Homomorphic Encryption for Privacy-Preserving Machine Learning
Speaker: Michael Reid
As machine learning has evolved, the systems that train and predict on machine learning models have increasingly become outsourced from those that own the data; for example, a company training its data in a public cloud like AWS, or Facebook giving away its user data to its partners. As an end user of these systems, we want to experience the personalized benefits that training sophisticated machine learning models enables, but also want to make sure our data is secure and not visible to unknown third parties. In this talk, we will look into how homomorphic encryption techniques can be used to allow neural networks to train atop data sets without being able to see the data underneath. We will discuss both fully and somewhat homomorphic cryptosystems, as well as the ongoing research being done to apply these to neural networks.
Michael Reid is the co-founder of Nautilus Analytics Technologies, a company that is building privacy-preserving machine learning software for businesses in government, healthcare, and other security-focused sectors. Previously, Mike worked as a software engineer at Hootsuite, Splunk, and Amazon Web Services, dealing with both machine learning and large-scale, high-volume distributed systems.
Mik Kersten is from Tasktop Technologies Inc, a local Vancouver company. This talk will interest anybody who uses the Eclipse IDE for software development.
Abstract: In under four years, Mylyn’s task-focused interface moved from a UBC whiteboard to the hands of hundreds of thousands of developers. For many, the idea of a tool that hides more of a program than it shows sounded wrong. But the rapid adoption of Mylyn is making the next big step in the evolution of the Integrated Development Environment (IDE) clear. Tasks are more important than files, focus is more important than features, and explicit context has provided developers the biggest productivity boost since structure views and code completion. This talk will discuss the inspirations and research behind Mylyn, demonstrate the tool and the impact it has had, overview the open source ecosystem and startup that we formed around the technology, and conclude with a vision of how the task-focused interface is poised to transform other fields of knowledge work.
Refreshments will be served at 6pm and the lecture will start at 6:30pm. Come early to chat with old friends, or make new friends.
For more information about each of these lectures, please check the web site.
Title: Physical Interface Design and Communication Through Touch
Abstract: I’m interested in how people communicate through the sense of touch, and how haptic information transfer interacts with perception in other modalities. As time and interest permit, I’ll discuss two very different kinds of haptic communication that my group has been studying recently. The first is abstract information, delivered one-way to your hand encoded in complex vibrations. We’ve found that humans are better at this than you might expect—depending on how the sensations are created; and the medium has potential for low-effort, background communication. In the second kind, we’re examining haptically communicated affect: what’s behind feels that we like or don’t like—can this be predicted or quantified? How do we communicate emotion haptically, to people or animals, and is this an essential part of emotional communication more generally? We’re building a highly sensed animatronic Creature as an experimental platform, which we plan to use for basic study and in a therapeutic setting.
We’ll meet this time in the HCI Learning Studio, a resource created with a special grant by the CS HCI faculty for studio-style courses. Snacks possibly available for early arrivals.