{"paper":{"title":"Visualising the Attractor Landscape of Neural Cellular Automata","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Neural cellular automata often show simple behavioral manifolds at the full-state level but complex ones when broken down to individual cells.","cross_cats":["cs.ET"],"primary_cat":"cs.NE","authors_text":"Alexander Mordvintsev, Harald Michael Ludwig, James Stovold, Mia-Katrin Kvalsund, Varun Sharma","submitted_at":"2026-04-12T13:39:17Z","abstract_excerpt":"As Neural Cellular Automata (NCAs) are increasingly applied outside of the toy models in Artificial Life, there is a pressing need to understand how they behave and to build appropriate routes to interpret what they have learnt. By their very nature, the benefits of training NCAs are balanced with a lack of interpretability: we can engineer emergent behaviour, but have limited ability to understand what has been learnt.\n  In this paper, we apply a variety of techniques to pry open the NCA black box and glean some understanding of what it has learnt to do. We apply techniques from manifold lear"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"When analysis is performed at a macroscopic level (i.e. taking the entire NCA state as a single data point), the underlying manifold is often quite simple and can be captured and analysed quite well. When analysis is performed at a microscopic level (i.e. taking the state of individual cells as a single data point), the manifold is highly complex and more complicated techniques are required in order to make sense of it.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the chosen manifold learning and topological techniques faithfully recover the true behavioral manifold without significant distortion or loss of dynamics that matter for the NCA's function.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Neural Cellular Automata show simple attractor manifolds at the full-state level but complex ones at the individual-cell level when analyzed with PCA, autoencoders, and persistent homology.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Neural cellular automata often show simple behavioral manifolds at the full-state level but complex ones when broken down to individual cells.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b7e95c3122564da93caf90528dd272363339964b9641017e7102163b863aea59"},"source":{"id":"2604.10639","kind":"arxiv","version":2},"verdict":{"id":"faeef505-c98c-496f-8d0c-529c041240b0","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T15:48:36.700178Z","strongest_claim":"When analysis is performed at a macroscopic level (i.e. taking the entire NCA state as a single data point), the underlying manifold is often quite simple and can be captured and analysed quite well. When analysis is performed at a microscopic level (i.e. taking the state of individual cells as a single data point), the manifold is highly complex and more complicated techniques are required in order to make sense of it.","one_line_summary":"Neural Cellular Automata show simple attractor manifolds at the full-state level but complex ones at the individual-cell level when analyzed with PCA, autoencoders, and persistent homology.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the chosen manifold learning and topological techniques faithfully recover the true behavioral manifold without significant distortion or loss of dynamics that matter for the NCA's function.","pith_extraction_headline":"Neural cellular automata often show simple behavioral manifolds at the full-state level but complex ones when broken down to individual cells."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.10639/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}