diff --git a/examples/cifar10/cifar10.jl b/examples/cifar10/cifar10.jl
index 2ef638f..a4b3f95 100644
--- a/examples/cifar10/cifar10.jl
+++ b/examples/cifar10/cifar10.jl
@@ -47,16 +47,15 @@ lr_policy = LRPolicy.Staged(
   (5000, LRPolicy.Fixed(0.00001)),
 )
 solver_params = SolverParameters(max_iter=70000,
-    regu_coef=0.004, mom_policy=MomPolicy.Fixed(0.9), lr_policy=lr_policy)
+    regu_coef=0.004, mom_policy=MomPolicy.Fixed(0.9), lr_policy=lr_policy,
+    load_from="snapshots")
 solver = SGD(solver_params)
 
 # report training progress every 200 iterations
 add_coffee_break(solver, TrainingSummary(), every_n_iter=200)
 
 # save snapshots every 5000 iterations
-add_coffee_break(solver,
-    Snapshot("snapshots", auto_load=true),
-    every_n_iter=5000)
+add_coffee_break(solver, Snapshot("snapshots"), every_n_iter=5000)
 
 # show performance on test data every 1000 iterations
 test_net = Net("CIFAR10-test", backend, [data_tt_layer, common_layers..., acc_layer])
diff --git a/examples/cifar10/convert.jl b/examples/cifar10/convert.jl
index 018b42e..fa6171d 100644
--- a/examples/cifar10/convert.jl
+++ b/examples/cifar10/convert.jl
@@ -17,7 +17,7 @@ for (key, sources) in datasets
     dset_data = d_create(h5, "data", datatype(Float32), 
         dataspace(width, height, channels, batch_size * length(sources)))
     dset_label = d_create(h5, "label", datatype(Float32), 
-        dataspace(1,1,1, batch_size * length(sources)))
+        dataspace(1, batch_size * length(sources)))
 
     for n = 1:length(sources)
       open("cifar-10-batches-bin/$(sources[n])") do f
@@ -47,7 +47,7 @@ for (key, sources) in datasets
 
         index = (n-1)*batch_size+1:n*batch_size
         dset_data[:,:,:,index] = img
-        dset_label[:,:,:,index] = label
+        dset_label[:,index] = label
       end
     end
 
diff --git a/examples/mnist/mnist_dropout_fc.jl b/examples/mnist/mnist_dropout_fc.jl
index 2a55975..4b66407 100644
--- a/examples/mnist/mnist_dropout_fc.jl
+++ b/examples/mnist/mnist_dropout_fc.jl
@@ -68,25 +68,24 @@ drop_layers = [drop_input, drop_fc1, drop_fc2]
 # put training net together, note that the correct ordering will automatically be established by the constructor
 net = Net("MNIST-train", backend, [data_layer, common_layers..., drop_layers..., loss_layer])
 
+base_dir = "snapshots_dropout_fc"
 # we let the learning rate decrease by 0.998 in each epoch (=600 batches of size 100)
 # and let the momentum increase linearly from 0.5 to 0.9 over 500 epochs
 # which is equivalent to an increase step of 0.0008
 # training is done for 2000 epochs
 params = SolverParameters(max_iter=600*2000, regu_coef=0.0,
                           mom_policy=MomPolicy.Linear(0.5, 0.0008, 600, 0.9),
-                          lr_policy=LRPolicy.Step(0.1, 0.998, 600))
+                          lr_policy=LRPolicy.Step(0.1, 0.998, 600),
+                          load_from=base_dir)
 solver = SGD(params)
 
-base_dir = "snapshots_dropout_fc"
 setup_coffee_lounge(solver, save_into="$base_dir/statistics.jld", every_n_iter=5000)
 
 # report training progress every 100 iterations
 add_coffee_break(solver, TrainingSummary(), every_n_iter=100)
 
 # save snapshots every 5000 iterations
-add_coffee_break(solver,
-                 Snapshot(base_dir, auto_load=true),
-                 every_n_iter=5000)
+add_coffee_break(solver, Snapshot(base_dir), every_n_iter=5000)
 
 # show performance on test data every 600 iterations (one epoch)
 data_layer_test = HDF5DataLayer(name="test-data", source="data/test.txt", batch_size=100)