I have this optimization problem:
$$\max_{a} \min_{\alpha} \ \sum_{i,j} w_{ij} (s_i - s_j)^2 I $$ $$I = 1 \text{ if } w_{ij} \ge \alpha \text{, and } (s_i - s_j)^2 \le a \ ; 0 \text{ otherwise} $$
Notes: $w_{ij} \text{ is normalized between }[0,1].$ $\text{It would be ok if the restriction }(s_i - s_j)^2 \le a \text{ was replaced by } |s_i - s_j| \le a$
I am not entirely sure where to start to solve this problem. But I would start by asking the questions: How to relax this optimization problem? And whether it fits well-known optimization problems.