Limits to connectionism

Marcus, G. F., Vijayan, S., Rao, S. B., & Vishton, P. M. (1999). Rule learning by seven-month-old infants. Science, 283, 77—80. Also read the responses.

• Your goal here should be to understand Marcus et al.’s experimental paradigm and why they think it means that human cognition cannot operate the way that neural networks do. Do you agree? Do you find the criticisms compelling?

McClelland, J. L., & Plaut, D. C. (1999). Does generalization in infant learning implicate abstract algebra-like rules?. Trends in Cognitive Sciences, 3, 166—168. Also read the Marcus response.

• This is a more detailed objection to the Marcus (1999) argument. Make sure you understand what they are suggesting that networks can learn, and also why Marcus is not impressed.

Marcus, G. (2018). Deep learning: A critical appraisal. arXiv preprint.

• Neural networks in 2018 are much more impressive than they were in 1999. And yet, Marcus is still concerned. As you read this, think about whether the same arguments are being made here as in his 1999 paper. Are previous arguments refuted? Are there new compelling arguments?