Cian Eastwood

I'm a Senior Research Scientist at Relation, working on generative models of biological data.

I am particularly interested in diffusion/flow models and ways to achieve generalisation, target-property guidance, and multimodality.

Previously, I was at Valence Labs / Recursion, working on generative models of cell microscopy images. I completed my PhD jointly at the University of Edinburgh (with Chris Williams) and the Max Planck Institute for Intelligent Systems (with Bernhard Schölkopf), working on generalisation and representation learning. During my PhD, I spent time at Google DeepMind, Meta AI (FAIR) and Spotify.

Selected Publications

	Elucidating the Design Space of Flow Matching for Cellular Microscopy C Jones, E Noutahi, J Hartford, C Eastwood** Preprint 2026 \| Code We develop a simple and scalable recipe for training flow-matching models on cell-microscopy images.
	GIVT: Generative Infinite-Vocabulary Transformers M Tschannen, C Eastwood, F Mentzer ECCV 2024 \| Code We introduce generative transformers for real-valued vector sequences, rather than discrete tokens from a finite vocabulary.
	Using Invariant Predictions to Harness Spurious Features C Eastwood, S Singh, A Nicolicioiu, M Vlastelica, J von Kügelgen, B Schölkopf NeurIPS 2023 \| Code We show how weak-but-robust predictions can be used to harness complementary spurious features in a new domain, boosting out-of-distribution performance. (Previously @ ICML 2023 Spurious Correlations Workshop)
	Probable Domain Generalization via Quantile Risk Minimization C Eastwood, A Robey, S Singh, J von Kügelgen, H Hassani, G J Pappas, B Schölkopf NeurIPS 2022 \| Code \| Video We learn predictors that generalise with a desired probability and argue for better evaluation protocols in domain generalisation.

Other Publications

	Unpaired Multimodal Learning for Biological Datasets Z Ji, C Eastwood, A Goldenberg, PP Liang, J Hartford, RG Krishnan, E Noutahi Medical Imaging with Deep Learning 2026 We propose a contrastive multimodal method that only requires weak treatment-group labels rather than paired data.
	TxPert: Leveraging biochemical relationships for out-of-distribution transcriptomic perturbation prediction F Wenkel, W Tu, C Masschelein, H Shirzad, C Eastwood, ST Whitfield, I Bendidi, C Russell, L Hodgson, Y El Mesbahi, J Ding, MM Fay, B Earnshaw, E Noutahi, AK Denton Preprint 2025 We explore how knowledge graphs of gene-gene relationships can improve out-of-distribution predictions.
	Towards scientific discovery with dictionary learning: Extracting biological concepts from microscopy foundation models K Donhauser, K Ulicna, GE Moran, A Ravuri, K Kenyon-Dean, C Eastwood, J Hartford ICML 2025 We extract interpretable biological concepts from microscopy foundation models. (Previously Oral @ NeurIPS 2024 workshop on Interpretable AI)
	Self-Supervised Disentanglement by Leveraging Structure in Data Augmentations C Eastwood, J von Kügelgen, L Ericsson, D Bouchacourt, P Vincent, B Schölkopf, M Ibrahim Preprint 2023 \| Code We use data augmentations to disentangle rather than discard. (Previously @ NeurIPS 2023 workshops on SSL and CRL)
	DCI-ES: An Extended Disentanglement Framework C Eastwood, A Nicolicioiu, J von Kügelgen, A Kekić, F Träuble, A Dittadi, B Schölkopf ICLR 2023* \| Code We extend the DCI framework by quantifying representation "explicitness". (Previously @ UAI 2022 Causal Repr. Learning Workshop)
	Source-Free Adaptation to Measurement Shift via Bottom-Up Feature Restoration C Eastwood, I Mason, CKI Williams, B Schölkopf ICLR 2022 (Spotlight) \| Code We address "measurement shift" (e.g., a new hospital scanner) by restoring the same features rather than learning new ones.
	Unit-Level Surprise in Neural Networks C Eastwood, I Mason, CKI Williams NeurIPS 2021 ICBINB Workshop (Spotlight & Didactic Award) \| Code \| Video We use unit-level activation "surprise" to determine which parameters to adapt for a given distribution shift.
	Learning Object-Centric Representations of Multi-Object Scenes from Multiple Views N Li, C Eastwood, B Fisher NeurIPS 2020 (Spotlight) \| Code \| Video We learn accurate, object-centric representations of 3D scenes by aggregating information from multiple 2D views/observations.
	A Framework for the Quantitative Evaluation of Disentangled Representations C Eastwood, CKI Williams ICLR 2018 \| Code We propose the DCI framework for evaluating "disentangled" representations. (Previously Spotlight @ NeurIPS 2017 Disentanglement workshop)

Website source