Adversarial Examples for Evaluating Reading Comprehension Systems by Robin Jia and Percy Liang
- Evaluation under ideal conditions is only a start. To test and improve systems–in this study language comprehension algorithms–we need adversarial evaluation.
- Generally speaking, image classification algorithms error because they are overly sensitive, while language comprehension algorithms error because they’re overly stable.
Meaningless comparisons lead to false optimism in medical machine learning by Orianna DeMasi, Konrad Kording, and Benjamin Recht
- Check the baselines. Check the baselines. Check the baselines.
- This is a vital lesson for philanthropy. Tired “Are you using machine learning?” Wired “Demonstrate that your machine learning is actually good”
Seven Days of Heroin: This is What an Epidemic Looks Like by Cincinnati Enquirer and Media Network of Central Ohio
“The woman says she’s lost two children to heroin in just the past year. She describes waking up in an abandoned building one morning to find her 25-year-old daughter at her side, cold and dead.”
Blood And Iron: Photographing The ‘End Of The Industrial Revolution’ by Vikor Macha
“I’m not even very interested in photography. It’s just the best medium for recording the production cycle.”
## make iris1
library(tidyverse)
library(ggdendro)
library(networkD3)
iris_clusters <- iris %>%
select(Sepal.Length:Petal.Width) %>%
dist() %>%
hclust(method = "average") %>%
as.dendrogram() %>%
dendro_data()
iris_clusters$segments %>%
log10() %>%
ggplot() +
geom_segment(aes(x=x*-1, y=y*-1, xend=xend-10, yend=yend), alpha = .1, color = "white") +
geom_point(aes(x = x*-1, y = y*-1), color = "grey", size = .01) +
coord_polar() +
theme_void() +
theme(panel.background = element_rect(fill = "black", color = "black"))