Disentanglement

Rotating Features for Object Discovery

This paper introduces several advancements for continuous and distributed object-centric representations, scaling them from simple toy to real-world data.

Causal Representation Learning for Instantaneous and Temporal Effects in Interactive Systems

We present iCITRIS, a causal representation learning method that can identify causal variables with instantaneous effects and their graph from temporal sequences. Published at ICLR 2023.

CITRIS - Causal Identifiability from Temporal Intervened Sequences

We present CITRIS, a causal representation learning algorithm for multidimensional causal factors identified from videos with interventions. Published at ICML 2022 (**Spotlight**).

Complex-Valued Autoencoders for Object Discovery

We perform unsupervised object-centric representation learning by exploiting complex-valued activations in an unconstrained autoencoder architecture.