream tasks such as question answering (Shen and Lapata, 2007) and open information extraction (Fader et al., 2011). Recently, the NLP community has seen the excitement around the self-attention-based model that makes heavy use of pretraining based on language modeling (Peters et al., 2018; Radford et al., 2018). We explore the use of syntax information to guide the attention of such model to improve the performance on SRL task. <br/></p><p><br/></p><p><br/></p>
Last Day of Classes Day of Classes2019-12-06T06:00:00Z
First Day of Winter Final Exams Day of Winter Final Exams2019-12-09T06:00:00Z
Last Day of Winter Final Exams Day of Winter Final Exams2019-12-18T06:00:00Z
Faculty Focused Researcher Symposium Focused Researcher Symposium2020-01-09T06:00:00Z<p>The Research Education & Information Office within the Office of the Vice Chancellor for Research (OVCR) will host a faculty focused symposium on a variety of topics pertaining to effective research management. The time and location will be announced at a later date.<br/></p><p> The event will help researchers navigate the complexities of the research environment by featuring a variety of sessions on topics such as:</p><ul><li>finding funding</li><li>clinical trials</li><li>working with international collaborators</li><li>data management</li><li>writing a protocol</li><li>managing relationships within your lab</li><li>postdoc recruitment and management</li><li>good clinical practice training<br/></li><li>entrepreneurship</li><li>much more!</li></ul><p>In order to provide sessions that are of the most interest to the Washington University research community, we ask you to complete this <a href="">brief (two-minute) survey</a> designed for faculty input. Your feedback will help shape the symposium and allow us to provide programming of most benefit to WashU faculty.<br/></p>