CITRIS: Causal Identifiability from Temporal Intervened Sequences
What this work does
Learns visual causal representations from temporal transitions with observed intervention targets and an intervention-conditioned latent transition model.
Connection to discovery
Shows why observation representation and available intervention diversity must be specified before measuring discovery-policy ability.
Evidence
Inspected the minimal-causal-variable formulation, a necessary intervention-separation condition and interventional visual benchmarks.
Scope and limitations
Guarantees are conditional on the temporal/generative assumptions and intervention coverage. Factors always manipulated together may not be separately identifiable.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.