Four short links: 31 October 2016
Faking Neural Nets, Watson Fintech, Changing Behaviors, and Time-Series Features
- Universal Adversarial Perturbations — generating images that neural nets will mistakenly classify, and the system generalizes well across universal neural nets.
- RegTech — IBM applying Watson to regulations.
- Knowledge-based Interventions are More Likely to Reduce Legal Disparities than are Implicit Bias Interventions — argues that trying to educate about implicit bias doesn’t change outcomes; instead, we should change people’s knowledge about how the structure of the social environment makes them complicit in the perpetuation of bias. Not sure how this will fare in today’s individualistic world, where we aren’t allowed to acknowledge that we operate in a cultural system of largely unspoken and completely arbitrary expectations, values, and other mutually constructed fictions. “Don’t teach THAT controversy!” (via Patrick Forscher)
- TSFRESH — Automatic extraction of relevant features from time series.
