diff --git a/app/src/models.py b/app/src/models.py
index 61d525e..436ce6a 100644
--- a/app/src/models.py
+++ b/app/src/models.py
@@ -1,5 +1,6 @@
 import json
 import logging
+from tensorflow.keras import regularizers
 from tensorflow.keras.callbacks import Callback, ModelCheckpoint, TensorBoard
 from tensorflow.keras.layers import (
     Activation,
@@ -19,6 +20,7 @@
 PARAM_ACTIVATION = "activation"
 PARAM_INPUT_SHAPE = "input_shape"
 PARAM_RETURN_SEQUENCES = "return_sequences"
+PARAM_POOL_SIZE = "pool_size"
 PARAM_UNITS = "units"
 PARAM_VERBOSE = "verbose"
 
@@ -47,6 +49,21 @@ def add_ic_layer(model, name, dropout=0.2, batchnorm=True):
 
 
 def add_layer(model, layer, name, ic_params={}, **kwargs):
+    for arg in kwargs:
+        if not arg.endswith("_regularizer"):
+            continue
+        config = kwargs[arg]
+        if config is None or isinstance(config, regularizers.Regularizer):
+            continue
+        reg = (
+            regularizers.L1L2(**config)
+            if type(config) is dict
+            else regularizers.get(config)
+        )
+        logger.debug(
+            f"Adding regularizer {type(reg).__name__} to {name}: {reg.get_config()}"
+        )
+        kwargs[arg] = reg
     logger.debug(f'Adding layer "{name}" - {layer.__name__}: {kwargs}')
     model.add(layer(name=name, **kwargs))
     if type(ic_params) is not dict:
@@ -56,7 +73,11 @@ def add_layer(model, layer, name, ic_params={}, **kwargs):
 
 def add_conv1d_layer(model, name, ic_params={}, pool=None, **kwargs):
     add_layer(model, Conv1D, name, ic_params=None, **kwargs)
-    if pool is not None:
+    if pool:
+        if pool is True:
+            pool = {}
+        elif type(pool) is int:
+            pool = {PARAM_POOL_SIZE: pool}
         add_layer(model, MaxPooling1D, f"{name}_pool", ic_params=None, **pool)
     add_ic_layer(model, name, **ic_params)
 
diff --git a/docs/Usage/Usage.md b/docs/Usage/Usage.md
index 399beb2..35fd511 100644
--- a/docs/Usage/Usage.md
+++ b/docs/Usage/Usage.md
@@ -80,8 +80,6 @@ Channel name for Right Aux. Must be provided if Right Aux has data, otherwise ch
 ### Train
 Builds a model (currently only LSTM is supported) and saves the built model and diagram image in `MODEL_DIR`. If `--epochs` is not 0, also trains the model on the data in `DATA_FILE` and saves the trained model and training history to `MODEL_DIR`. Even if `--epochs` is not 0, `DATA_FILE` is required to determine the input size to the LSTM.
 
-Note that if `--shuffle-samples` is not true, only samples belonging to contiguous sequences of length `--sequence-size` are used.
-
 ```bash
 train [arguments] DATA_FILE MODEL_DIR
 ```
@@ -153,3 +151,10 @@ Save TensorBoard logs every epoch. Default is false
 
 **`-g, --gradient-metrics`**  
 Print metrics in Gradient chart format every epoch. Default is false
+
+#### Usage Notes
+* If `--shuffle-samples` is not true, only samples belonging to contiguous sequences of length `--sequence-size` are used.
+* Layer specification parameters (e.g. `--lstm`, `--conv1d`) should be list of objects in JSON format, where each object in the list contains the specification of a single layer. To create only one layer, you can also use a single JSON object instead of an array of length 1. Please refer to the YAML files in the .ps_project folder for examples.
+* Each layer specification object can include one or more regularization parameters (e.g. `kernel_regularizer`). The value of these parameters should be either a string or a dict. For example:
+  * `"l1"` or `"l2"` will create the corresponding regularizer with default parameters.
+  * `{"l1": 0.1, "l2": 0.2}` will create an l1_l2 regularizer with parameters `l1=0.1, l2=0.2`. You don't need to include both l1 and l2 values.
\ No newline at end of file