Hardly a week goes by without a report about another cyberattack. With almost every major organization having been victim, including most government organizations, such Target, Sony, NSA, US Office of Personnel Management, why would you expect your organization to be immune? By many projections, the worse is yet to come. Although much progress is being made in improving hardware and software, studies have reported that between 50-70% of all cyberattacks are aided or abetted by insiders (usually unintentionally), so understanding the cybersecurity governance and organizational culture is increasingly important. In this session, we will discuss the managerial, organizational, and strategic aspects of cybersecurity with an emphasis on the protection of the nation's critical infrastructure.
2016 MIT Information and Communication Technologies Conference
This talk will discuss the critical role of mapping and localization in the development of self-driving cars and autonomous underwater vehicles (AUVs). After a discussion of some of the recent amazing progress and open technical challenges in the development of self-driving vehicles, we will discuss the past, present and future of Simultaneous Localization and Mapping (SLAM) in robotics. We will review the history of SLAM research and will discuss some of the major challenges in SLAM, including choosing a map representation, developing algorithms for efficient state estimation, and solving for data association and loop closure. We will describe some of the challenges using SLAM for AUVs, and we will also present recent results on object-based mapping in dynamic environments and real-time dense mapping using RGB-D cameras.
Joint work with Sudeep Pillai, Tom Whelan, Michael Kaess, John McDonald, Hordur Johannsson, Maurice Fallon, David Rosen, Ross Finman, Paul Huang, Liam Paull, Nick Wang, and Dehann Fourie.
Encryption as a means of data control (privacy and security):
For a long time, interaction on Web has been less private or secure than many end-users expect and prefer. Now, however, the widespread deployment of encryption helps us to change that.
* Making encryption widespread. For years we have known how to do encryption, but it wasn't widely used, because it wasn't part of overall system design. In response, particularly as we've become aware of capabilities for network-scale monitoring, standards groups including IETF and W3C have worked to encrypt more of those network connections at the protocol and API-design phase, and to make it easier to deploy and use encrypted protocols such as HTTPS. Encryption won't necessarily stop a targeted attack (attackers can often break end-user systems where they can't brute-force break the encryption), but it raises the effort required for surveillance and forces transparency on other network participants who want to see or shape traffic.
* Secure authentication. Too many of our "secure" communications are protected by weak password mechanisms, leaving users open to password database breaches and phishing attacks. Strong new authentication mechanisms, being worked on for web-wide standards, can replace the password; helping users and applications to secure accounts more effectively. Strong secure authentication will enable users to manage their personal interactions and data privacy, as well as securing commercial data exchange.
Brian Anthony
Benedetto Marelli Paul M. Cook Career Development Professor, Associate Professor of Civil, and Environmental Engineering, MIT CEE
Song Han Assistant Professor, Department of Electrical Engineering and Computer Science, MIT
Contact
This talk introduces a new generation of machine learning methods that provide state of the art performance and are very interpretable, introducing optimal classification (OCT) and regression (ORT) trees for prediction and prescription with and without hyperplanes. This talk shows that (a) Trees are very interpretable, (b) They can be calculated in large scale in practical times, and (c) In a large collection of real world data sets, they give comparable or better performance than random forests or boosted trees. Their prescriptive counterparts have a significant edge on interpretability and comparable or better performance than causal forests. Finally, we show that optimal trees with hyperplanes have at least as much modeling power as (feedforward, convolutional, and recurrent) neural networks and comparable performance in a variety of real world data sets. These results suggest that optimal trees are interpretable, practical to compute in large scale, and provide state of the art performance compared to black box methods.