Thank you. Tapering at _just_ the right dose to not interfere with the grades in the master degree I am doing is quite tough.
Which makes me wonder...there's common regimens like the Ashton scale and whatnot for benzos but it's very boilerplate.
Unicorn Idea:
Quantitative Withdrawal
User enters dosage each time they use. Datetime is auto filled, but can be altered for dosages that are entered belatedly.
ML is used to show charts with sliders based on speed of taper and severity of side effects. A time series showing the reduction in withdrawal effects over time with an ETA and other statistics. With labeled sections for certain parts of the withdrawal that are more severe (think a phase change diagram.) Seizure/epilespy zone would be clearly large on a configuration where the user chooses a ridiculously fast taper. The app would show a color, red in this case, warning of these symptoms and recommending against it. Baseline taper recommendations could be based on the medical literature out there with clinical trials. There is plenty of labeled data especially from the NIH.
The user can log their current symptoms to help the model learn their individual brain chemistry.
And vitals like HR, pulse, and o2 that are easily measured via pleasant APIs like Healthkit on iOS and Android. (Would be by proxy optionally compatible with iWatch, FitBit and other such sensors.)
These vitals are great features that the model can learn from.
The user can answer questions regarding the current state of their withdrawal symptoms, providing the model with labeled data to learn from.
Models can be pretrained on an individual in close proximity to the MLE on the distribution of human neurochemistry. And thus would work out of the box pretty well before the users input and vitals start to vastly improve it until it helps the user maintain AND gain :-)
Which makes me wonder...there's common regimens like the Ashton scale and whatnot for benzos but it's very boilerplate.
Unicorn Idea: Quantitative Withdrawal
User enters dosage each time they use. Datetime is auto filled, but can be altered for dosages that are entered belatedly.
ML is used to show charts with sliders based on speed of taper and severity of side effects. A time series showing the reduction in withdrawal effects over time with an ETA and other statistics. With labeled sections for certain parts of the withdrawal that are more severe (think a phase change diagram.) Seizure/epilespy zone would be clearly large on a configuration where the user chooses a ridiculously fast taper. The app would show a color, red in this case, warning of these symptoms and recommending against it. Baseline taper recommendations could be based on the medical literature out there with clinical trials. There is plenty of labeled data especially from the NIH.
The user can log their current symptoms to help the model learn their individual brain chemistry.
And vitals like HR, pulse, and o2 that are easily measured via pleasant APIs like Healthkit on iOS and Android. (Would be by proxy optionally compatible with iWatch, FitBit and other such sensors.)
These vitals are great features that the model can learn from.
The user can answer questions regarding the current state of their withdrawal symptoms, providing the model with labeled data to learn from.
Models can be pretrained on an individual in close proximity to the MLE on the distribution of human neurochemistry. And thus would work out of the box pretty well before the users input and vitals start to vastly improve it until it helps the user maintain AND gain :-)
just ideas..