McConville
Half a century ago, photographs of Earth from space were first broadcast around the world. Initially celebrated for their sublime beauty, they have since transformed into ambivalent icons representing both planetary consciousness and existential angst. Today, remote sensing technologies increasingly serve as humanity's sensory prosthetics, expanding potential interpretations of the meaning of the "whole Earth." They illuminate phenomena across previously invisible spatial, temporal, and spectral scales, revealing the complexity and interconnectedness of Earth's systems. In this presentation, I explore how visualizations of satellite observations can transform perceptions of our home planet and their paradoxical effects on collective sensemaking.

Siegel
Spiritual and religious traditions have historically cornered the market on paths to peace, connection, and wellbeing. Science and technology are truly ubiquitous and incredibly beneficial, but they do little to point us toward deep flourishing. With all of our digital connectivity, by many measures we are not getting any happier. What if we applied our incredible engineering capacity toward our inner world, not just to cure disease, but to self-actualization? What if we took an innovation approach to spirituality itself, and redesigned it for a modern secular world? What if tech that provided the benefit of 10,000 hours of meditation was as accessible as cell phones? The quality of our relationships is one of the biggest contributors to our overall wellbeing. Thanks to modern technology we are becoming increasingly connected, but we are also increasingly lonely and socially isolated. Something is missing. In this talk we will explore a three projects attempting to leverage modern technology to support deep and meaningful human connection. We will also discuss current work leading up to a one of the first technology augmented meditation retreats.

Huang
Machine learning models have been shown to be vulnerable to adversarial examples, inputs maliciously constructed to force misclassification. This has been extensively studied in the context of computer vision applications of deep learning. In this talk, I will present recent work that shows neural network policies, trained with deep reinforcement learning, are also vulnerable to adversarial-example attacks. Specifically, existing adversarial example crafting techniques can be used to significantly degrade test-time performance of trained policies. This is true across tasks and training algorithms, in both white-box and black-box settings. Even small adversarial perturbations that do not interfere with human perception can cause a significant drop in performance. I will also describe results on constructing dormant adversarial examples, which have a delayed effect on the targeted policy's actions.