Events for the 2nd week of September
-
Fall 2019 Joint CSC@USC/CommNetS-MHI Seminar Series
Mon, Sep 09, 2019 @ 02:00 PM - 03:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Kimon Drakopoulos, University of Southern California
Talk Title: Misinformation in platforms: Persuasion and inundation
Abstract: In the first part of the talk, we study information design in social networks. We consider a setting, where agents actions exhibit positive local network externalities. There is uncertainty about the underlying state of the world, which impacts agents payoffs. The platform can choose a signaling mechanism that sends informative signals to agents upon realization of this uncertainty, thereby influencing their actions. We investigate how the platform should design its signaling mechanism to achieve a desired outcome.. We find that in the case where the platform seeks only to minimize misinformation (regardless of the induced engagement), common threshold mechanisms with identical thresholds across agents are optimal. This is in contrast to the engagement maximization setting, where when agents are heterogeneous in terms of their network positions, common threshold mechanisms induce substantially lower engagement than the optimal mechanisms. We also study the frontier of the engagement/misinformation levels that can be achieved via different mechanisms and characterize when common threshold mechanisms achieve optimal tradeoffs.
In the second part of the talk, we study a model of information consumption where consumers sequentially interact with a platform that offers a menu of signals (posts) about an underlying state of the world (fact). At each time, incapable of consuming all posts, consumers screen the posts and only select (and consume) one from the offered menu. We show that in the presence of uncertainty about the accuracy of these posts, and as the number of posts increases, adverse effects such as slow learning and polarization arise. Specifically, we establish that, in this setting, bias emerges as a consequence of the consumer screening process. Namely, consumers, in their quest to choose the post that reduces their uncertainty about the state of the world, choose to consume the post that is closest to their own beliefs. We study the evolution of beliefs and we show that such a screening bias slows down the learning process, and the speed of learning decreases with the menu size. Further, we show that the society becomes polarized during the prolonged learning process even in situations where the society belief distribution was not a priori polarized.
Biography: Kimon Drakopoulos is an Assistant Professor of Operations Management at the Marshall School of Business, University of Southern California. His research focuses on the operations of complex networked systems, social networks, stochastic modeling, game theory and information economics. Kimon, prior to joining USC, completed his PhD at the Laboratory for Information and Decision systems at MIT focusing on the analysis and control of contagion processes on networks.
Host: Ketan Savla, ksavla@usc.edu
More Info: http://csc.usc.edu/seminars/2019Fall/drakopoulos.html
More Information: 190905 Kimon Drakopoulos CSCUSC Seminar.pdf
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 132
Audiences: Everyone Is Invited
Contact: Brienne Moore
Event Link: http://csc.usc.edu/seminars/2019Fall/drakopoulos.html
-
Integrating Blockchain & Big Data
Mon, Sep 09, 2019 @ 07:00 PM - 10:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Wyatt Meldman-Floch, Constellation Labs
Talk Title: Integrating Blockchain & Big Data
Abstract: The main limitation to traditional linear blockchain technology is scalability. Most approaches to scalability improvements utilize L2 solutions such as sharding or partitioning. However, a limitation of these L2 approaches is a lack of resilience to node failures due to the stateful nature of blockchain protocols. Backend systems that can dynamically adapt to changes in throughput or outright resource failure are know as elastic infrastructure, which are a core feature of most tools for large scale data processing. In order to achieve native integration with traditional backend systems, stateful P2P networks need elastic infrastructure. MEME, an online machine learning model created at Constellation Labs was created to address elasticity in stateful peer to peer networks such as blockchain protocols and cryptocurrencies associated to them. MEME is an ensamble model comprised of three known approaches to quantify performance and influence of a participant in P2P networks that can be incorporated with any proof model such as proof of work (PoW), proof of stake (PoS) or proof of reputable observation (PRO). The focus of this presentation is on elastic infrastructure and MEME, an approach for maintaining elasticity in blockchain/DAG clusters created and used by Constellation Labs.
Biography: Wyatt is the CTO and cofounder of Constellation Labs, where he developed an asynchronous DAG protocol to powering a decentralized data marketplace. He is a software engineer based in San Francisco with over six years of professional experience specializing in distributed systems and machine learning. Wyatt's career began at NASAs SETI Institute where he contributed to the Kepler project and implemented an entropy-based algorithm to detect intelligent (alien) communication. Prior to cofounding Constellation Labs, he served as a software engineer for Rally Health, Radius Intelligence and Zignal Labs, where he built scalable data processing pipelines for data mining, distributed graph based NLP models, and stream processing platforms for data enrichment at the Twitter firehose at scale.
Host: Bhaskar Krishnamachari, CCI
More Info: https://www.meetup.com/Hyperledger-Los-Angeles/events/264393286/
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 132
Audiences: Everyone Is Invited
Contact: Brienne Moore
Event Link: https://www.meetup.com/Hyperledger-Los-Angeles/events/264393286/
-
Center for Cyber-Physical Systems and Internet of Things and Ming Hsieh Institute Seminar
Wed, Sep 11, 2019 @ 02:00 PM - 03:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Professor Arun Kumar, Department of Computer Science & Engineering & Halicioglu Data Science Institute, University of California, San Diego
Talk Title: Democratizing Machine Learning-based Data Analytics
Series: Center for Cyber-Physical Systems and Internet of Things
Abstract: As machine learning (ML) permeates data-driven applications in enterprise, Web, and scientific domains, data management and systems bottlenecks in ML are proving increasingly critical. The overarching goal of my research is to mitigate such bottlenecks and improve the efficiency of ML systems and productivity of ML users, which in turn can help reduce costs and democratize ML-based analytics. Toward this grand goal, we are building abstractions, algorithms, and systems to improve the processes of sourcing and preparing data for ML, performing iterative ML model selection, and integrating ML models with data-driven applications.
In this talk, I will give an overview of our recent work on all these fronts, focusing specifically on a new direction that could transform how ML systems are built: multi-query optimization for ML. Drawing on the lessons of decades of work on query optimization in relational systems, I will talk about some of our recent work on connecting linear algebra, learning theory, and optimization theory with scalable system design and implementation to accelerate the model selection process in ML systems. Our approach is a step towards bridging the large gap between current ML system abstractions and the level at which ML users think, has implications for both statistical models and deep learning, and could lay a principled systems foundation for new AutoML frameworks.
Biography: Arun Kumar is an Assistant Professor in the Department of Computer Science and Engineering and Halicioglu Data Science Institute at the University of California, San Diego. He is a member of the Database Lab and Center for Networked Systems and an affiliate member of the AI Group. His primary research interests are in data management and systems for machine learning/artificial intelligence-based data analytics. Systems and ideas based on his research have been released as part of the MADlib open-source library, shipped as part of products from EMC, Oracle, Cloudera, and IBM, and used internally by Facebook, LogicBlox, Microsoft, and other companies. He is a recipient of two SIGMOD research paper awards in 2019 and 2014, three distinguished reviewer awards from SIGMOD/VLDB in 2019 and 2017, the 2016 PhD dissertation award from UW-Madison CS, a 2016 Google Faculty Research Award, a 2018 Hellman Fellowship. Research webpage: https://adalabucsd.github.io/
Host: Paul Bogdan
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 132
Audiences: Everyone Is Invited
Contact: Talyia White
-
Ming Hsieh Institute Seminar Series on Integrated Systems
Fri, Sep 13, 2019 @ 02:00 PM - 03:30 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Dr. Hai Li, Associate Professor, Duke University
Talk Title: Highly Efficient Neuromorphic Computing Systems with Emerging Nonvolatile Memories
Host: Profs. Hossein Hashemi, Mike Chen, Dina El-Damak, and Mahta Moghaddam
More Information: MHI Seminar Series IS - Hai Li.pdf
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 132
Audiences: Everyone Is Invited
Contact: Jenny Lin