diff --git a/bayesflow/networks/summary_network.py b/bayesflow/networks/summary_network.py
index e821be3f3..d7df0b476 100644
--- a/bayesflow/networks/summary_network.py
+++ b/bayesflow/networks/summary_network.py
@@ -53,4 +53,6 @@ def compute_metrics(self, x: Tensor, stage: str = "training", **kwargs) -> dict[
 
     @classmethod
     def from_config(cls, config, custom_objects=None):
+        if hasattr(cls.get_config, "_is_default") and cls.get_config._is_default:
+            return cls(**config)
         return cls(**deserialize(config, custom_objects=custom_objects))
diff --git a/bayesflow/workflows/basic_workflow.py b/bayesflow/workflows/basic_workflow.py
index c30271eb1..34fa03794 100644
--- a/bayesflow/workflows/basic_workflow.py
+++ b/bayesflow/workflows/basic_workflow.py
@@ -914,6 +914,7 @@ def build_optimizer(self, epochs: int, num_batches: int, strategy: str) -> keras
             self.optimizer = keras.optimizers.Adam(learning_rate, clipnorm=1.5)
         else:
             self.optimizer = keras.optimizers.AdamW(learning_rate, weight_decay=5e-3, clipnorm=1.5)
+        return self.optimizer
 
     def _fit(
         self,
@@ -955,9 +956,10 @@ def _fit(
             else:
                 kwargs["callbacks"] = [model_checkpoint_callback]
 
-        self.build_optimizer(epochs, dataset.num_batches, strategy=strategy)
-
-        if not self.approximator.built:
+        # returns None if no new optimizer was built and assigned to self.optimizer, which indicates we do not have
+        # to (re)compile the approximator.
+        optimizer = self.build_optimizer(epochs, dataset.num_batches, strategy=strategy)
+        if optimizer is not None:
             self.approximator.compile(optimizer=self.optimizer, metrics=kwargs.pop("metrics", None))
 
         try: