add reference

Summary:
We have these information in Reagent which has been useful for customers and ourselfs as well.

This quick fix aligned with reagent code base.

Reviewed By: rodrigodesalvobraz

Differential Revision: D53694510

fbshipit-source-id: fa53d69028faabb457b90553dc3aa03fa68699b4

pearl/policy_learners/contextual_bandits/neural_linear_bandit.py

@@ -50,6 +50,9 @@ class NeuralLinearBandit(ContextualBanditBase):
     Here _deep_represent_layers can be treated as featuer processing,
     and then processed features are fed into a linear regression layer to output predicted score.
     For example : features --> neural networks --> LinUCB --> UCB score
+
+    The implementation of LinearBandit refers to https://arxiv.org/pdf/1003.0146.pdf
+    The implementation of NeuralLinearBandit refers to https://arxiv.org/pdf/2012.01780.pdf
     """
 
     def __init__(