This paper can more or less be summed into one simple question; “How could machines learn as efficiently as humans and animals?”. Not a simple question to answer and perhaps why this paper serves simply as a proposal to future directions of study rather than a conventional technical paper that shares novel research.
While prior methods typically decouple the two jobs with separate models, deep learning, provides us with a general and simple baseline: combine an off-the-shelf RL algorithm with a recurrent neural network (RNN; e.g., LSTM (Hochreiter & Schmidhuber, 1997) and GRU (Chung et al., 2014)).
SimCLR learning process The authors tested numerous augmentations, as shown below.
SimCLRv2 with a teacher network, which is trained on unmarked data using the initial SimCLR pattern, and a more compact student network, trained on the markup provided by the first model.
The Brain Team at Google...