Four short links: 18 September 2019

1. Extracting Insights from the Shape of Complex Data using Topology (Nature) — three properties of topology that make it useful: coordinate-free, so you can work on data with different coordinate systems; “small” deformation invariance, so it’s resistant to some forms of noise; and it works on the compressed representations of shapes so you can use it to throw away features but still keep fundamental properties of the shape of the data. I’m paraphrasing the intro here.
2. Getting Tired of Your Friends: The Dynamics of Venture Capital Relationships — Based on the relationships of the top 50 US venture capital firms, this paper focuses on the strengths of relationships and their dynamic evolution. Empirical estimates indicate that having a deeper relationship leads to fewer, not more future co-investments. Moreover, deeper relationships lead to lower exit performance, even after controlling for endogeneity. Interestingly, deeper relationships first lead to lower performance, and subsequently lead to a slowdown in the relationship intensity. The more I know about you, the less I want to do business with you because I know we’ll lose money together.
3. ImageNet Roulette (Kate Crawford and Trevor Paglen) — show it a picture, it’ll label using ImageNet. A wonderful reality check about the limitations of general-purpose computer perception.
4. Social Science One Ships — Facebook’s project to release data to registered researchers took longer than expected because they implemented differential privacy (perturbing the data to preserve statistical properties while preventing reconstruction of any one person’s attributes). The data describe web page addresses (URLs) that have been shared on Facebook starting January 1, 2017, up to and including February 19, 2019. URLs are included if shared with public privacy settings more than on average 100 times (±Laplace(5) noise to minimize information leakage). We have conducted post-processing on the URLs … to remove potentially private and/or sensitive data This paper goes into detail on steps taken to protect privacy.

from Radar https://ift.tt/2M2LGoz