initially as the number of parameters increases, the bias of the model reduces and at one point (the valley) the number of parameters is enough to fit the model and bias is minimum. After this, we see the variance of the model increasing again till d ~ n. After that we see that graph descends again for the second time (called the double descent phenomenon). This part is called the overparameterized regime, because here the model is overparameterized with d > n