diff --git a/src/nncf/common/tensor_statistics/collectors.py b/src/nncf/common/tensor_statistics/collectors.py
index 2ee9cd362b9..a3666b6718e 100644
--- a/src/nncf/common/tensor_statistics/collectors.py
+++ b/src/nncf/common/tensor_statistics/collectors.py
@@ -928,9 +928,13 @@ def __init__(self, num_samples: int | None = None):
 
     def _register_reduced_input_impl(self, x: Tensor) -> None:
         trace = fns.sum(fns.multiply(x, x))
-        # NOTE: average trace?? divide by number of diagonal elements
-        # TODO(dlyakhov): revise this formula as possibly it is with an error; adopted from previous HAWQ implementation
-        self._container = (self._container + trace) / x.size
+        # We normalize the trace by the number of elements in the tensor so that larger matrices do not dominate
+        # the sensitivity scores
+        self._container += trace / x.size
+
+    def reset(self) -> None:
+        self._collected_samples = 0
+        self._container = Tensor(0.0)
 
     def _aggregate_impl(self) -> Tensor:
         return self._container * 2 / self._collected_samples
diff --git a/tests/common/test_reducers_and_aggregators.py b/tests/common/test_reducers_and_aggregators.py
index 53b4fa98a6f..b0912751393 100644
--- a/tests/common/test_reducers_and_aggregators.py
+++ b/tests/common/test_reducers_and_aggregators.py
@@ -714,8 +714,8 @@ def test_aggregators_hash(self, aggregator_cls):
 
     HAWQ_AGGREGATOR_REFERENCE_VALUES = [
         ([np.arange(10)], 57.0),
-        ([np.arange(12).reshape((2, 6)), np.arange(24).reshape((4, 6))], 181.92361111111111),
-        ([np.arange(8 * i).reshape((1, 8, i)) for i in range(1, 5)], 165.61627197265625),
+        ([np.arange(12).reshape((2, 6)), np.arange(24).reshape((4, 6))], 222.33333333333331),
+        ([np.arange(8 * i).reshape((1, 8, i)) for i in range(1, 5)], 300.3333333333333),
     ]
 
     @pytest.mark.parametrize("inputs,reference_output", HAWQ_AGGREGATOR_REFERENCE_VALUES)