3 ms·

> it uses the phone’s touchscreen and repurposes the existing battery temperature sensors to gather data that a machine learning model uses to estimate people’s

> it uses the phone’s touchscreen and repurposes the existing battery temperature sensors to gather data that a machine learning model uses to estimate people’s core body temperatures > FeverPhone estimated patient core body temperatures with an average error of about 0.41 degrees Fahrenheit (0.23 degrees Celsius), which is in the clinically acceptable range of 0.5 C

nomel 3y ago
> with an average error Eww, that's not how it works. The goal isn't to make the average near the limit, since the average of -100 and 100 means 0 average error! The goal is to make sure the measurement have a distribution that gives some reasonable confidence that it's within the limit. This is PTT: https://www.spcpress.com/pdf/DJW244.pdf https://www.spcpress.com/pdf/DJW244.pdf
- glibby 3y ago
  I would have naively calculated average error as avg(max( abs(smallest), largest)). so it would be 100 in your case. I'm sure that's not statistically correct but it makes sense to me.
- glial 3y ago
  Abstract says "mean absolute error of 0.229 °C"
- srcreigh 3y ago
  It’d be hard to get average error so low, since you need precision to get low error values for varying temps. You can’t just return 100 and -100 randomly, since the actual temp is unknown and you’d need to evenly sample around that.
  - nomel 3y ago
    A normal distribution will average to 0, with enough samples. The center and spread is important, since individual measurements can be considered a random sampling within that distribution. But, this is overly simplifying things, and assume random error, which AI systems often don't exhibit. It appears this is moot, because the quote did not reflect the actual value or method used.
    - srcreigh 3y ago
      The distribution of human temperature levels is already known, you could just give the average temp as every guess and get maybe 1.5 F average error. My point is, you can’t really get 0.44F avg error unless it’s predictive. The “std dev” of abs(reality - model) is too high
Rebelgecko 3y ago
Skimming through the study, it seems like most of the control temps were within a 0.6° C range. I wonder what the average absolute error would be of a thermometer that always returned 37°C
- h0l0cube 3y ago
  According to a sibling comment the abstract claims "a mean absolute error of 0.229 °C". According to wiki the normal temperature window is 36.5–37.5 °C and that of a fever is 37.2-37.7 °C or higher, with variations for both of about 0.5 °C throughout a 24 hour period. > An early morning temperature higher than 37.2 °C (99.0 °F) or a late afternoon temperature higher than 37.7 °C (99.9 °F) is normally considered a fever, assuming that the temperature is elevated due to a change in the hypothalamus's setpoint.[14] Lower thresholds are sometimes appropriate for elderly people.[14] The normal daily temperature variation is typically 0.5 °C (0.90 °F), but can be greater among people recovering from a fever. https://en.wikipedia.org/wiki/Human_body_temperature https://en.wikipedia.org/wiki/Human_body_temperature
progrus 3y ago
Amazing!

nomel 3y ago

> with an average error Eww, that's not how it works. The goal isn't to make the average near the limit, since the average of -100 and 100 means 0 average error! The goal is to make sure the measurement have a distribution that gives some reasonable confidence that it's within the limit. This is PTT: https://www.spcpress.com/pdf/DJW244.pdf https://www.spcpress.com/pdf/DJW244.pdf

glibby 3y ago
I would have naively calculated average error as avg(max( abs(smallest), largest)). so it would be 100 in your case. I'm sure that's not statistically correct but it makes sense to me.
glial 3y ago
Abstract says "mean absolute error of 0.229 °C"
srcreigh 3y ago
It’d be hard to get average error so low, since you need precision to get low error values for varying temps. You can’t just return 100 and -100 randomly, since the actual temp is unknown and you’d need to evenly sample around that.
- nomel 3y ago
  A normal distribution will average to 0, with enough samples. The center and spread is important, since individual measurements can be considered a random sampling within that distribution. But, this is overly simplifying things, and assume random error, which AI systems often don't exhibit. It appears this is moot, because the quote did not reflect the actual value or method used.
  - srcreigh 3y ago
    The distribution of human temperature levels is already known, you could just give the average temp as every guess and get maybe 1.5 F average error. My point is, you can’t really get 0.44F avg error unless it’s predictive. The “std dev” of abs(reality - model) is too high

Rebelgecko 3y ago

Skimming through the study, it seems like most of the control temps were within a 0.6° C range. I wonder what the average absolute error would be of a thermometer that always returned 37°C

h0l0cube 3y ago
According to a sibling comment the abstract claims "a mean absolute error of 0.229 °C". According to wiki the normal temperature window is 36.5–37.5 °C and that of a fever is 37.2-37.7 °C or higher, with variations for both of about 0.5 °C throughout a 24 hour period. > An early morning temperature higher than 37.2 °C (99.0 °F) or a late afternoon temperature higher than 37.7 °C (99.9 °F) is normally considered a fever, assuming that the temperature is elevated due to a change in the hypothalamus's setpoint.[14] Lower thresholds are sometimes appropriate for elderly people.[14] The normal daily temperature variation is typically 0.5 °C (0.90 °F), but can be greater among people recovering from a fever. https://en.wikipedia.org/wiki/Human_body_temperature https://en.wikipedia.org/wiki/Human_body_temperature

progrus 3y ago

Amazing!