A talk by Raphaël Millière as part of a 2023 workshop on The Philosophy of Deep Learning: wp.nyu.edu/consciousness/the-...
Пікірлер: 2
@robinranabhat312511 ай бұрын
Time Stapms : 19:34: Compositional Generalization 20:24 : Synthetic Datasets to test if models learn Compositional Representation ## Methods for Understanding Black boxes 24:09 : Probing 25:54 : Tensor Products ( Learning Compositional Representation in continuous space ) , People like Paul Smolensky 30:11 : Tree Projection (Chris mannings lab) 32:26 : Mechanistic Interpretability 44:04 : Do DNN's implement classical Architecture ? 47:10 : Benefits of DNN-style Compositionality 48:00 : Conclusion
@robinranabhat312511 ай бұрын
47:41 Could someone provide a specific example with problem of Tractability on Classical systems ? I am already overwhelmed and don't want to discourse into trying to understand how classical systems work. Just an simple example or reference link would be enough. Thanks !