#positive-homogeneity

Question 1: Single choice

Permuting the hidden units is one symmetry of an MLP. Which other change of the parameters leaves the function computed by a ReLU network exactly the same for every input?

Add the same constant to all incoming weights of one hidden unit1
Scale one hidden unit's incoming weights and bias by $c > 0$ and its outgoing weights by $1/c$ 2
Flip the signs of one hidden unit's incoming weights, of its bias and of all its outgoing weights3
Multiply one hidden unit's incoming weights and its outgoing weights both by $c > 0$ 4

#positive-homogeneity

Timed test