S-DAM
A research effort that injects Spelke core-knowledge priors from developmental psychology into modern Hopfield networks. All experimental results are disclosed through pre-registration, exploring how cognitive structural priors influence neural network learning efficiency.
Timeline (last 90 days)
S-DAM (Spelke-seeded Dense Associative Memory) is a study that introduces the core knowledge theory from developmental psychology into modern Hopfield networks, and publicly disclosed all results in a pre-registered manner, including one failed experiment.
The core idea of S-DAM is to encode structural prior knowledge as the initialization or memory patterns (seeding) of a modern Hopfield network, and to examine whether such priors can improve the model's learning efficiency or generalization ability.
S-DAM is not a work pursuing state-of-the-art performance, but rather an exploration in methodology and research attitude.
S-DAM represents a research direction distinct from purely large-data-plus-large-model approaches, advocating for improving sample efficiency by injecting structured priors, forming a complement or tension with the mainstream scaling paradigm.
All Facts (4)
S-DAM (Spelke-seeded Dense Associative Memory) is a study that introduces the core knowledge theory from developmental psychology into modern Hopfield networks, and publicly disclosed all results in a pre-registered manner, including one failed experiment.
50%UnverifiedThe core idea of S-DAM is to encode structural prior knowledge as the initialization or memory patterns (seeding) of a modern Hopfield network, and to examine whether such priors can improve the model's learning efficiency or generalization ability.
50%UnverifiedS-DAM is not a work pursuing state-of-the-art performance, but rather an exploration in methodology and research attitude.
50%UnverifiedS-DAM represents a research direction distinct from purely large-data-plus-large-model approaches, advocating for improving sample efficiency by injecting structured priors, forming a complement or tension with the mainstream scaling paradigm.
50%