Quick links

CS Department Colloquium Series

Data-Driven Security

Date and Time
Monday, March 27, 2017 - 12:30pm to 1:30pm
Location
Friend Center 008
Type
CS Department Colloquium Series
Host
Prof. Arvind Narayanan

Zakir Durumeric
For too long, computer science has approached security in an ad hoc and reactionary manner. In order to make meaningful progress, we need to shift our defensive approach to be data-driven and epidemiological. Over the course of my Ph.D., I have built systems to facilitate a data-driven approach to security, and I have applied this methodology to tackle some of the most pressing real-world security problems. In this talk, I will first highlight two of these systems, ZMap and Censys. Second, I will show how Internet-wide scanning has enabled us to identify weaknesses in cryptographic keys, uncover real-world attacks against email delivery, and guide users in patching vulnerabilities. I will conclude by discussing how, in the future, I want to elevate data-driven security beyond individual systems and tools to make it a fundamental part of the Internet ecosystem.

Zakir Durumeric is a Ph.D. Candidate and Google Research Fellow in Computer Science and Engineering at the University of Michigan where his research focuses on systems and network security. His work has received the IRTF Applied Networking Research Prize and best paper awards from USENIX Security, ACM CCS, and ACM IMC.

Beyond "User Engagement": Designing for Intentional Technology Use

Date and Time
Wednesday, March 15, 2017 - 12:30pm to 1:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
Prof. Arvind Narayanan
Have you ever found yourself checking your phone at an inappropriate moment? Or looking up from a casual game and wondering where the time went? If so, you’re not alone, as people report widespread feelings of dissatisfaction with the way they engage with technology and a lack of control over their own behaviors. “User engagement” is a fundamental design goal for many consumer-facing products, and the resulting technologies are often irresistibly engaging. Though modern technology offers enormous value and convenience, many people report a desire to disengage and an inability to change their habits. In this talk, I will describe a series of studies to understand people’s feelings about their engagement patterns and new systems and tools for both children and adults to support intentional engagement. Based on findings from this work, I will discuss what individuals, developers, and policy-makers can do to tackle this problem and improve our collective self-efficacy.

Alexis Hiniker is a Ph.D. candidate at the University of Washington in the department of Human Centered Design and Engineering where she studies the relationship between technology design and human well-being. She holds a master’s degree in Learning, Design, and Technology from Stanford and a bachelor’s degree in Computer Science from Harvard. She is the technical co-founder of Go Go Games Studios, a startup company that builds educational games for children on the autism spectrum. Her research has been covered by The New York Times, TIME Magazine, NPR and many other prominent national media outlets. And she has been the recipient of numerous accolades for her industry work, including the Parents’ Choice Gold award, the “academy award” of children’s media. Her past and current research has been supported by the National Science Foundation, Sesame Workshop, the Google Anita Borg Scholarship, the Facebook Fellowship Program, the Duca Fund, and more. Before returning to graduate school, Hiniker spent six years as a software engineer and engineering manager at Microsoft working on Windows and the .NET Framework.

Scalable Systems for Fast and Easy Machine Learning

Date and Time
Monday, March 13, 2017 - 12:30pm to 1:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
Prof. Andrew Appel

Machine learning models trained on massive datasets power a number of applications; from machine translation to detecting supernovae in astrophysics. However the end of Moore’s law and the shift towards distributed computing architectures presents many new challenges for building and executing such applications in a scalable fashion.

In this talk I will present my research on systems that make it easier to develop new machine learning applications and scale them while achieving high performance. I will first present programming models that let users easily build distributed machine learning applications. Next, I will show how we can exploit the structure of machine learning workloads to build low-overhead performance models that can help users understand scalability and simplify large scale deployments. Finally, I will describe scheduling techniques that can improve scalability and achieve high performance when using distributed data processing frameworks.

Shivaram Venkataraman is a PhD Candidate at the University of California, Berkeley and is advised by Mike Franklin and Ion Stoica. His research interests are in designing systems and algorithms for large scale data processing and machine-learning. He is a recipient of the Siebel Scholarship and best-of-conference citations at VLDB and KDD. Before coming to Berkeley, he completed his M.S at the University of Illinois, Urbana-Champaign.

Grounding natural language with autonomous interaction

Date and Time
Friday, March 10, 2017 - 12:30pm to 1:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
Prof. Barbara Engelhardt

Photo of Karthik Narasimhan
The resurgence of deep neural networks has resulted in impressive advances in natural language processing (NLP). However, this success is dependent on access to large amounts of structured supervision, often manually constructed and unavailable for many applications and domains. In this talk, I will present novel computational models that integrate reinforcement learning with language understanding to induce grounded representations of semantics using unstructured feedback. These techniques not only enable task-optimized representations which reduce dependence on high quality annotations, but also exploit language in adapting control policies across different environments.  First, I will describe an approach for learning to play text-based games, where all interaction is through natural language and the only source of feedback is in-game rewards. Second, I will exhibit a framework for utilizing textual descriptions to assist cross-domain policy transfer for reinforcement learning. Finally, I will demonstrate how reinforcement learning can enhance traditional NLP systems in low resource scenarios. In particular, I describe an autonomous agent that can learn to acquire and integrate external information to improve information extraction.

Karthik Narasimhan is a PhD candidate working with Prof. Regina Barzilay at CSAIL, MIT. His research interests are in natural language understanding and deep reinforcement learning. His current focus is on developing autonomous systems that can acquire language understanding through interaction with their environment while also utilizing textual knowledge to drive their decision making. His work has received a best paper award at EMNLP 2016 and an honorable mention for best paper at EMNLP 2015. Karthik received a B.Tech in Computer Science and Engineering from IIT Madras in 2012 and an S.M in Computer Science from MIT in 2014.

 

Rethinking Distributed Systems for the Datacenter

Date and Time
Monday, March 6, 2017 - 12:30pm to 1:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
Prof. Michael Freedman

Prof. Dan Ports

Today's most popular applications are deployed as massive-scale distributed systems in the datacenter. Keeping data consistent and available despite server failures and concurrent updates is a formidable challenge. Two well-known abstractions, strongly consistent replication and serializable transactions, can free developers from these challenges by transparently masking failures and treating complex updates as atomic units. Yet the conventional wisdom is that these techniques are too expensive to deploy in high-performance systems.

I will demonstrate a new approach to designing distributed systems that allows strongly consistent distributed systems to be built with little to no performance cost. Taking advantage of the properties and capabilities of the datacenter environment, we can co-design distributed protocols and the network layer. Specifically, I will describe two systems for state machine replication, Speculative Paxos and Network-Ordered Paxos, and one for distributed transaction processing, Eris, built using this approach. They are able to achieve 5-17x performance improvements over conventional designs. Moreover, they achieve performance within 2% of their weakly consistent alternatives, demonstrating that strong consistency and high performance are not incompatible.

Dan Ports is Research Assistant Professor in Computer Science and Engineering at the University of Washington, where he leads the distributed systems research group. His group's research focuses on building practical distributed systems with strong theoretical underpinnings. Prior to joining the faculty at UW in 2015, Dan received the Ph.D. from MIT (2012), where he was advised by Barbara Liskov, and completed a postdoc at UW CSE. His research has been recognized with best paper awards at NSDI and OSDI.

Towards a Theory of Safe and Interactive Autonomy

Date and Time
Tuesday, February 28, 2017 - 4:30pm to 5:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
CS MAE ELE

Today’s society is rapidly advancing towards cyber-physical systems (CPS) that interact and collaborate with humans, e.g., semi-autonomous vehicles interacting with drivers and pedestrians, medical robots used in collaboration with doctors, or service robots interacting with their users in smart homes. The safety-critical nature of these systems requires us to provide provably correct guarantees about their performance in interaction with humans. The goal of my research is to enable such human-cyber-physical systems (h-CPS) to be safe and interactive. I aim to develop a formalism for design of algorithms and mathematical models that facilitate correct-by-construction control for safe and interactive autonomy.

In this talk, I will first discuss interactive autonomy, where we use algorithmic human-robot interaction to be mindful of the effects of autonomous systems on humans, and further leverage these effects for better safety, efficiency, coordination, and estimation. I will then talk about safe autonomy, where we provide correctness guarantees, while taking into account the uncertainty arising from the environment. Further, I will discuss a diagnosis and repair algorithm for systematic transfer of control to the human in unrealizable settings. While the algorithms and techniques introduced can be applied to many h-CPS applications, in this talk, I will focus on the implications of my work for semi-autonomous driving.

Dorsa Sadigh is a Ph.D. candidate in the Electrical Engineering and Computer Sciences department at UC Berkeley. Her research interests lie in the intersection of control theory, formal methods, and human-robot interactions. Specifically, she works on developing a unifying framework for safe and interactive autonomy. Dorsa received her B.S. from Berkeley EECS in 2012. She was awarded the NDSEG and NSF graduate research fellowships in 2013. She was the recipient of the 2016 Leon O. Chua department award and the 2011 Arthur M. Hopkin department award for achievement in the field of nonlinear science, and she received the Google Anita Borg Scholarship in 2016.

Reliability and Performance Challenges at Cloud Scale

Date and Time
Friday, March 3, 2017 - 12:30pm to 1:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
Prof. Kyle Jamieson

Asaf Cidon
Cloud computing is quickly becoming the backbone for critical societal applications, such as self-driving cars, tele-medicine and Internet of Things. However, the unprecedented scale of cloud infrastructure, with millions of servers spread across hundreds of data centers, introduces new fundamental challenges that were not faced by past computer systems. In my talk, I will present two examples of systems, Copysets and Cliffhanger, which leverage an analytical approach to tackle novel challenges introduced by the scale of the cloud. In Copysets, I present a novel replication framework that reduces the probability of data loss by over 10,000 times for the common scenario of simultaneous server failures. In Cliffhanger, I present a key-value cache that dynamically adapts to changing cloud application workloads, which reduces the number of misses by 35% or more. Both of these systems provide an example for my research approach, of building practical systems that tackle novel unstudied problems introduced by the scale of the cloud.

Asaf Cidon is the Vice President, Content Security Services at Barracuda Networks. He currently also leads the cloud caching research project at Stanford. Asaf completed his PhD at Stanford under Mendel Rosenblum and Sachin Katti. His research on cloud and mobile systems was published in NSDI and Sigcomm, and received the Best Student Paper Award in Usenix ATC. Asaf was also the founder and CEO of Sookasa, a cloud storage security startup, which was acquired by Barracuda Networks. His research focuses on how to provide reliability and performance guarantees in large-scale cloud environments, and was adopted by several companies, including Facebook, NetApp, and Chartbeat.

Towards New Systems for Mobile/Cloud Applications

Date and Time
Thursday, March 2, 2017 - 12:30pm to 1:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
Prof. Jennifer Rexford

Irene Zhang

The proliferation of datacenters, smartphones, personal sensing and tracking devices, and home automation products is fundamentally changing the applications we interact with daily.  Today's popular user applications are no longer limited to a single desktop computer but now commonly span many mobile devices and cloud servers. As a result, existing systems often do not meet the needs of modern mobile/cloud applications.  In this talk, I will present three systems designed to tackle the challenges of mobile/cloud applications: Sapphire, Diamond and TAPIR.  These systems represent a first step towards better end-to-end support for mobile/cloud applications in runtime management, data management, and distributed transactional storage. Together, they significantly simplify the development of new mobile/cloud applications.

Irene Zhang is a fifth-year PhD student at the University of Washington. She works with Hank Levy and Arvind Krishnamurthy in the Computer Systems Lab. Her current research focuses on systems for large-scale, distributed applications, including distributed runtime systems and transactional storage systems. Before starting her PhD, Irene worked for three years at VMware in the virtual machine monitor group. Irene received her S.B. and M. Eng. in computer science from MIT, where she worked with Frans Kaashoek in the Parallel and Distributed Operating Systems Lab.

Low Latency and Strong Guarantees for Scalable Storage

Date and Time
Monday, February 27, 2017 - 12:30pm to 1:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
Prof. Nick Feamster

Photo of Professor Wyatt Lloyd
Scalable storage systems, where data is sharded across many machines, are necessary to support web services whose data is too large for a single machine to handle.  An ideal system would provide the lowest latency—to make the web services built on top of it fast—and the strongest guarantees—to make programming the web service easier.  Theoretical results prove that such an ideal system is impossible, but all hope is not lost!  Our work has made progress on this problem from both directions: providing stronger guarantees for low latency systems and providing lower latency for systems with strong guarantees.  I will cover one of these systems, Rococo, in detail.  I will also touch on our recent impossibility result, the SNOW Theorem, and how it guided us in building new systems with latency-optimal read-only transactions.

Wyatt Lloyd is a third-year Assistant Professor of Computer Science at the University of Southern California.  His research interests include the theory, design, implementation, evaluation, and deployment of large-scale distributed systems.  He received his Ph.D. from Princeton University in 2013 for the COPS and Eiger systems, which demonstrated stronger semantics were compatible with low latency for scalable geo-replicated storage.  He then spent a year as a Postdoctoral Researcher at Facebook, and he continues to collaborate with its engineers on projects related to media processing, storage, and delivery.

Putting language in context: social networks and discourse structures for robust language technology

Date and Time
Thursday, February 16, 2017 - 12:30pm to 1:30pm
Location
Computer Science Small Auditorium (Room 105)
Type
CS Department Colloquium Series
Host
Prof. Elad Hazan

Jacob Eisenstein

The creation of software that can use and understand natural human language would be a transformative technological advance, enabling computational inference to be brought to bear on the vast stores of knowledge that are encoded only as text. Supervised machine learning approaches have brought this possibility into view, but existing methods rely on annotated datasets that focus overwhelmingly on a narrow set of news texts, while failing to generalize to high-impact application domains such as social media and electronic health records. Human readers and listeners successfully comprehend language under difficult circumstances by relying on various forms of contextual information -- knowing who is speaking, and what they might be trying to say. My research brings this same contextual awareness to automated language processing, using deep learning architectures that are constructed to reflect theoretical ideas from sociolinguistics and discourse semantics. I will describe three applications of this methodology: (1) incorporating sociolinguistic variation into document classification; (2) linking textual references to canonicalized entities; (3) predicting the discourse relations that hold between sentences. I will also briefly describe research that uses computational linguistic analysis to obtain new evidence on sociocultural affinity and influence.


Jacob Eisenstein is an Assistant Professor in the School of Interactive Computing at Georgia Tech. He works on statistical natural language processing, focusing on computational sociolinguistics, social media analysis, discourse, and machine learning. He is a recipient of the NSF CAREER Award, a member of the Air Force Office of Scientific Research (AFOSR) Young Investigator Program, and was a SICSA Distinguished Visiting Fellow at the University of Edinburgh. His work has also been supported by the National Institutes for Health, the National Endowment for the Humanities, and Google. Jacob was a Postdoctoral researcher at Carnegie Mellon and the University of Illinois. He completed his Ph.D. at MIT in 2008, winning the George M. Sprowls dissertation award. Jacob's research has been featured in the New York Times, National Public Radio, and the BBC. Thanks to his brief appearance in If These Knishes Could Talk, Jacob has a Bacon number of 2.

Follow us: Facebook Twitter Linkedin