Filter Options

How To Do Healthy Machine Learning in Health
While clinical AI and medical risk scores have received much attention for their potential to achieve above-human performance, there are many concerns about their ability to mimic societal bias. In this talk, Dr. Ghassemi explores the difficulty of making state-of-the-art machine learning models behave as we say, not as we do, and how technical choices that seems natural in other settings may not work well in health.

Deep Equilibrium Models (DEQs) are a class of machine learning models that computes the fixed point of a single nonlinear operator in lieu of a traditional multi-layer network. The resulting models are conceptually simpler than traditional models and benefit from smaller memory requirements.

Most discussions about cyber-security today focus on harms to the enterprise (e.g., ransomware or data breach) or harms to the individual (e.g., identity theft or fraud). But what does the Internet itself, as a packet transport system, have to do with this. In this talk, David will discuss vulnerabilities and abuses of the key systems that make up the Internet: addressing, routing, naming and key management, and how these systems are exploited as part of a wide range of malicious activities. 

CSAIL researcher Henry Corrigan-Gibbs will present the design and implementation of SafetyPin, a system for encrypted mobile-device backups. Like existing cloud-based mobile-backup systems, including those of Apple and Google, SafetyPin requires users to remember only a short PIN and defends against brute-force PIN-guessing attacks using hardware security protections. Unlike today's systems, SafetyPin splits trust over a cluster of hardware security modules (HSMs) in order to provide security guarantees that scale with the number of HSMs.

Efficient Privacy Enhancing Technologies (PET) in the coming decade will enable us to perform a host of tasks while adhering to privacy regulations and without losing competitive edge while going outside one's own firm. In this talk Shafi will discuss current efficient verification methods which can maintain privacy of digital documents and contracts while verifying compliance and validity, as well as efficient Homomorphic Encryption based methods which can allow the extraction of knowledge from a collection of data owned by different entities without getting direct access to the data itsel