Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning..

📖 Description

Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning Sreejan Kumar 1* , Ishita Dasgupta 2 , Nathaniel D. Daw 13 , Jonathan. D. Cohen 13 , Thomas L. Grif fiths 34 1 Princeton Neuroscience Institute 2 DeepMind 3 Princeton University Department of Psychology 4 Princeton University Department of Computer Science *: Corresponding author: [email removed] Abstract The ability to acquire abstract knowledge is a hallmark of human intelligence and is believed by many to be one of the core dif ferences between humans and neural network models. Agents can be endowed with an inductive bias towards abstraction through meta-learning, where they are trained on a distribution of tasks that share some abstract structure that can be learned and applied. However , because neural networks are hard to interpret, it can be dif ficult to tell whether agents have learned the underlying abstraction, or alternatively statistical patterns that are characteristic of t...

📊 Game Impacts

💭 Reactions

🔗 Sources

• [1] https://arxiv.org/abs/2204.01437