Back to Events

Seminar Series

April 11

EECS Special Seminar: Learning-based Learning Systems

Tianqi Chen
Unviersity of Washington

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-04-11 16:00:00 2019-04-11 17:00:00 America/New_York EECS Special Seminar: Learning-based Learning Systems ABSTRACTData, models, and computing are the three pillars that enable machine learning to solve real-world problems at scale. Making progress on these three domains requires not only disruptive algorithmic advances but also systems innovations that can continue to squeeze more efficiency out of modern hardware. Learning systems are in the center of every intelligent application nowadays. However, the ever-growing demand for applications and hardware specialization creates a huge engineering burden for these systems, most of which rely on heuristics or manual optimization.In this talk, I will present a new approach that uses machine learning to automate system optimizations. I will describe our approach in the context of deep learning deployment problem. I will first discuss how to design invariant representations that can lead to transferable statistical cost models, and apply these representations to optimize tensor programs used in deep learning applications. I will then describe the system improvements we made to enable diverse hardware backends. TVM, our end-to-end system, delivers performance across hardware back-ends that are competitive with state-of-the-art, hand-tuned deep learning frameworks. Finally, I will discuss how to generalize our approach to do full-stack optimization of the model, system, hardware jointly, and how to build systems to support life-long evolution of intelligent applications. Host: Martin Rinard BIOTianqi Chen is a Ph.D. candidate in the Paul G. Allen School of Computer Science & Engineering at the University of Washington, working with Carlos Guestrin on the intersection of machine learning and systems. He has created three major learning systems that are widely adopted: XGBoost, TVM, and MXNet(co-creator). He is a recipient of the Google Ph.D. Fellowship in Machine Learning. 32-G449

April 10

EECS Special Seminar: Organizing Computation for High-Performance Graphics and Visual Computing

Jonathan Ragan-Kelley
University of California, Berkeley

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-04-10 16:00:00 2019-04-10 17:00:00 America/New_York EECS Special Seminar: Organizing Computation for High-Performance Graphics and Visual Computing Abstract:In the face of declining returns to Moore’s law, future visual computing applications—from photorealistic real-time rendering, to 4D light field cameras, to pervasive sensing with deep learning—still demand orders of magnitude more computation than we currently have. From data centers to mobile devices, performance and energy scaling is limited by locality (the distance over which data has to move, e.g., from nearby caches, far away main memory, or across networks) and parallelism. Because of this, I argue that we should think of the performance and efficiency of an application as determined not just by the algorithm and the hardware on which it runs, but critically also by the organization of its computations and data. For algorithms with the same complexity—even the exact same set of arithmetic operations—the order and granularity of execution and placement of data can easily change performance by an order of magnitude because of locality and parallelism. To extract the full potential of our machines, we must treat the organization of computation as a first-class concern, while working across all levels, from algorithms and data structures, to programming languages, to hardware.This talk will present facets of this philosophy in systems I have built for image processing, 3D graphics, and machine learning. I will show that, for the data-parallel pipelines common in these data-intensive applications, the possible organizations of computations and data, and the effect they have on performance, are driven by the fundamental dependencies in a given problem. Then I will show how, by exploiting domain knowledge to define structured spaces of possible organizations and dependencies, we can enable radically simpler high-performance programs, smarter compilers, and more efficient hardware. Finally, I will show how we use these structured spaces to unlock the power of machine learning for optimizing systems.Bio:Jonathan Ragan-Kelley is an assistant professor of Computer Science at UC Berkeley. He works on high-efficiency visual computing, including systems, compilers, and architectures for image processing and vision, 3D graphics, and machine learning. He is a recipient of the NSF CAREER award, the William A. Martin and Firestone thesis prizes, and multiple CACM Research Highlights. He was previously a visiting researcher at Google, a postdoc at Stanford, and earned his PhD from MIT in 2014, where he built the Halide language. Halide is used throughout industry to process billions of images every day, from data centers to billions of smartphones. Before Halide, Jonathan built the Lightspeed preview system, which was used on over a dozen films at Industrial Light & Magic and was a finalist for an Academy technical achievement award, and he worked in GPU architecture, compilers, and research at NVIDIA, Intel, and ATI. 32-G449

April 08

EECS Special Seminar: Interactive Autonomy: Learning and Control for Human-Robot Systems

Dorsa Sadigh
Stanford University

Part Of

4:00P
- 5:00P

Location

34-401
Grier Room A & B
Add to Calendar 2019-04-08 16:00:00 2019-04-08 17:00:00 America/New_York EECS Special Seminar: Interactive Autonomy: Learning and Control for Human-Robot Systems Abstract:Today’s society is rapidly advancing towards robotics systems 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. My research is about algorithm design for these autonomous and intelligent systems that interact with people.Today, I plan to talk about: humans, interactions, and societal implications of interactions. I will first discuss our recent results on active learning of humans’ preferences for robotics tasks. We develop data efficient techniques that learn computational models of humans’ preferences and compare our method with learning from demonstration. I will then formalize interactive autonomy, and our approach in design of learning and control algorithms that influence humans’ actions for better safety and coordination. Finally, I will discuss our approach on studying societal implications of autonomous systems. Specifically, I will talk about routing and decision making algorithms for autonomous cars that reduce congestion on mixed-autonomy roads.Bio:Dorsa Sadigh is an assistant professor in Computer Science and Electrical Engineering at Stanford University. Her research interests lie in the intersection of robotics, learning and control theory, and algorithmic human-robot interaction. Specifically, she works on developing efficient algorithms for autonomous systems that safely and reliably interact with people. Dorsa has received her doctoral degree in Electrical Engineering and Computer Sciences (EECS) at UC Berkeley in 2017, and has received her bachelor’s degree in EECS at UC Berkeley in 2012. She is awarded the Amazon Faculty Research Award, the NSF and NDSEG graduate research fellowships as well as the Leon O. Chua departmental award. 34-401

April 03

EECS Special Seminar: Secure Computer Hardware in the Age of Pervasive Security Attacks

Mangjia Yan
University of Illinois at Urbana-Champaign

Part Of

4:00P
- 5:00P

Location

32-G449
Add to Calendar 2019-04-03 16:00:00 2019-04-03 17:00:00 America/New_York EECS Special Seminar: Secure Computer Hardware in the Age of Pervasive Security Attacks Abstract: Recent attacks such as Spectre and Meltdown have shown how vulnerable modern computer hardware is. The root cause of the problem is that computer architects have traditionally focused on performance and energy efficiency. Security has never been a first-class requirement. Moving forward, however, this has to radically change: we need to rethink computer architecture from the ground-up for security.As an example of this vision, in this talk, I will focus on speculative execution in out-of-order processors --- a core computer architecture technology that is the target of the recent attacks. I will describe InvisiSpec, the first robust hardware defense mechanism against speculative (a.k.a transient) execution attacks. The idea is to make loads invisible in the cache hierarchy, and only reveal their presence at the point when they are safe. Once an instruction is deemed safe, our hardware is able to cheaply modify the cache coherence state in a consistent manner. Further, to reduce the cost of InvisiSpec and increase its protection coverage, I propose Speculative Taint Tracking (STT). This is a novel form of information flow tracking that is specifically designed for speculative execution. It reduces cost by allowing tainted instructions to become safe early, and by effectively leveraging the predictor hardware that is ubiquitous in modern processors. Further improvements of InvisiSpec-STT can be attained with new compiler techniques. Finally, I will conclude my talk by describing ongoing and future directions towards designing secure processors. Host: Martin RinardBIO: Mengjia Yan is a Ph.D. student at the University of Illinois at Urbana-Champaign (UIUC), working with Professor Josep Torrellas. Her research interest lies in the areas of computer architecture and hardware security, with a focus on defenses against transient execution attacks and cache-based side channel attacks. Her work has appeared in some of the top venues in computer architecture and security, and has sparked a large research collaboration initiative between UIUC and Intel. Mengjia received the UIUC College of Engineering Mavis Future Faculty Fellow, the Computer Science W.J. Poppelbaum Memorial Award, a MICRO TopPicks in Computer Architecture Honorable Mention, and was invited to participate in two Rising Stars workshops. 32-G449

April 02

EECS Special Seminar: Learning to Synthesize Images

Jun-Yan Zhu
MIT

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-04-02 16:00:00 2019-04-02 17:00:00 America/New_York EECS Special Seminar: Learning to Synthesize Images Abstract: People are avid consumers of visual content. Every day, we watch videos, play games, and share photos on social media. However, there is an asymmetry – while everybody is able to consume visual content, only a chosen few (e.g., painters, sculptors, film directors) are talented enough to express themselves visually. For example, in modern computer graphics workflows, professional artists have to explicitly specify everything “just right” including geometry, materials, and lighting, for a human to perceive an image as realistic. To automate this tedious process, I present several general-purpose machine learning algorithms for image synthesis. Our methods can discover the structure of the visual world from the data itself and learn to synthesize realistic high-dimensional outputs directly. I then demonstrate applications in different fields such as vision, graphics, and robotics, as well as usages by developers and visual artists. Finally, I discuss our ongoing efforts on learning to synthesize 3D objects and high-resolution videos, with the ultimate goal of building machines that can recreate the visual world and help everyone tell their visual stories.Bio: Jun-Yan Zhu is a postdoctoral researcher at MIT CSAIL. He obtained his Ph.D. in computer science from UC Berkeley after studying at CMU and UC Berkeley, and before that, received his B.E. from Tsinghua University. He studies computer graphics, computer vision, and machine learning, with the goal of building intelligent machines, capable of recreating the visual world. He is the recipient of Facebook Fellowship, ACM SIGGRAPH Outstanding Doctoral Dissertation Award, and UC Berkeley EECS David J. Sakrison Memorial Prize for outstanding doctoral research. His work has been covered in the New Yorker, the New York Times, and the Economist. Jun-Yan has served as a Technical Paper Committee member at SIGGRAPH Asia 2018, a guest editor of International Journal of Computer Vision (IJCV), and a co-instructor of the Deep Learning course at Udacity. 32-G449

April 01

EECS Special Seminar: Natacha Crooks, "A Client-Centric Approach to Transactional Datastores"

Natacha Crooks

Part Of

4:00P
- 5:00P

Location

32-G449 Refreshments at 3:45pm
Add to Calendar 2019-04-01 16:00:00 2019-04-01 17:00:00 America/New_York EECS Special Seminar: Natacha Crooks, "A Client-Centric Approach to Transactional Datastores" Abstract: Modern applications must collect and store massive amounts of data. Cloud storage offers these applications simplicity: the abstraction of a failure-free, perfectly scalable black-box. While appealing, offloading data to the cloud is not without challenges. Cloud storage systems often favour weaker levels of isolation and consistency. These weaker guarantees introducebehaviours that, without care, can break application logic. Offloading data to an untrusted third party like the cloud also raises questions of security and privacy. This talk summarises my efforts to improve the performance, the semantics and the security of transactional cloud storage systems. It centers around a simple idea: defining consistency guarantees from the perspective of the applications that observe these guarantees, rather than from the perspective of the systems that implement them. I will discuss how this new perspective brings forth several benefits. First, it offers simpler and cleaner definitions of weak isolation and consistency guarantees. Second, it enables more scalable implementations of existing guarantees like causal consistency. Finally, I will discuss its applications to security: our client-centricperspective allows us to add obliviousness guarantees to transactional cloud storage systems.Host: Nickolai ZeldovichBio: Natacha Crooks is a PhD candidate at the University of Texas at Austin and a visiting student at Cornell University. Her research interests are in distributed systems, distributed computing and databases. She is the recipient of a Google Doctoral Fellowship in Distributed Computing and a Microsoft Research Women Fellowship. 32-G449 Refreshments at 3:45pm

March 21

A unified program synthesis framework for automating end-user programming tasks

Xinyu Wang
University of Texas, Austin

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-03-21 16:00:00 2019-03-21 17:00:00 America/New_York A unified program synthesis framework for automating end-user programming tasks Abstract: Programming has started to become an essential skill for an increasing number of people, including novices without formal programming background. As a result, there is an increasing need for technology that can provide basic programming support to such non-expert computer end-users. Program synthesis, as a technique for automatically generating programs from high-level specifications, has been used to automate real-world programming tasks in a number of application domains (such as spreadsheet programming and data science) that non-expert users struggle with. However, developing specialized synthesizers for these domains is notoriously hard. In this talk, I will describe a unified program synthesis framework that can be applied broadly to automating tasks across different application domains. This framework is also efficient and achieves orders of magnitude improvement in terms of synthesis speed compared to existing techniques. In particular, I have used this framework to build synthesizers for three different application domains and achieved up to 450x speed-up compared to state-of-the-art synthesis techniques. Bio: Xinyu Wang is a PhD candidate at UT Austin advised by Isil Dillig. He works at the intersection of programming languages, software engineering and formal methods. He is interested in developing foundational program synthesis techniques that are applicable to automating real-world programming tasks. 32-G449

March 19

Self-Directed Learning

Deepak Pathak
University of California Berkeley

Part Of

4:00P
- 5:00P

Location

34-401
Grier Combined
Add to Calendar 2019-03-19 16:00:00 2019-03-19 17:00:00 America/New_York Self-Directed Learning Abstract:Generalization, i.e., the ability to adapt to novel scenarios, is the hallmark of human intelligence. While we have systems that excel at recognizing objects, cleaning floors, playing complex games and occasionally beating humans, they are incredibly specific in that they only perform the tasks they are trained for and are miserable at generalization. In this talk, I will present our initial efforts toward endowing artificial agents with a human-like ability to generalize in diverse scenarios. The main insight is to allow the agent to learn general-purpose skills in a completely self-directed manner, without optimizing for any external goal. These skills are then later repurposed to solve complex tasks. I will discuss how this framework can be instantiated to develop curiosity-driven agents (virtual as well as real) that can learn to play games, learn to walk, and learn to perform real-world object manipulation without any rewards or supervision. These self-directed robotic agents, after exploring the environment, can find their way in office environments, tie knots using rope, rearrange object configuration, and compose their skills in a modular fashion.Bio:Deepak Pathak is a Ph.D. candidate in Computer Science at UC Berkeley, advised by Prof. Trevor Darrell and Prof. Alexei A. Efros. His research spans computer vision, machine learning, and robotics. Deepak is a recipient of the Facebook Graduate Fellowship, the NVIDIA Fellowship, and the Snapchat Fellowship, and his research has been featured in popular press outlets, including The Wall Street Journal, The Economist, Quanta Magazine, Wired, and MIT Technology Review. Deepak received his Bachelor's in Computer Science from IIT Kanpur and was a recipient of the Gold Medal and the best undergraduate thesis award. He has also spent time at Facebook, Microsoft and founded VisageMap Inc. which was later acquired by FaceFirst Inc.For papers and open-sourced code see: https://people.eecs.berkeley.edu/~pathak/ 34-401

March 18

EECS Special Seminar: The Next Billion Medical Devices

Edward Wang
University of Washington

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-03-18 16:00:00 2019-03-18 17:00:00 America/New_York EECS Special Seminar: The Next Billion Medical Devices Abstract:Medical care today revolves mainly around the clinic, with devices and care designed with accuracy as the main metric of success. Many medical tests can be accomplished with just a small vial of blood, and in the ICU we can probe a variety of physiological parameters continuously. In the search for accuracy, accessibility, however, has often taken a backseat. In the majority of situations outside of clinical care, there will not be access to that vial of blood or the myriads of sensors on an ICU monitor. Grounded in this premise, my research looks to invent the next generation of "mobile medical devices" that enables medical sensing beyond the current clinical paradigm in two major directions. First, I will discuss how we can re-appropriate existing mobile sensor infrastructure to enable low-cost and widespread medical sensing. The second major theme of my work explores novel wearable sensors to perform continuous health tracking for passive, long-term insights to a person’s health. Through this talk, I will discuss how I have enabled smartphones to perform physiological measurements comparable to their medical device counterparts and developed novel wearable devices that continuously track blood pressure and user activity in a low-power and highly wearable fashion. My work has taken me from the lab bench building prototypes, clinical validations against state-of-the-art medical devices, to working with actual users and patients at villages in the jungles of South America. Finally, I will discuss my future explorations into largescale data-driven development of accessible mobile health systems that will aim to negotiate scale, human-in-the-loop effects, privacy, and accuracy. Bio:Edward Wang is a Ph.D. candidate in Electrical and Computer Engineering in the Ubiquitous Computing (UbiComp) Lab at the University of Washington, advised by Shwetak Patel. His research work explores practical solutions to address real-world medical needs drawn from collaborations with clinicians and world health organizations, but solved in new, creative ways that leverage state-of-the-art applied machine learning, embedded systems, and mobile sensors. He was awarded the NSF Graduate Fellowship, the ARCS Foundation Fellowship, and 3 paper awards during his time at the University of Washington. He received a B.Sci in Engineering from Harvey Mudd College in 2012, where he was advised by Elizabeth Orwin and specialized in biomedical engineering. 32-G449

EECS Special Seminar: Agency in the Era of Learning Systems

Jakob Foerster
University of Oxford

Part Of

3:00P
- 4:00P

Location

34-401
Add to Calendar 2019-03-18 15:00:00 2019-03-18 16:00:00 America/New_York EECS Special Seminar: Agency in the Era of Learning Systems Abstract: We commonly think of machine learning problems, such as machine translation, as supervised tasks consisting of a static set of inputs and desired outputs. Even reinforcement learning, which tackles sequential decision making, typically treats the environment as a stationary black box. However, as machine learning systems are deployed in the real world, these systems start having impact on each other and their users, turning their decision making into a multi-agent problem. It is time we start thinking of these problems as such, by directly accounting for the agency of other learning systems in the environment. In this talk we look at recent advances in the field of multi-agent learning, where accounting for agency can have drastic effects. In the “Bayesian Action Decoder” (BAD) agents directly reason over the beliefs of other agents in order to learn communication protocols in settings with limited public knowledge and actions that can be used to share information. BAD can be seen as a step towards a kind of “theory of mind” for AI agents and achieves a new state-of-the-art on the cooperative, partial-information, card-game Hanabi (“Spiel des Jahres” in 2013), an exciting new benchmark for measuring AI progress. Short Bio:Jakob Foerster recently obtained his PhD in AI at the University of Oxford, under the supervision of Shimon Whiteson. Using deep reinforcement learning (RL) he studies how accounting for agency can address multi-agent problems, ranging from the emergence of communication to non-stationarity, reciprocity and multi-agent credit-assignment. His papers have gained prestigious awards at top machine learning conferences (ICML, AAAI) and have helped push deep multi-agent RL to the forefront of AI research. During his PhD Jakob interned at Google Brain, OpenAI, and DeepMind. Prior to his PhD Jakob obtained a first-class honours Bachelor’s and Master’s degree in Physics from the University of Cambridge and also spent four years working at Goldman Sachs and Google. Previously he has also worked on a number of research projects in systems neuroscience, including work at MIT and research at the Weizmann Institute. 34-401

EECS Special Seminar: On the Foundations of Deep Learning: SGD, Overparametrization, and Generalization

Jason Lee
University of Southern California

Part Of

1:00P
- 2:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-03-18 13:00:00 2019-03-18 14:00:00 America/New_York EECS Special Seminar: On the Foundations of Deep Learning: SGD, Overparametrization, and Generalization ABSTRACT: Deep Learning has had phenomenal empirical successes in many domains including computer vision, natural language processing, and speech recognition. To consolidate and boost the empirical success, we need to develop a more systematic and deeper understanding of the elusive principles of deep learning.In this talk, I will provide theoretical analysis of several elements of deep learning including non-convex optimization, overparametrization, and generalization error. First, we show that gradient descent and many other algorithms are guaranteed to converge to a local minimizer of the loss. For several interesting problems including the matrix completion problem, this guarantees that we converge to a global minimum. Then we will show that gradient descent converges to a global minimizer for deep overparametrized networks. Finally, we analyze the generalization error by showing that a subtle combination of SGD, logistic loss, and architecture combine to promote large margin classifiers, which are guaranteed to have low generalization error. Together, these results show that on overparametrized deep networks SGD finds solution of both low train and test error.BIO: Jason Lee is an assistant professor in Data Sciences and Operations at the University of Southern California. Prior to that, he was a postdoctoral researcher at UC Berkeley working with Michael Jordan. Jason received his PhD at Stanford University advised by Trevor Hastie and Jonathan Taylor. His research interests are in statistics, machine learning, and optimization. Lately, he has worked on high dimensional statistical inference, analysis of non-convex optimization algorithms, and theory for deep learning. 32-G449

March 14

EECS Special Seminar: Protecting Privacy by Splitting Trust

Henry Corrigan-Gibbs
Stanford University

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-03-14 16:00:00 2019-03-14 17:00:00 America/New_York EECS Special Seminar: Protecting Privacy by Splitting Trust ABSTRACT:When the maker of my phone, smart-watch, or web browsercollects data about how I use it, must I trust themanufacturer to protect that sensitive information fromtheft? When I use the cryptographic hardware module in mylaptop, need I trust that it will keep my secrets safe? WhenI use a messaging app to chat with friends, must I trust theapp vendor not to sell the details of my messaging activityfor profit?This talk will show that we can get the functionality wewant from our systems without having to put blind faith inthe correct behavior of the companies collecting our data,building our hardware, or designing our apps. The principleis to split our trust -- among organizations, or devices, orusers. I will introduce new cryptographic techniques andsystems-level optimizations that make it practical to splittrust in a variety of settings. Then, I will present threebuilt systems that employ these ideas, including one thatnow ships with the Firefox browser.BIO: Henry Corrigan-Gibbs is a Ph.D. candidate at Stanford,advised by Dan Boneh. His research interests are in computersecurity, applied cryptography, and online privacy. Henryand his collaborators have received the Best YoungResearcher Paper Award at Eurocrypt 2018, the 2016 CasparBowden Award for Outstanding Research in Privacy EnhancingTechnologies, and the 2015 IEEE Security and PrivacyDistinguished Paper Award, and Henry's work has been citedby IETF and NIST. 32-G449

March 12

EECS Special Seminar: Democratizing Web Automation: Programming for Social Scientists and Other Domain Experts

Sarah Chasins
University of California, Berkeley

Part Of

4:00P
- 5:00P

Location

32-D463
Star Conference Room
Add to Calendar 2019-03-12 16:00:00 2019-03-12 17:00:00 America/New_York EECS Special Seminar: Democratizing Web Automation: Programming for Social Scientists and Other Domain Experts ABSTRACT:We have promised social scientists a data revolution, but it hasn’t arrived. What stands between practitioners and the data-driven insights they want? Acquiring the data. In particular, acquiring the social media, online forum, and other web data that was supposed to help them produce big, rich, ecologically valid datasets. Web automation programming is resistant to high-level abstractions, so end-user programmers end up stymied by the need to reverse engineer website internals—DOM, JavaScript, AJAX. Programming by Demonstration (PBD) offered one promising avenue towards democratizing web automation. Unfortunately, as the web matured, the programs became too complex for PBD tools to synthesize, and web PBD progress stalled.In this talk, I’ll describe how I reformulated traditional web PBD around the insight that demonstrations are not always the easiest way for non-programmers to communicate their intent. By shifting from a purely Programming-By-Demonstration view to a Programming-By-X view that accepts a variety of user-friendly inputs, we can dramatically broaden the class of programs that come in reach for end-user programmers. My Helena ecosystem combines (i) usable PBD-based program drafting tools, (ii) learnable programing languages, and (iii) novel programming environment interactions. The end result: non-coders write Helena programs in 10 minutes that can handle the complexity of modern webpages, while coders attempt the same task and time out in an hour. I’ll conclude with predictions about the abstraction-resistant domains that will fall next—robotics, analysis of unstructured texts, image processing—and how hybrid PL-HCI breakthroughs will vastly expand access to programming.BIOGRAPHY:Sarah Chasins is a Ph.D. candidate at UC Berkeley, advised by Ras Bodik. Her research interests lie at the intersection of programming languages and human-computer interaction. Much of her work is shaped by ongoing collaborations with social scientists, data scientists, and other non-traditional programmers. She has been awarded an NSF graduate research fellowship and a first place award in the ACM Student Research Competition. 32-D463

March 11

EECS Special Seminar: Using Computer Vision to Study Society: Methods and Challenges

Timnit Gebru
Google

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-03-11 16:00:00 2019-03-11 17:00:00 America/New_York EECS Special Seminar: Using Computer Vision to Study Society: Methods and Challenges Abstract: Targeted socio-economic policies require an accurate understanding of a country's demographic makeup. To that end, the United States spends more than 1 billion dollars a year gathering census data such as race, gender, education, occupation and unemployment rates. Compared to the traditional method of collecting surveys across many years which is costly and labor intensive, data-driven, machine learning driven approaches are cheaper and faster--with the potential ability to detect trends in close to real time. In this work, we leverage the ubiquity of Google Street View images and develop a computer vision pipeline to predict income, per capita carbon emission, crime rates and other city attributes from a single source of publicly available visual data. We first detect cars in 50 million images across 200 of the largest US cities and train a model to determine demographic attributes using the detect cars. To facilitate our work, we used a graph based algorithm to collect a challenging fine-grained dataset consisting of over 2600 classes of cars comprised of images from Google Street View and other web sources. Our prediction results correlate well with ground truth income (r=0.82), race, education, voting, sources investigating crime rates, income segregation, per capita carbon emission, and other market research. Data mining based works such as this one can be used for many types of applications--some ethical and others not. I will finally discuss work (inspired by my experiences while working on this project), on auditing and exposing biases found in computer vision systems. Using recent work on exposing the gender and skin type bias found in commercial gender classification systems as a case study, I will discuss how the lack of standardization and documentation in AI is leading to biased systems used in high stakes scenarios. I will end with the concept of AI datasheets for datasets, and model cards for model reporting to standardize information for datasets and pre-trained models, to push the field as a whole towards transparency and accountability. Host: Antonio Torralba.Bio: Timnit Gebru is a research scientist in the Ethical AI team at Google and just finished her postdoc in the Fairness Accountability Transparency and Ethics (FATE) group at Microsoft Research, New York. Prior to that, she was a PhD student in the Stanford Artificial Intelligence Laboratory, studying computer vision under Fei-Fei Li. Her main research interest is in data mining large-scale, publicly available images to gain sociological insight, and working on computer vision problems that arise as a result, including fine-grained image recognition, scalable annotation of images, and domain adaptation. She is currently studying the ethical considerations underlying any data mining project, and methods of auditing and mitigating bias in sociotechnical systems. The New York Times, MIT Tech Review and others have recently covered her work. As a cofounder of the group Black in AI, she works to both increase diversity in the field and reduce the negative impacts of racial bias in training data used for human-centric machine learning models. 32-G449

AlphaGo, Hamiltonian descent, and the computational challenges of machine learning

Chris Maddison
University of Oxford

Part Of

3:00P
- 4:00P

Location

32-D463
Star Conference Room
Add to Calendar 2019-03-11 15:00:00 2019-03-11 16:00:00 America/New_York AlphaGo, Hamiltonian descent, and the computational challenges of machine learning Abstract: Many computational challenges in machine learning involve the three problems of optimization, integration, and fixed-point computation. These three can often be reduced to each other, so they may also provide distinct vantages on a single problem. In this talk, I present a small part of this picture through a discussion of my work on AlphaGo and Hamiltonian descent methods. AlphaGo is the first computer program to defeat a world-champion player, Lee Sedol, in the board game of Go. My work laid the groundwork of the neural net components of AlphaGo, and culminated in our Nature publication describing AlphaGo's algorithm, at whose core hide these three problems. The work introducing Hamiltonian descent methods presents a family of gradient-based optimization algorithms inspired by the Monte Carlo literature and recent work on reducing the problem of optimization to that of integration. These methods can achieve fast linear convergence without strong convexity by using a non-standard kinetic energy to condition the optimization. I conclude by bringing this basic idea back to the study of the classical gradient descent method.Bio: Chris Maddison is a PhD candidate in the Statistical Machine Learning Group in the Department of Statistics at the University of Oxford. He is an Open Philanthropy AI Fellow and spends two days a week as a Research Scientist at DeepMind. His research is broadly focused on the development of numerical methods for deep learning and machine learning. He has worked on methods for variational inference, numerical optimization, and Monte Carlo estimation with a specific focus on those that might work at scale with few assumptions. Chris received his MSc. from the University of Toronto. He received a NIPS Best Paper Award in 2014, and was one of the founding members of the AlphaGo project. 32-D463

March 07

AI for Imperfect-Information Game Settings

Noam Brown
Carnegie Mellon University

Part Of

4:00P
- 5:00P

Location

32-G449 Patil/Kiva
Add to Calendar 2019-03-07 16:00:00 2019-03-07 17:00:00 America/New_York AI for Imperfect-Information Game Settings Abstract: The field of artificial intelligence has had a number of high-profile successes in the domain of perfect-information games. But real-world strategic interactions are not like chess or Go where all participants know the exact state of the world. Instead, they typically involve hidden information, such as in negotiations, cybersecurity attacks, and financial markets. Techniques used in perfect-information games fall apart when applied to such imperfect-information games, with poker serving as the classic example. Libratus is an AI that, in a 120,000-hand competition, decisively defeated four top professionals in heads-up no-limit Texas hold’em poker, the leading benchmark for imperfect-information games and a long-standing challenge problem for AI in general. In this talk I will explain why past techniques intended for perfect-information multi-agent and imperfect-information single-agent settings break down both in theory and in practice in imperfect-information multi-agent settings, and the advances in Libratus and my later work that overcome those challenges. In particular, I will describe new general methods I developed for state-of-the-art equilibrium finding and real-time planning in imperfect-information games. These techniques all have theoretical guarantees in addition to strong empirical performance, and are domain-independent. I will conclude by discussing applications of this work and future research directions in the area of multi-agent artificial intelligence. Host: Tommi Jaakkola.Bio: Noam Brown is a PhD candidate in computer science at Carnegie Mellon University advised by Tuomas Sandholm and a Research Scientist at Facebook AI Research. His research combines computational game theory and machine learning to develop AI systems capable of strategic reasoning in large imperfect-information multi-agent settings. He has applied this research to creating Libratus, the first AI to defeat top humans in no-limit poker, which was one of 12 finalists for Science Magazine's Scientific Breakthrough of the Year. Noam received a NeurIPS Best Paper award in 2017, the 2017 Allen Newell Award for Research Excellence, an Outstanding Paper Honorable Mention at AAAI 2019, and the 2019 Marvin Minsky Medal for Outstanding Achievements in AI. His PhD was supported by an Open Philanthropy Project AI fellowship and a Tencent AI Lab fellowship. 32-G449 Patil/Kiva

March 04

February 21

EECS Special Seminar: Context-Driven Implicit Interactions

Gierad Laput
Carnegie Mellon University

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-02-21 16:00:00 2019-02-21 17:00:00 America/New_York EECS Special Seminar: Context-Driven Implicit Interactions AbstractAs computing proliferates into everyday life, systems that understand people’s context of use are of paramount importance. Regardless of whether the platform is a mobile device, a wearable, or embedded in the environment, context offers an implicit dimension that will become highly important if we are to power more human-centric experiences. Context-driven sensing will become a foundational element for many high-impact applications, from specific domains such as elder care, health monitoring, and empowering people with disabilities, to much broader areas such as smart infrastructures, robotics, and novel interactive experiences for consumers.In this talk, I discuss the construction and evaluation of sensing technologies that can be practically deployed and yet still greatly enhance contextual awareness, primarily drawing upon machine learning to unlock a wide range of applications. I attack this problem area on two fronts: 1) supporting sensing expressiveness via context-sensitive wearable devices, and 2) achieving general-purpose sensing through sparse environment instrumentation. I discuss algorithms and pipelines that extract meaningful signals and patterns from sensor data to enable high-level abstraction and interaction. I also discuss system and human-centric challenges, and I conclude with a vision of how rich contextual awareness can enable more powerful experiences across broader domains. BioGierad is Ph.D. candidate at Carnegie Mellon University’s School of Computer Science. His research in HCI lies at the intersection of interactive systems, sensing, and applied machine learning. He designs, builds, and evaluates novel interactive technologies that greatly enhance input expressivity for users and contextual awareness for devices. His research has been recognized with a Google Ph.D. Fellowship, a Swartz Entrepreneurial Fellowship, a Qualcomm Innovation Fellowship, an Adobe Research Fellowship, and a Disney Research Fellowship. He is also a recipient of the Fast Company Innovation by Design Award, along with 6 Best Paper Awards and Nominations at premier venues in human-computer interaction. He is also Editor-in-Chief of XRDS, ACM’s flagship magazine for students. Homepage: www.gierad.comHost: Stefanie Mueller 32-G449

February 20

Private Event

ROOM CHANGE: EECS Special Seminar Algorithms and data structures in the brain

Saket Navlakha
Salk Institute for Biological Studies

Part Of

4:00P
- 5:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-02-20 16:00:00 2019-02-20 17:00:00 America/New_York ROOM CHANGE: EECS Special Seminar Algorithms and data structures in the brain Abstract: A fundamental challenge in neuroscience is to understand the algorithms that neural circuits have evolved to solve computational problems critical for survival. In this talk, I will describe how the olfactory circuit in the fruit fly brain has evolved simple yet effective algorithms to process and store odors. First, I will describe how fruit flies use a variant of a traditional computer science algorithm (called locality-sensitive hashing) to perform efficient similarity searches. Second, I will describe how this circuit uses a variant of a classic data structure (called a Bloom filter) to perform novelty detection for odors. In both cases, we show that tricks from biology can be translated to improve machine computation, while also raising new hypotheses about neural function. I will conclude by arguing that the search for "algorithms in nature" is not limited to only the brain and could include many other areas of biology, including plant biology.Bio: Saket Navlakha is an assistant professor in the Integrative Biology Laboratory at the Salk Institute for Biological Studies. He received an A.A. from Simon's Rock College in 2002, a B.S. from Cornell University in 2005, and a Ph.D. in computer science from the University of Maryland College Park in 2010. He was then a post-doc in the Machine Learning Department at Carnegie Mellon University before starting his lab at the Salk Institute in 2014. His lab studies algorithms in nature, i.e., how collections of molecules, cells, and organisms process information and solve computational problems. In 2018, he was named a Pew Biomedical Scholar, and in 2019, he was awarded an NSF CAREER award. Host: Nancy Lynch 32-G449

February 19

EECS Special Seminar: Microrobots as the Future of Tools: Designing Effective Platforms and Collaborative Swarms

Daniel Drew
Stanford University

Part Of

3:00P
- 4:00P

Location

32-G449
Patil/Kiva
Add to Calendar 2019-02-19 15:00:00 2019-02-19 16:00:00 America/New_York EECS Special Seminar: Microrobots as the Future of Tools: Designing Effective Platforms and Collaborative Swarms Abstract: In the near future, swarms of millimeter scale robots will be vital and common tools in industrial, commercial, and personal settings. With applications ranging from distributed chemical sensing to tangible 3D interfaces, providing mobility platforms to low-power sensing and actuation nodes will push us that much closer to the dream of ubiquitous computing. In this talk I will present my efforts to develop a flying microrobot, the “ionocraft”, which uses atmospheric ion thrusters to move completely silently and with no mechanical moving parts. Spanning from development of novel MEMS actuators to incorporation of onboard sensor packages for control, I will discuss system design at the resource-constrained edge of robotics. Even given a working mobility platform, a bevy of interdisciplinary challenges remain to make microrobots useful tools; I will further discuss strategies for enabling future autonomous swarm deployments as well as for studying human-robot interaction outside the context of traditional social robotics.Bio: Daniel Drew received his Ph.D. in Electrical Engineering from UC Berkeley under the supervision of Professor Kris Pister. His research focused on the design and fabrication of centimeter-scale robotic systems and human-computer interaction in the context of novel debugging and development tools. He recently began as a postdoctoral scholar at Stanford University in Mechanical Engineering, working with Professor Sean Follmer on human-swarm interaction and swarm platform development. Daniel received an NSF Graduate Research Fellowship in 2013 and an Intelligence Community Postdoctoral Research Fellowship in 2019. 32-G449