Feedback
feedback
Latest
Comparing supervised learning dynamics: Deep neural networks match human data efficiency but show a generalisation lag
Recent research has seen many behavioral comparisons between humans and deep neural networks (DNNs) in the domain of image classification. Often, comparison studies focus on the end-result of the learning process by measuring and comparing the similarities in the representations of object categories once they have been formed. However, the process of how these representations emerge—that is, the behavioral changes and intermediate stages observed during the acquisition—is less often directly and empirically compared. In this talk, I'm going to report a detailed investigation of the learning dynamics in human observers and various classic and state-of-the-art DNNs. We develop a constrained supervised learning environment to align learning-relevant conditions such as starting point, input modality, available input data and the feedback provided. Across the whole learning process we evaluate and compare how well learned representations can be generalized to previously unseen test data. Comparisons across the entire learning process indicate that DNNs demonstrate a level of data efficiency comparable to human learners, challenging some prevailing assumptions in the field. However, our results also reveal representational differences: while DNNs' learning is characterized by a pronounced generalisation lag, humans appear to immediately acquire generalizable representations without a preliminary phase of learning training set-specific information that is only later transferred to novel data.
Removing information from working memory
Holding information in working memory is essential for cognition, but removing unwanted thoughts is equally important. There is great flexibility in how we can manipulate information in working memory, but the processes and consequences of these operations are poorly understood. In this talk I will discuss our recent findings using multivariate pattern analyses of fMRI brain data to demonstrate the successful removal of information from working memory using three different strategies: suppressing a specific thought, replacing a thought with a different one, and clearing the mind of all thought. These strategies are supported by distinct brain regions and have differential consequences on the encoding of new information. I will discuss implications of these results on theories of memory and I will highlight some new directions involving the use of real-time neurofeedback to investigate causal links between brain and behavior.
feedback coverage
2 items