Events

April 28, 2018

The Future of Vision, Learning and Graphics: A special workshop to celebrate Bill Freeman's 60th

9:00A

- 5:00P

Location

32-123

MIT Stata Center, Kirsch Auditorium

Add to Calendar 2018-04-28 9:00:00 2018-04-28 17:00:00 America/New_York The Future of Vision, Learning and Graphics: A special workshop to celebrate Bill Freeman's 60th Agenda:9-9:20 Andrew Zisserman (Oxford) Visual Sounds and Voices9:20-9:40 Eero Simoncelli (NYU) EigenDistortions of Hierarchical Image Representations9:40-10:00 Yair Weiss (Hebrew University) Invariances in Deep Networks and Shiftable Transforms10:00-10:20 Miki Rubinstein (Google) Looking to Listencoffee break10:50-11:10 Ted Adelson (MIT) Junctions: The Road Not Taken11:10-11:30 Jonathan Yedidia (Analog Devices) A Message Passing Algorithm for General Optimization Problems11:30-11:50 Josh Tenenbaum (MIT) Style, Content, and MetaLearning11:50-12:10 Alyosha Efros (Berkeley) The "Corrupt Professor's Algorithm" and other datadriven adventureslunch14:00-14:20 Ce Liu (Google) World of Motion: from papers to products serving billions of users14:20-14:40 Pietro Perona (Caltech) Seeing into Darkness14:40-15:00 Fredo Durand (MIT) Revealing the Invisible15:00-15:20 Rob Fergus (Facebook AI Research) Synthesizing Future Video Framescoffee break16:00-16:20 Antonio Torralba (MIT) Accidental Pinholes ++16:20-16:40 flash talks by Bill's former students and postdocs16:40-17:00 Bill Freeman (MIT) closing remarks 32-123 Belfer sarah_donahue@hks.harvard.edu

April 30, 2018

Thesis Defense: Flexible models for understanding and shaping complex populations

Jonas Mueller

MIT CSAIL

3:30P

- 4:30P

Location

32-G882 (Hewlett Room)

Refreshments

3:15P

Add to Calendar 2018-04-30 15:30:00 2018-04-30 16:30:00 America/New_York Thesis Defense: Flexible models for understanding and shaping complex populations Data analysis is often driven by the goals of understanding or shaping of some population of interest. The first of these two objectives aims to produce insights regarding characteristics of the underlying population, often to facilitate scientific understanding. Crucially, this requires models which produce results that are highly interpretable to the analyst. On the other hand, notions of interpretability are not necessarily as central for determining how to shape populations, where the aim is to build data-driven systems which learn how to act upon individuals in a manner that maximally improves certain outcomes of interest across the population. In this thesis, we develop interpretable (yet flexible) modeling frameworks for addressing the former goal, as well as black-box nonparametric methods for addressing the latter. Throughout, we demonstrate various empirical applications of our algorithms, primarily in the biological context of modeling gene expression in large cell populations.For better understanding populations, we introduce two nonparametric models that can accurately reflect interesting characteristics of complex distributions without reliance on restrictive assumptions, while simultaneously remaining highly interpretable through their use of the Wasserstein (optimal transport) metric to summarize changes over an entire population. One such approach is principal differences analysis, a projection-based technique that interpretably characterizes differences between two arbitrary high-dimensional probability distributions. Another approach is the nonparametric TRENDS model, which quantifies the underlying effects of temporal progression in a sequence of distributions that also vary due to confounding noise. While the aforementioned techniques fall under the frequentist regime, we subsequently present a Bayesian framework for the task of shaping populations. Drawing upon the Gaussian process toolkit, our method learns how to best conservatively intervene upon heterogeneous populations in settings with limited data and substantial uncertainty about the underlying relationship between actions and outcomes. 32-G882 (Hewlett Room) Belfer sarah_donahue@hks.harvard.edu

May 01, 2018

Cybersecurity: Technology, Application and Policy Online Course | Starts May 1, 2018

12:00A

- 11:00P

Location

MIT xPro Online Course

Add to Calendar 2018-05-01 0:00:00 2018-05-01 23:00:00 America/New_York Cybersecurity: Technology, Application and Policy Online Course | Starts May 1, 2018 Course Description:Are you and your organization safe from cyberattacks? Discover how to protect yourself and your company with MIT xPRO’s Cybersecurity course. In partnership with MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), this fast-paced, 6-week online course provides a holistic look at cybersecurity technologies, techniques, and systems.Taught by a team of world-renowned security experts in hardware, software, cryptography, and policy from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), the Cybersecurity course provide a comprehensive view of key technologies and techniques. Through video lectures and case studies, you will develop a broad understanding of technological advances in cybersecurity and how they apply to today’s systems. By the end of the course, you’ll be equipped with new tools to help develop and implement your own cybersecurity solutions. Faculty:Daniela Rus, Director, MIT CSAIL, and Faculty Co-DirectorHoward Shrobe, Principal Research Scientist, MIT CSAIL, and Faculty Co-DirectorAdam Chlipala, Associate Professor, MIT CSAILDavid Clark, Senior Research Scientist, MIT CSAIL MIT xPro Online Course Belfer sarah_donahue@hks.harvard.edu

Internet of Things: Roadmap to a Connected World Online Course | Starts May 1, 2018

12:00A

- 11:00P

Location

MIT xPro Online Course

Add to Calendar 2018-05-01 0:00:00 2018-05-01 23:00:00 America/New_York Internet of Things: Roadmap to a Connected World Online Course | Starts May 1, 2018 By 2020, there will be 50 billion devices connected to the Internet. How will you and your organization capitalize on this tremendous opportunity? Join this new evolution in hardware, software, and data by registering for this six-week online offering.While the promise of the Internet of Things (IoT) brings many new business prospects, it also presents significant challenges ranging from technology architectural choices to security concerns. MIT xPRO’s Internet of Things: Roadmap to the Connected World course offers important insights on how to overcome these challenges and thrive in this exciting space. At the end of the program, you will be able to understand how to develop and implement your own IoT technologies, solutions, and applications.Faculty:Daniela Rus, Director, MIT CSAIL, and Faculty Co-DirectorSanjay Sarma, VP for Open Learning, MIT ODLTim Berners-Lee, Professor, MIT CSAILSam Madden, Faculty Co-Director, STL@CSAIL MIT xPro Online Course Belfer sarah_donahue@hks.harvard.edu

Thesis Defense: Computational Perception for Multimodal Document Understanding

Zoya Bylinskii

MIT CSAIL

10:00A

- 11:00A

Location

Allen&Haus RLE seminar rooms (36-462 & 36-428)

Refreshments

9:45A

Add to Calendar 2018-05-01 10:00:00 2018-05-01 11:00:00 America/New_York Thesis Defense: Computational Perception for Multimodal Document Understanding Multimodal documents occur in a variety of forms, as graphs in technical reports, diagrams in textbooks, and graphic designs in bulletins. Humans can efficiently parse the visual and textual information contained within to make decisions on topics including business, healthcare, and science. Building the computational tools to understand multimodal documents can therefore have important applications for web search, information retrieval, automatic captioning, and design tools. In this talk, I will discuss our machine learning approaches for detecting and parsing the visual and textual elements in multimodal documents for topic prediction and automatic summarization. Inspired by human perception, I will also present our models that predict where people look in graphic designs and information visualizations, along with the interactive design applications that they enable. The work in this thesis makes contributions to the fields of human vision, computer vision, and human-computer interaction.Committee: Fredo Durand, Aude Oliva, Hanspeter Pfister, Rob Miller, Bill Freeman Allen&Haus RLE seminar rooms (36-462 & 36-428) Belfer sarah_donahue@hks.harvard.edu

Avian Inspired Design

David Lentink

Stanford University

11:00A

- 12:00P

Location

32-G449 Patil/Kiva

Add to Calendar 2018-05-01 11:00:00 2018-05-01 12:00:00 America/New_York Avian Inspired Design Many organisms fly in order to survive and reproduce. My lab focusses on understanding bird flight to improve flying robots—because birds fly further, longer, and more reliable in complex visual and wind environments. I use this multidisciplinary lens that integrates biomechanics, aerodynamics, and robotics to advance our understanding of the evolution of flight more generally across birds, bats, insects, and autorotating seeds. The development of flying organisms as an individual and their evolution as a species are shaped by the physical interaction between organism and surrounding air. The organism’s architecture is tuned for propelling itself and controlling its motion. Flying animals and plants maximize performance by generating and manipulating vortices. These vortices are created close to the body as it is driven by the action of muscles or gravity, then are ‘shed’ to form a wake (a trackway left behind in the fluid). I study how the organism’s architecture is tuned to utilize these and other aeromechanical principles to compare the function of bird wings to that of bat, insect, and maple seed wings. The experimental approaches range from making robotic models to training birds to fly in a custom-designed wind tunnel as well as in visual flight arena’s—and inventing methods to 3D scan birds and measure the aerodynamic force they generate—nonintrusively—with a novel aerodynamic force platform. The studies reveal that animals and plants have converged upon the same solution for generating high lift: A strong vortex that runs parallel to the leading edge of the wing, which it sucks upward. Why this vortex remains stably attached to flapping animal and spinning plant wings is elucidated and linked to kinematics and wing morphology. While wing morphology is quite rigid in insects and maple seeds, it is extremely fluid in birds. I will show how such ‘wing morphing’ significantly expands the performance envelope of birds during flight, and will dissect the mechanisms that enable birds to morph better than any aircraft can. Finally, I will show how these findings have inspired my students to design new flapping and morphing aerial robots.Bio: Professor Lentink's multidisciplinary lab studies how birds fly to develop better flying robots—integrating biomechanics, fluid mechanics, and robot design. He has a BS and MS in Aerospace Engineering (Aerodynamics, Delft University of Technology) and a PhD in Experimental Zoology cum laude (Wageningen University). During his PhD he visited the California institute of Technology for 9 months to study insect flight. His postdoctoral training at Harvard was focused on studying bird flight. Publications range from technical journals to cover publications in Nature and Science. He is an alumnus of the Young Academy of the Royal Netherlands Academy of Arts and Sciences, recipient of the Dutch Academic Year Prize, the NSF CAREER award, he has been recognized in 2013 as one of 40 scientists under 40 by the World Economic Forum, and he is the inaugural winner of the Steven Vogel Young Investigator Award from the Journal Bioinspiration & Biomimetics for early career brilliance. 32-G449 Patil/Kiva Belfer sarah_donahue@hks.harvard.edu

May 02, 2018

Human-Computer Interaction User Experience Design Online Course | Starts May 2, 2018

12:00A

- 11:00P

Location

GetSmarter and MIT xPro Online Course

Add to Calendar 2018-05-02 0:00:00 2018-05-02 23:00:00 America/New_York Human-Computer Interaction User Experience Design Online Course | Starts May 2, 2018 The human-computer interaction (HCI) industry is dynamic and ever-changing and, as a result, requires professionals committed to improving the discipline with a focus on the future.Eight experts from MIT CSAIL have dedicated their research efforts to exploring the ways in which HCI can be enhanced and now they’re opening that research up to you.You have the opportunity to access their work to imagine your own future for HCI and the user experience by reflecting on non-traditional interaction, speech recognition, and rapid prototyping, among others.Take your understanding beyond interface by integrating experimental HCI into your work, all backed by the research of expert MIT faculty and verified with a certificate from MIT xPRO.Faculty:Daniela Rus, Director, MIT CSAIL, Faculty Co-DirectorDaniel Jackson, Professor, MIT CSAILRandall Davis, Professor, MIT CSAILWojciech Matusik, Professor, MIT CSAIL GetSmarter and MIT xPro Online Course Belfer sarah_donahue@hks.harvard.edu

Impact of negative selection on the genetic architecture of diseases and complex traits

Alkes Price

Harvard T.H. Chan School of Public Health

Part Of

Bioinformatics Seminar Series 2018

11:30A

- 1:00P

Location

32-G575

Refreshments

11:15A

Add to Calendar 2018-05-02 11:30:00 2018-05-02 13:00:00 America/New_York Impact of negative selection on the genetic architecture of diseases and complex traits Abstract: It is widely known that negative selection against geneticvariants that reduce fitness causes them to be enriched forlower-frequency variants, so that lower-frequency variants have largercausal effects on diseases and complex traits. Here, we explore twoother ways in which negative selection impacts disease and traitarchitectures. First, we show that (conditional on minor allelefrequency) variants with low levels of linkage disequilibrium havelarger causal effects. We show that much of this signal can beexplained by the fact that (conditional on minor allele frequency)more recent variants have larger causal effects, since negativeselection has had less time to remove them; for example, the youngest20% of common variants explain 4x more heritability than the oldest20% of common variants. Second, we show that functional annotationsstrongly impacted by negative selection have larger enrichment forlow-frequency variant heritability compared to their enrichment forcommon variant heritability, both empirically and in forwardsimulations. Cell-type-specific regulatory annotations that areenriched for common variant heritability tend to be similarly enrichedfor low-frequency variant heritability for most annotations andtraits, but more enriched for brain-related annotations and traits.For example, H3K4me3 marks in brain DPFC explain 57±12% oflow-frequency variant heritability vs. 12±2% of common variantheritability for neuroticism, implicating the action of negativeselection on low-frequency variants affecting gene regulation in thebrain. 32-G575 Belfer sarah_donahue@hks.harvard.edu

Artificial Intelligence: Implications for Business Strategy Online Course | Starts May 2, 2018

12:00P

- 1:00P

Location

GetSmarter and MIT xPro Online Course

Add to Calendar 2018-05-02 12:00:00 2018-05-02 13:00:00 America/New_York Artificial Intelligence: Implications for Business Strategy Online Course | Starts May 2, 2018 Reports reveal that the majority of enterprise executives say AI builds new roles and even makes workers more productive.*To capitalize on the value automation could bring to your business, you need the ability to appreciate its implications and the confidence to apply it in your teams and organizations.This online program from the MIT Sloan School of Management and the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) challenges common misconceptions surrounding AI and will equip and encourage you to embrace AI as part of a transformative toolkit.With a focus on the organizational and managerial implications of these technologies, rather than on their technical aspects, you’ll leave this course armed with the knowledge and confidence you need to pioneer its successful integration in business.Faculty:Daniela Rus, Director, MIT CSAIL, and Faculty Co-DirectorTom Malone, Professor, MIT Sloan School of Management and Faculty Co-DirectorRegina Barzilay, Professor, MIT CSAILPatrick Winston, Professor, MIT CSA GetSmarter and MIT xPro Online Course Belfer sarah_donahue@hks.harvard.edu

Sanctum: Minimal Hardware Extensions for Strong Software Isolation

Ilia Lebedev

MIT

2:00P

- 3:00P

Location

32-G882 (Hewlett Room)

Add to Calendar 2018-05-02 14:00:00 2018-05-02 15:00:00 America/New_York Sanctum: Minimal Hardware Extensions for Strong Software Isolation ABSTRACTWith the increasing prevalence of remote computation via cloud services and ubiquitous IoT devices, we must accept the security vulnerabilities inevitably found in these complex systems. Software systems built in practice are far too large for contemporary formal verification, as they must include all privileged software systems in the trusted computing base of any unprivileged workload. The staggering complexity of modern hardware further complicates matters, undermining simple guarantees such as process isolation.TXT, SGX, TrustZone and similar work attempt to employ trusted hardware to construct an isolated software container and exclude some software from the TCB. While well-intentioned, these defy security analysis and present complex or vague threat models, all undermining their potential role as foundations for secure systems.With our work (the Sanctum processor), we advocate for a transparent system with isolated containers. We offer integrity and confidentiality guarantees against a broad and straightforward threat model (a remote adversary capable of subverting arbitrary software on the system). While this work is certainly not the end-all to computer systems security, we propose a compelling threat model to inform future systems, and bootstrap one trusted platform on which secure systems may be built.BIOIlia Lebedev is a PhD candidate at the Computer Science and Artificial Intelligence Lab at MIT. He works in the area of computer systems security, currently focused on bootstrapping trust for remotely attested computation. He deeply values accessible education and has received the 2016 Hazen Award for outstanding teaching. Outside of his academic career, he is a sailing ship captain, a bartender, and a generally questionable individual. 32-G882 (Hewlett Room) Belfer sarah_donahue@hks.harvard.edu

Cybersecurity@CSAIL Lecture Series | Cybersecurity & AI: A Venture into Reality

Harold Moss, Senior Director of Strategy and Business Development

Akamai Technologies

3:00P

- 4:00P

Location

Ray and Maria Stata Center

Star Conference Room, 32-D463

Add to Calendar 2018-05-02 15:00:00 2018-05-02 16:00:00 America/New_York Cybersecurity@CSAIL Lecture Series | Cybersecurity & AI: A Venture into Reality Abstract:Discussion about the role of AI for cyber professionals as well as how it is weaponized and how that might evolve. In this talk we will discuss how AI is being used today to attack organizations, and how AI for cyber professionals will evolve to be symbiotic with Security analysts.Bio:Harold Moss is a Sr. Director of Strategy and Business development for Akamai’s flourishing web security business. Leveraging his experience as a consumer of technology to protect the enterprise, as well as a purveyor of security technology Harold is focused on identifying opportunities for better protecting Akamai customers. Prior to joining Akamai Harold led Architecture, Strategy and technology for companies such as EMC in the global CISO office, and held various roles in emerging technologies with a focus on security for IBM. Ray and Maria Stata Center Belfer sarah_donahue@hks.harvard.edu

Dertouzos Distinguished Lecture: From Utopia to Dystopia in 29 Short Years

Tim Berners-Lee

W3C/CSAIL

4:30P

- 5:30P

Location

32-123/Kirsch Auditorium

Refreshments

4:15P

Add to Calendar 2018-05-02 16:30:00 2018-05-02 17:30:00 America/New_York Dertouzos Distinguished Lecture: From Utopia to Dystopia in 29 Short Years This year marks a milestone in the history of the World Wide Web: more than half of the world's population will be online. However, threats to the web are real and many; from misinformation and questionable political advertising to a loss of control over our personal data.If a future web were to empower the hopes we had for the original web, what would it look like? A new web is possible in which people have complete control of their own data; applications and data are separated from each other; and a user's choice for each are separate. Let's assemble the brightest minds from business, technology, government, civil society, the arts and academia to build a new web which will again empower science and democracy. 32-123/Kirsch Auditorium Belfer sarah_donahue@hks.harvard.edu

May 03, 2018

Innovative Design and Analysis for Intelligent and Connected Vehicles

Chung-Wei Lin

Toyota InfoTechnology Center

11:00A

- 12:00P

Location

Seminar Room D507

Add to Calendar 2018-05-03 11:00:00 2018-05-03 12:00:00 America/New_York Innovative Design and Analysis for Intelligent and Connected Vehicles Abstract: Advanced Driver Assistance Systems (ADAS), autonomous functions, and connected applications bring a revolution to automotive systems and software. In this talk, several ongoing research topics in the domain of automotive systems and software will be introduced: (1) security-aware design and analysis considering timing and game theory, (2) a distributed graph-based conflict resolution algorithm for intersection management, (3) a certification protocol for plug-and-play systems to certify outsourced software but keep source codes confidential, and (4) a case study of runtime monitoring for two ADAS applications, cooperative pile-up mitigation system and false-start prevention system. Current achievements and research challenges will be covered in the talk.Biography: Chung-Wei Lin is a researcher at Toyota InfoTechnology Center. He received the B.S. degree in computer science and the M.S. degree in electronics engineering from National Taiwan University. He received the Ph.D. degree in electrical engineering and computer sciences from University of California, Berkeley. His research includes design and analysis of automotive systems, security, and design automation. He serves as an Automotive Committee member of Design Automation Conference (DAC) and TPC members of DAC, ASP-DAC, SIES, and other conferences. He won ACM Transactions on Design Automation of Electronic Systems (TODAES) 2016 Best Paper Award. Seminar Room D507 Belfer sarah_donahue@hks.harvard.edu

Interpretable Representation Learning for Visual Intelligence

Bolei Zhou

4:00P

- 5:00P

Location

Kiva 32-G449

Add to Calendar 2018-05-03 16:00:00 2018-05-03 17:00:00 America/New_York Interpretable Representation Learning for Visual Intelligence Recent progress of deep neural networks in computer vision and machine learning has enabled transformative applications across robotics, healthcare, and security. However, despite the superior performance of the deep neural networks, it remains challenging to understand their inner workings and explain their output predictions. My thesis work has pioneered several novel approaches for opening up the “black box” of neural networks used in vision tasks. In this talk, I will first show that objects and other meaningful concepts emerge as a consequence of recognizing scenes. A network dissection approach is further introduced to automatically identify the emergent concepts and quantify their interpretability. Then I will describe an approach that can efficiently explain the output prediction for any given image. It sheds light on the decision-making process of the networks and why they succeed or fail. Finally, I will talk about ongoing efforts toward learning efficient and interpretable deep representations for video event understanding and applications in robotics and medical image analysis.Committee: Antonio Torralba, Aude Oliva, Bill Freeman Kiva 32-G449 Belfer sarah_donahue@hks.harvard.edu

May 04, 2018

Learning representations of the visual world

Jon Shlens

Google Brain

Part Of

Brains, Minds and Machines Seminar Series 2017/2018

2:00P

- 3:00P

Location

Singleton Auditorium (46-3002)

Singleton Auditorium, 3rd Fl., MIT Bldg. 46, 43 Vassar Street, Cambridge MA 02139

Add to Calendar 2018-05-04 14:00:00 2018-05-04 15:00:00 America/New_York Learning representations of the visual world Abstract: Recent advances in machine learning have profoundly influenced our study of computer vision. Successes in this field have demonstrated the expressive power of learning representations directly from visual imagery — both in terms of practical utility and unexpected expressive abilities. In this talk I will discuss several contributions which have helped improve our ability to learn representations of images. First, I will describe recent advances for constructing models for extracting semantic information from images by leveraging transfer learning and meta-learning techniques. Such learned models outperform human-invented architectures and are readily scalable across a range of computational budgets. Second, I will highlight recent efforts focused on the converse problem of synthesizing images through the rich visual vocabulary of painting styles and visual textures. This work permits a unique exploration of visual space and offers a window on to the structure of the learned representation of visual imagery. My hope is that these works will highlight common threads in machine and human vision and point towards opportunities for future research.Speaker Bio.: Jon Shlens is a senior research scientist at Google since 2010. Prior to joining Google Research he was a research fellow at the Howard Hughes Medical Institute and a Miller Fellow at UC Berkeley. His research interests include machine perception, statistical signal processing, machine learning and biological neuroscience. Singleton Auditorium (46-3002) Belfer sarah_donahue@hks.harvard.edu

May 07, 2018

Responsibility Sensitive Safety of Self-Driving Cars

Shai Shalev-Shwartz

Hebrew University of Jerusalem, Mobileye

3:00P

- 4:00P

Location

32-G449 (Stata Center - Patil/Kiva Conference Room)

Add to Calendar 2018-05-07 15:00:00 2018-05-07 16:00:00 America/New_York Responsibility Sensitive Safety of Self-Driving Cars Abstract: In recent years, car makers and tech companies have been racing towards self driving cars. It seems that the main parameter in this race is who will have the first car on the road. The goal of the talk is to add to the equation two additional crucial parameters. The first is standardization of safety assurance --- what are the minimal requirements that every self-driving car must satisfy, and how can we verify these requirements. The second parameter is scalability --- engineering solutions that lead to unleashed costs will not scale to millions of cars, which will push interest in this field into a niche academic corner, and drive the entire field into a "winter of autonomous driving". In the first part of the talk I will show why statistical guarantees give a very weak safety and will propose instead a white-box, interpretable, mathematical model for safety assurance, which we call Responsibility-Sensitive Safety (RSS). Time permitting, the second part of the talk will involve reinforcement learning techniques for scalable driving policy.Joint work with Shaked Shammah and Amnon Shashua.Bio:Shai Shalev-Shwartz is a VP Technology at Mobileye, a Senior Fellow at Intel, and a professor in the Faculty of Computer Science and Engineering at the Hebrew University of Jerusalem. Prof. Shalev-Shwartz is the author of the book “Online Learning and Online Convex Optimization,” and a co-author of the book “Understanding Machine Learning: From Theory to Algorithms”. His research is around the theoretical foundations of machine learning as well as crafting theoretically justified optimization algorithms that make learning more efficient. He received several best paper awards, was listed in the Aminer top 100 most influential scholar list of 2016, and was listed at the 17'th place in The-Marker's top 100 most influential people in Israel, under the title “the brain behind artificial intelligence”, acknowledging his contributions to Mobileye's technology. 32-G449 (Stata Center - Patil/Kiva Conference Room) Belfer sarah_donahue@hks.harvard.edu

Machine Learning Seminar: Benjamin Recht "A Dynamical View of Optimization and Learning"

Benjamin Recht

4:00P

- 5:00P

Location

34-401 Grier Combined

Refreshments

3:45P

Add to Calendar 2018-05-07 16:00:00 2018-05-07 17:00:00 America/New_York Machine Learning Seminar: Benjamin Recht "A Dynamical View of Optimization and Learning" Abstract: Noting that all iterative algorithms are dynamical systems, I will illustrate how most of the popular optimization methods used in machine learning can be cast as a family of feedback systems long studied in control theory. Leaning on this abstraction enables us to apply powerful, control-theoretic methods to algorithm analysis. I will show how the convergence rates of most common algorithms—including gradient descent, mirror descent, Nesterov’s method, etc.—can be verified using a unified set of potential functions. I will then describe how such potential functions can themselves be found by solving small semidefinite programming problems. These techniques can be used to search for optimization algorithms with desired performance characteristics and provide a new methodology for algorithm design. I will close with several additional examples of how a dynamical view can provide new insights into how to make machine learning systems more safe and reliable as they interact with increasingly complex and uncertain environments.Bio: Benjamin Recht is an Associate Professor in the Department of Electrical Engineering and Computer Sciences at the University of California, Berkeley. Ben's research group studies the theory and practice of optimization algorithms with a focus on applications in machine learning and data analysis. Ben is the recipient of a Presidential Early Career Awards for Scientists and Engineers, an Alfred P. Sloan Research Fellowship, the 2012 SIAM/MOS Lagrange Prize in Continuous Optimization, the 2014 Jamon Prize, the 2015 William O. Baker Award for Initiatives in Research, and the 2017 NIPS Test of Time Award. 34-401 Grier Combined Belfer sarah_donahue@hks.harvard.edu

Three-dimensional Cardiovascular Magnetic Resonance for Children with Congenital Heart Disease

Mehdi H. Moghari

Children's Hospital Boston; Harvard Medical School

Part Of

Biomedical Imaging and Analysis 2017/2018

5:00P

- 6:00P

Location

32-D407

Add to Calendar 2018-05-07 17:00:00 2018-05-07 18:00:00 America/New_York Three-dimensional Cardiovascular Magnetic Resonance for Children with Congenital Heart Disease Severe congenital heart disease (CHD) affects approximately 1.2% of children and is the leading cause of birth defect-related deaths. Patients with severe CHD require multiple palliative surgeries during the first day of their lives to survive. Despite considerable improvement in the survival of these patients, there is increasing morbidity and mortality over time. It remains unclear why some of these patients fail their surgical repairs while others remain relatively well. Clinicians often rely on 2-dimensional (2D) images acquired from echocardiograms, catheterizations, or cardiovascular magnetic resonance (CMR) exams to assess these patients and qualitatively choose the optimal surgical repair. The 2D images, however, lead to a suboptimal understanding of the complex 3D spatial relationships and hemodynamics, and limit efficient decision making. To address this deficiency, we have developed a comprehensive 3D CMR sequence to acquire 3D block of images from patients with CHD. These 3D dataset is used for simultaneous assessment of cardiac function and hemodynamics. In this talk, we will introduce the conventional 2D CMR exam and discuss its limitations. The new 3D CMR exam will be then presented and compared with the 2D CMR exam in terms of cardiac function and hemodynamics. 32-D407 Belfer sarah_donahue@hks.harvard.edu

May 09, 2018

Preventing (Network) Time Travel with Chronos

Michael Schapira

Hebrew University of Jerusalem, School of Computer Science and Engineering

2:00P

- 3:00P

Location

32-G882 (Hewlett Room)

Add to Calendar 2018-05-09 14:00:00 2018-05-09 15:00:00 America/New_York Preventing (Network) Time Travel with Chronos ABSTRACTThe Network Time Protocol (NTP) synchronizes time across computer systems over the Internet. Unfortunately, NTP is highly vulnerable to “time shifting attacks”, which has severe implications for time-sensitive applications and for security mechanisms. We present Chronos, a new NTP client whose design leverages ideas from distributed computing theory to achieve good synchronization even in the presence of powerful man-in-the-middle attackers. Chronos is backwards compatible with legacy NTP and involves no changes whatsoever to NTP servers. Our results indicate that to shift time at a Chronos client by over 100ms from the universal time (UTC), even powerful man-in-the-middle attackers require over 20 years of effort in expectation.Joint work with Omer Deutsch, Neta Rozen Schiff, and Danny DolevSPEAKER BIOMichael Schapira is an associate professor and the co-leader of the Fraunhofer Cybersecurity Center at the Hebrew University. His research focuses on the design and analysis of (Inter)network architectures and protocols. Prior to joining Hebrew U, he worked at Google NYC's Infrastructure Networking Group and was a postdoctoral researcher at UC Berkeley, Yale University, and Princeton University. He is a recipient of the Microsoft Research Faculty Fellowship, 2 IETF/IRTF Applied Networking Research Prizes, a Google Faculty Research Award, and the ERC Starting Grant. 32-G882 (Hewlett Room) Belfer sarah_donahue@hks.harvard.edu

The Discrete-Logarithm Problem with Preprocessing

Henry Corrigan-Gibbs

Dept. of Computer Science, Stanford University

4:00P

- 5:00P

Location

32-G882

Hewlett Room

Add to Calendar 2018-05-09 16:00:00 2018-05-09 17:00:00 America/New_York The Discrete-Logarithm Problem with Preprocessing ABSTRACTIn this talk, I will present some recent work on discrete-log algorithmsthat use preprocessing. In our model, an adversary may use a very largeamount of precomputation to produce an "advice" string about a specificgroup (e.g., NIST P-256). In a subsequent online phase, the adversary'stask is to use the preprocessed advice to quickly compute discretelogarithms in the group. Motivated by surprising recent preprocessingattacks on the discrete-log problem, we study the power and limits ofsuch algorithms.In particular, we focus on generic algorithms -- these are algorithmsthat operate in every cyclic group. We show that any genericdiscrete-log algorithm with preprocessing that uses an S-bit advicestring, runs in online time T, and succeeds with probability \epsilon ina group of prime order N must satisfy ST^2 = \tilde{\Omega}(\epsilon N).Our lower bound, which is tight up to logarithmic factors, uses asynthesis of incompressibility techniques and classic methods forgeneric-group lower bounds. We apply our techniques to prove relatedlower bounds for the CDH, DDH, and multiple-discrete-log problems.Finally, we demonstrate two new generic preprocessing attacks: one forthe multiple-discrete-log problem and one for certain decisional-typeproblems in groups. This latter result demonstrates that, for genericalgorithms with preprocessing, distinguishing tuples of the form (g,g^x, g^(x^2)) from random is much easier than the discrete-log problem.This talk is based on joint work with Dmitry Kogan.BIOHenry Corrigan-Gibbs is a PhD candidate at Stanford, advised by Dan Boneh. He builds systems for messaging, data analysis, and web browsing that protect the private data and metadata of their users. For these research efforts, Henry and his co-authors have received the Best Young Researcher Paper Award at Eurocrypt 2018, the 2016 Caspar Bowden Award for Outstanding Research in Privacy Enhancing Technologies, and the 2015 IEEE Security and Privacy Distinguished Paper Award. 32-G882 Belfer sarah_donahue@hks.harvard.edu

Learning Mixtures of Product Distributions via Higher Multilinear Moments

Sitan Chen

MIT

Part Of

Algorithms and Complexity Seminar Series 2017/2018

4:00P

- 5:00P

Location

32-G575

Add to Calendar 2018-05-09 16:00:00 2018-05-09 17:00:00 America/New_York Learning Mixtures of Product Distributions via Higher Multilinear Moments Abstract: Learning mixtures of k binary product distributions is a central problem in computational learning theory, but one where there are wide gaps between the best known algorithms and lower bounds (even for restricted families of algorithms). We narrow many of these gaps by developing novel insights about how to reason about higher order multilinear moments. Our results include: 1) an n^{O(k^2)} time algorithm for learning mixtures of binary product distributions, giving the first improvement on the n^{O(k^3)} time algorithm of Feldman, O'Donnell and Servedio (FOCS '05), 2) an n^{\Omega(\sqrt{k})} statistical query lower bound, improving on the quasipolynomial lower bound that is based on connections to sparse parity with noise, and 3) a quasipolynomial time algorithm for learning mixtures of $k$ subcubes. This special case can still simulate many other hard learning problems, but is much richer than any of them alone. As a corollary, we obtain more flexible algorithms for learning decision trees under the uniform distribution, that work with stochastic transitions, when we are only given positive examples and with a polylogarithmic number of samples for any fixed k.Joint work with Ankur Moitra. 32-G575 Belfer sarah_donahue@hks.harvard.edu

May 11, 2018

Kevin Fu: Physics of Cybersecurity: Sensors, Acoustics, Cuba

Kevin Fu, University of Michigan

Part Of

Cryptography and Information Seminar (CIS) 2018

10:30A

- 12:00P

Location

Hewlett, G882

Add to Calendar 2018-05-11 10:30:00 2018-05-11 12:00:00 America/New_York Kevin Fu: Physics of Cybersecurity: Sensors, Acoustics, Cuba Abstract: Medical devices, autonomous vehicles, and the Internet of Things depend on the integrity and availability of trustworthy data from sensors to make safety-critical, automated decisions. How can such cyberphysical systems remain secure against an adversary using intentional interference to fool sensors? Physics-based cybersecurity risks can bubble up into operating systems as bizarre, undefined behavior. Transduction attacks using audible acoustic, ultrasonic, and radio interference can manipulate sensors found in devices ranging from fitbits to hard drives to implantable medical devices with implications to file system integrity and human safety. Defenders can fight back with physics and more trustworthy software APIs. I’ll wrap up by explaining how ultrasonic exfiltration could have caused the symptoms experienced by diplomats harmed in Cuba.Biography: Kevin Fu enjoys early-stage, interdisciplinary research where there are still many par-baked problems to define and solve. Kevin was recognized as an IEEE Fellow, Sloan Research Fellow, and MIT Technology Review TR35 Innovator of the Year. He serves as chair of the CCC Cybersecurity Task Force. His students received some best paper awards. Kevin earned a certificate of artisanal bread making from the French Culinary Institute. http://web.eecs.umich.edu/~kevinfu/ Hewlett, G882 Belfer sarah_donahue@hks.harvard.edu

Sophia Yakoubov: "Fuzzy Password-Authenticated Key Exchange"

Sophia Yakoubov, Boston University

Part Of

Cryptography and Information Seminar (CIS) 2018

1:00P

- 2:00P

Location

Hewlett, G882

Refreshments

12:00P

Add to Calendar 2018-05-11 13:00:00 2018-05-11 14:00:00 America/New_York Sophia Yakoubov: "Fuzzy Password-Authenticated Key Exchange" Abstract:Consider key agreement by two parties who start out knowing a common secret (which we refer to as “pass-string”, a generalization of “password”), but face two complications: (1) the pass-string may come from a low-entropy distribution, and (2) the two parties’ copies of the pass-string may have some noise, and thus not match exactly. We provide the first efficient and general solutions to this problem that enable, for example, key agreement based on commonly used biometrics such as iris scans.The problem of key agreement with each of these complications individually has been well studied in literature. Key agreement from low-entropy shared pass-strings is achieved by password-authenticated key exchange (PAKE), and key agreement from noisy but high-entropy shared pass-strings is achieved by information-reconciliation protocols as long as the two secrets are “close enough.” However, the problem of key agreement from noisy low-entropy pass-strings has never been studied.We introduce (universally composable) fuzzy password-authenticated key exchange (fPAKE), which solves exactly this problem. fPAKE does not have any entropy requirements for the pass-strings, and enables secure key agreement as long as the two pass-strings are “close” for some notion of closeness. We also give two constructions. The first construction achieves our fPAKE definition for any (efficiently computable) notion of closeness, including those that could not be handled before even in the high-entropy setting. It uses Yao’s Garbled Circuits in a way that is only two times more costly than their use against semi-honest adversaries, but that guarantees security against malicious adversaries. The second construction is more efficient, but achieves our fPAKE definition only for pass-strings with low Hamming distance. It builds on very simple primitives: robust secret sharing and PAKE.Joint work with Pierre-Alain Dupont, Julia Hesse, David Pointcheval, and Leonid Reyzin. To appear at Eurocrypt 2018. Hewlett, G882 Belfer sarah_donahue@hks.harvard.edu

May 14, 2018

Multidimensional diffusion imaging: Why we should leave conventional diffusion MRI behind

Filip Szczepankiewicz

Harvard Medical School / Brigham and Women's Hospital

Part Of

Biomedical Imaging and Analysis 2017/2018

5:00P

- 6:00P

Location

32-D407

Add to Calendar 2018-05-14 17:00:00 2018-05-14 18:00:00 America/New_York Multidimensional diffusion imaging: Why we should leave conventional diffusion MRI behind The lecture will survey what features of tissue that can be probedwith diffusion MRI once we abandon conventional encoding techniques infavor of multidimensional diffusion encoding. The theory andmethodology of tensor-valued diffusion encoding will be studied indetail, and results will be demonstrated from healthy brain and braintumors. Time permitting, the methodology will be put in a largercontext together with related advanced imaging methods. 32-D407 Belfer sarah_donahue@hks.harvard.edu