diff --git a/datajoint/autopopulate.py b/datajoint/autopopulate.py
index 753e42e5b..8efdcb37b 100644
--- a/datajoint/autopopulate.py
+++ b/datajoint/autopopulate.py
@@ -19,6 +19,7 @@ class AutoPopulate(metaclass=abc.ABCMeta):
     must define the property populated_from, and must define the callback method _make_tuples.
     """
     _jobs = None
+    _populated_from = None
 
     @property
     def populated_from(self):
@@ -28,13 +29,16 @@ def populated_from(self):
                 join of the parent relations. Users may override to change the granularity
                 or the scope of populate() calls.
         """
-        parents = [FreeRelation(self.target.connection, rel) for rel in self.target.parents]
-        if not parents:
-            raise DataJointError('A relation must have parent relations to be able to be populated')
-        ret = parents.pop(0)
-        while parents:
-            ret *= parents.pop(0)
-        return ret
+        if self._populated_from is None:
+            self.connection.dependencies.load()
+            parents = [FreeRelation(self.target.connection, rel) for rel in self.target.parents]
+            if not parents:
+                raise DataJointError('A relation must have parent relations to be able to be populated')
+            ret = parents.pop(0)
+            while parents:
+                ret *= parents.pop(0)
+            self._populated_from = ret
+        return self._populated_from
 
     @abc.abstractmethod
     def _make_tuples(self, key):
@@ -70,17 +74,16 @@ def populate(self, restriction=None, suppress_errors=False,
         if order not in valid_order:
             raise DataJointError('The order argument must be one of %s' % str(valid_order))
 
-        self.connection.dependencies.load()
-
-        if not isinstance(self.populated_from, RelationalOperand):
+        todo = self.populated_from
+        if not isinstance(todo, RelationalOperand):
             raise DataJointError('Invalid populated_from value')
+        todo.restrict(restriction)
 
         error_list = [] if suppress_errors else None
 
         jobs = self.connection.jobs[self.target.database]
         table_name = self.target.table_name
-        unpopulated = (self.populated_from & restriction) - self.target.project()
-        keys = unpopulated.fetch.keys()
+        keys = (todo - self.target.project()).fetch.keys()
         if order == "reverse":
             keys = list(keys).reverse()
         elif order == "random":
@@ -113,14 +116,14 @@ def populate(self, restriction=None, suppress_errors=False,
                             jobs.complete(table_name, key)
         return error_list
 
-
     def progress(self, restriction=None, display=True):
         """
         report progress of populating this table
         :return: remaining, total -- tuples to be populated
         """
-        total = len(self.populated_from & restriction)
-        remaining = len((self.populated_from & restriction) - self.target.project())
+        todo = self.populated_from & restriction
+        total = len(todo)
+        remaining = len(todo - self.target.project())
         if display:
             print('%-20s' % self.__class__.__name__, flush=True, end=': ')
             print('Completed %d of %d (%2.1f%%)   %s' %
diff --git a/datajoint/dependencies.py b/datajoint/dependencies.py
index abc42c84d..722fd5244 100644
--- a/datajoint/dependencies.py
+++ b/datajoint/dependencies.py
@@ -3,18 +3,6 @@
 from . import DataJointError
 from functools import wraps
 
-def load_dependencies(func):
-    """
-    Decorator that ensures that dependencies are loaded
-    """
-
-    @wraps(func)
-    def f(*args, **kwargs):
-        args[0].load()
-        return func(*args, **kwargs)
-    return f
-
-
 
 class Dependencies:
     """
@@ -32,23 +20,18 @@ def __init__(self, conn):
 
 
     @property
-    @load_dependencies
     def parents(self):
         return self._parents
 
-
     @property
-    @load_dependencies
     def children(self):
         return self._children
 
     @property
-    @load_dependencies
     def references(self):
         return self._references
 
     @property
-    @load_dependencies
     def referenced(self):
         return self._referenced
 
diff --git a/datajoint/erd.py b/datajoint/erd.py
index 5655f08e3..fefbc8031 100644
--- a/datajoint/erd.py
+++ b/datajoint/erd.py
@@ -19,15 +19,11 @@
     pygraphviz_layout = None
 
 import matplotlib.pyplot as plt
-from . import DataJointError
-from functools import wraps
-from .utils import to_camel_case
+from inspect import isabstract
 from .base_relation import BaseRelation
 
 logger = logging.getLogger(__name__)
 
-from inspect import isabstract
-
 
 def get_concrete_descendants(cls):
     desc = []
@@ -153,7 +149,7 @@ def plot(self):
         # draw non-primary key relations
         nx.draw_networkx_edges(self, pos, self.non_pk_edges, style='dashed', arrows=False)
         apos = np.array(list(pos.values()))
-        xmax = apos[:, 0].max() + 200 #TODO: use something more sensible then hard fixed number
+        xmax = apos[:, 0].max() + 200  # TODO: use something more sensible than hard fixed number
         xmin = apos[:, 0].min() - 100
         ax.set_xlim(xmin, xmax)
         ax.axis('off')  # hide axis
@@ -378,7 +374,7 @@ def remove_edges_in_path(self, path):
         Removes all shared edges between this graph and the path
         :param path: a list of nodes defining a path. All edges in this path will be removed from the graph if found
         """
-        if len(path) <= 1: # no path exists!
+        if len(path) <= 1:  # no path exists!
             return
         for a, b in zip(path[:-1], path[1:]):
             self.remove_edge(a, b)
diff --git a/datajoint/relational_operand.py b/datajoint/relational_operand.py
index d2faade5a..698201015 100644
--- a/datajoint/relational_operand.py
+++ b/datajoint/relational_operand.py
@@ -436,6 +436,7 @@ def restrict(self, *restrictions):
         has_restriction = any(isinstance(r, RelationalOperand) or r for r in restrictions)
         do_subquery = has_restriction and self.heading.computed
         if do_subquery:
+            # TODO fix this for the case (r & key).aggregate(m='compute')
             raise DataJointError('In-place restriction on renamed attributes is not allowed')
         super().restrict(*restrictions)