2
$\begingroup$

I'm currently working on a research project where I try to apply different kinds of Machine Learning on some existing software I wrote a few years ago.

This software will scan for people in the room continuously. Some of these detections are either True or False. However, this is not known, so I cannot use supervised learning to train a network to make a distinction. I do however have a number that is correlated to the number of detections that should be True in a given period of time (let's say 30 seconds - 2 minutes), which can be used as an output feature to train a regression model. But the problem is... How can I give these multiple "detections" as an input? The way I see it now, would be something like this:

+--------------------------------------------------------------+-----------+------------+------------+----------------+--+
|                          Detections                          | Variable1 | Variable 2 | Variable n | Output Feature |  |
+--------------------------------------------------------------+-----------+------------+------------+----------------+--+
| {person a, person b, person h, person z}                     |       132 |        189 |          5 |             50 |  |
| {person a, person b, person c, person d, person k, person m} |         1 |         50 |        147 |             80 |  |
| {person c, person e, person g, person f}                     |       875 |        325 |          3 |             20 |  |
+--------------------------------------------------------------+-----------+------------+------------+----------------+--+

Each of these persons would be a tuple of values: var_1, var_2, var_3, var_4. These values are not constant however! They do change between observations.

Different approach to explain it: there's multiple observations (variable amount) in each time segment (duration of time segment is a fixed integer to be chosen). These observations have a few variables that would indicate whether the observation is true or false. However, the threshold for it being true or false, is very much dependant on the other variables, that are not tied to the information of the persons. (These variables are the same for all of them, but vary in between time segments. Let's call'm "environment features") Lastly, the output feature is the product of the count of persons that resulted in "True" and a (varying) factor that is correlated to the environment features.

So I've been thinking about probabilistic AI, but the problem is that there isn't a known distribution between True/False.

  • Is there any technique I can apply to be able to use this kind of data as an input of a Neural Network (or other forms of ML)? Or is there a specific form of ML that is used for this kind of problems?

Thanks in advance!

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.