def linear_backward(dZ, cache):

    A_prev, W, b = cache
    m = A_prev.shape[1]

    dW, db, dA_prev = None, None, None
    dW = (1/m) * np.dot(dZ, cache[0].T) # cache[0] is A_prev or A[l-1]
    db = (1/m) * np.sum(dZ, axis = 1, keepdims = True)
    dA_prev = np.dot(cache[1].T, dZ) # here we use use W, actually W^T
    
    assert (dA_prev.shape == A_prev.shape)
    assert (dW.shape == W.shape)
    assert (db.shape == b.shape)

    return dA_prev, dW, db