M Tschannen, C Eastwood, F Mentzer

We introduce generative transformers for real-valued vector sequences, rather than discrete tokens from a finite vocabulary.

C Eastwood*, S Singh*, A Nicolicioiu, M Vlastelica, J von Kügelgen, B Schölkopf

Code

We show how a weak-but-stable training signal can be used to harness complementary spurious features, boosting performance.

(Previously @ ICML 2023 Spurious Correlations Workshop)

C Eastwood*, A Robey*, S Singh, J von Kügelgen, H Hassani, G J Pappas, B Schölkopf

Code / Video

We learn predictors that generalize with a desired probability and argue for better evaluation protocols in domain generalization.

C Eastwood*, I Mason*, CKI Williams, B Schölkopf

Code

We address "measurement shift" (e.g., a new hospital scanner) by restoring the same features rather than learning new ones.

C Eastwood, CKI Williams

Code

We propose the DCI framework for evaluating "disentangled" representations.

(Previously Spotlight @ NeurIPS 2017 Disentanglement workshop)

C Eastwood, J von Kügelgen, L Ericsson, D Bouchacourt, P Vincent, B Schölkopf, M Ibrahim

We use data augmentations to disentangle rather than discard.

(Previously @ NeurIPS 2023 workshops on Self-Supervised Learning and Causal Representation Learning)

C Eastwood*, A Nicolicioiu*, J von Kügelgen*, A Kekić, F Träuble, A Dittadi, B Schölkopf

Code

We extend the DCI framework by quantifying the ease-of-use or explicitness of a representation.

(Previously @ UAI 2022 Causal Repr. Learning Workshop)

C Eastwood*, N Li*, CKI Williams

We propose a framework for explaining object-image differences in terms of the underlying object properties.

C Eastwood*, I Mason*, CKI Williams

Code / Video

We use surprising unit-level activations to determine which parameters to adapt for a given distribution shift.

N Li, C Eastwood, B Fisher

Code / Video

We learn accurate, object-centric representations of 3D scenes by aggregating information from multiple 2D views/observations.