{"paper":{"title":"Twincher: Bijective Representation Learning for Robust Inversion of Continuous Systems","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Twincher learns bijective representations of outputs aligned with parameters to enable robust inversion of continuous systems under noise.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Arkady Gonoskov","submitted_at":"2026-05-13T12:57:17Z","abstract_excerpt":"Recent advances in AI have been primarily driven by large-scale neural architectures that excel at function approximation, rather than by tailored inductive biases and inference or learning strategies that could be important for resource-efficient real-world perception and planning through the solution of inverse problems. In this work, we consider the possibility of enabling robust inversion of continuous forward processes $p \\mapsto y$ by learning representations of $y$ that are bijectively aligned with $p$ while remaining insensitive to perturbations in $y$ caused by noise or model mismatch"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Twincher enables robust and efficient iterative inverse inference by learning bijective representations of y that are aligned with p while remaining insensitive to perturbations in y caused by noise or model mismatch, exhibiting improved data efficiency and robustness compared to a baseline inverse-modeling approach.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That stacks of structured diffeomorphic transformations combined with adversarial training will produce representations that remain bijectively aligned and perturbation-insensitive when applied to real-world continuous systems beyond the synthetic examples shown.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Twincher learns bijective representations of observations aligned with continuous system parameters to enable robust iterative inversion, showing better data efficiency and noise tolerance than standard inverse modeling on synthetic systems.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Twincher learns bijective representations of outputs aligned with parameters to enable robust inversion of continuous systems under noise.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b1e01904efdc8ebf329b00dc08968036a44fa003a57a5bbe5055c69ea0812664"},"source":{"id":"2605.13470","kind":"arxiv","version":1},"verdict":{"id":"6e5b52cc-492f-4f97-a3ca-e6cdc0c9bc61","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:09:05.488911Z","strongest_claim":"Twincher enables robust and efficient iterative inverse inference by learning bijective representations of y that are aligned with p while remaining insensitive to perturbations in y caused by noise or model mismatch, exhibiting improved data efficiency and robustness compared to a baseline inverse-modeling approach.","one_line_summary":"Twincher learns bijective representations of observations aligned with continuous system parameters to enable robust iterative inversion, showing better data efficiency and noise tolerance than standard inverse modeling on synthetic systems.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That stacks of structured diffeomorphic transformations combined with adversarial training will produce representations that remain bijectively aligned and perturbation-insensitive when applied to real-world continuous systems beyond the synthetic examples shown.","pith_extraction_headline":"Twincher learns bijective representations of outputs aligned with parameters to enable robust inversion of continuous systems under noise."},"references":{"count":51,"sample":[{"doi":"","year":2020,"title":"Scaling laws for neural language models","work_id":"b48c8baf-a113-4a87-b196-8c68ba20002a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1998,"title":"Gradient-based learning applied to document recognition.Proc","work_id":"ab3edb5c-5be8-4527-84cf-1099f7ff4fce","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Gomez, Lukasz Kaiser, and Illia Polosukhin","work_id":"d33dc346-1e7f-4df8-9c56-7ee1cda28d9e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"On the inductive bias of neural tangent kernels, 2019","work_id":"96438aa8-cab6-4796-9fa4-7162495cc90d","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Challenging common assumptions in the unsupervised learning of disentangled representations","work_id":"ba4ff1c1-53b6-437d-b3b2-79ec991316d3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":51,"snapshot_sha256":"4bf67b2bdfe8922c5a9c29244614d825271b66418795e58c54c5348c92d57ee4","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"}