Problem: You toss a coin on a 2D table with known dimension. There are certain regions on the table where the probability of get heads is high. At the maximum you can toss N=20 times at an arbitrary 2D location on the table. For each location, you get heads(1) or tails(0) as an outcome.

Location of Heads form input A, location of Tails form input B. That means the input size to the network can be $N_{h} \times 2$(heads) and $N_{t} \times 2$(tails (could be split as A: 15x2, B:5x2 etc), and it is sequentially updated after every toss.

How do we predict the next location for tossing the coin which maximizes the probability of getting heads?

inputs: 2D location of heads, 2D location of tails

output: Next 2D location to toss

Idea: Initial thinking is to frame it as a binary classifier. Since the input data is extremely sparse(and sequentially recieved), I am struggling to find a way to approach this problem. Any suggestion will be hugely appreciated.



You must log in to answer this question.

Browse other questions tagged .