Pith. sign in

Representation Alignment in Neural Networks

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it
abstract

It is now a standard for neural network representations to be trained on large, publicly available datasets, and used for new problems. The reasons for why neural network representations have been so successful for transfer, however, are still not fully understood. In this paper we show that, after training, neural network representations align their top singular vectors to the targets. We investigate this representation alignment phenomenon in a variety of neural network architectures and find that (a) alignment emerges across a variety of different architectures and optimizers, with more alignment arising from depth (b) alignment increases for layers closer to the output and (c) existing high-performance deep CNNs exhibit high levels of alignment. We then highlight why alignment between the top singular vectors and the targets can speed up learning and show in a classic synthetic transfer problem that representation alignment correlates with positive and negative transfer to similar and dissimilar tasks.

citation-role summary

background 1

citation-polarity summary

fields

cs.LG 1

years

2025 1

verdicts

CONDITIONAL 1

roles

background 1

polarities

background 1

representative citing papers

Model alignment using inter-modal bridges

cs.LG · 2025-05-18 · conditional · novelty 6.0

A semi-supervised conditional flow matching method, guided by a new inter-modal bridge cost, aligns latent spaces of distinct pre-trained models using only a small set of paired samples.

citing papers explorer

Showing 1 of 1 citing paper.

  • Model alignment using inter-modal bridges cs.LG · 2025-05-18 · conditional · none · ref 10 · internal anchor

    A semi-supervised conditional flow matching method, guided by a new inter-modal bridge cost, aligns latent spaces of distinct pre-trained models using only a small set of paired samples.