MAP IT to Visualize Representations
Permanent lenke
https://hdl.handle.net/10037/36771Dato
2024-01-16Type
Journal articleTidsskriftartikkel
Peer reviewed
Forfatter
Jenssen, RobertSammendrag
MAP IT visualizes representations by taking a fundamentally different approach to dimensionality reduction. MAP IT aligns distributions over discrete marginal probabilities in the input space versus the target space, thus capturing information in wider local regions, as opposed to current methods which align based on pairwise probabilities between states only. The MAP IT theory reveals that alignment
based on a projective divergence avoids normalization of weights (to obtain true probabilities) entirely, and further reveals a dual viewpoint via continuous densities and kernel smoothing. MAP IT is shown to produce visualizations which capture class structure better than the current state of the art.
Forlag
ICLR (International Conference on Learning Representation)Sitering
Jenssen. MAP IT to Visualize Representations. International Conference on Learning Representations. 2024Metadata
Vis full innførselSamlinger
Copyright 2024 The Author(s)