Sequence learning recodes cortical representations instead of strengthening initial ones.

Kristjan Kalm,Dennis Norris,Ronald Van Den Berg

doi:10.1371/journal.pcbi.1008969

Abstract

We contrast two computational models of sequence learning. The associative learner posits that learning proceeds by strengthening existing association weights. Alternatively, recoding posits that learning creates new and more efficient representations of the learned sequences. Importantly, both models propose that humans act as optimal learners but capture different statistics of the stimuli in their internal model. Furthermore, these models make dissociable predictions as to how learning changes the neural representation of sequences. We tested these predictions by using fMRI to extract neural activity patterns from the dorsal visual processing stream during a sequence recall task. We observed that only the recoding account can explain the similarity of neural activity patterns, suggesting that participants recode the learned sequences using chunks. We show that associative learning can theoretically store only very limited number of overlapping sequences, such as common in ecological working memory tasks, and hence an efficient learner should recode initial sequence representations.

Highlights

We investigate the neural mechanism involved in learning short visual sequences
We show that associative learning without recoding is not theoretically capable of supporting long-term memory of short ecological sequences present in every day tasks such as reading, speaking, or navigating
The crucial difference between these two learning approaches is that the dimensionality of sequence representations changes: for associative learning the sequence representations remain the same, whilst new codes are inferred with recoding (Fig 1D)

Summary

Introduction

We investigate the neural mechanism involved in learning short visual sequences. An optimal learner is an agent whose internal model reflects the statistics of the environment [5, 6], and human learning has been shown to follow the optimal model in a wide range of domains such as speech and language [7, 8], visual scenes and objects [9,10,11,12,13], and sensorimotor control [14, 15]. An optimal learner would update the association weights as new data comes in to reflect the statistics of the environment. Learning can proceed by recoding frequently occurring associations using new latent representations. The latter approach has been termed ‘chunking’ in cognitive literature [16, 17] to describe learning where complex objects (words, faces) are constructed from lower-level features (phonemes, syllables, oriented lines, [10]). We can dissociate between these two mechanisms by comparing neural representations of novel sequences to learned ones

Results

Discussion

Conclusion