{"paper":{"title":"Perceiver IO: A General Architecture for Structured Inputs & Outputs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Perceiver IO adds a flexible querying mechanism to the Perceiver so one architecture processes arbitrary structured inputs and produces outputs of any size or type while scaling linearly.","cross_cats":["cs.CL","cs.CV","cs.SD","eess.AS"],"primary_cat":"cs.LG","authors_text":"Andrew Brock, Andrew Jaegle, Andrew Zisserman, Carl Doersch, Catalin Ionescu, Daniel Zoran, David Ding, Evan Shelhamer, Jean-Baptiste Alayrac, Jo\\=ao Carreira, Matthew M. Botvinick, Olivier H\\'enaff, Oriol Vinyals, Sebastian Borgeaud, Skanda Koppula","submitted_at":"2021-07-30T17:53:34Z","abstract_excerpt":"A central goal of machine learning is the development of systems that can solve many problems in as many data domains as possible. Current architectures, however, cannot be applied beyond a small set of stereotyped settings, as they bake in domain & task assumptions or scale poorly to large inputs or outputs. In this work, we propose Perceiver IO, a general-purpose architecture that handles data from arbitrary settings while scaling linearly with the size of inputs and outputs. Our model augments the Perceiver with a flexible querying mechanism that enables outputs of various sizes and semanti"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The same architecture achieves strong results on tasks spanning natural language and visual understanding, multi-task and multi-modal reasoning, and StarCraft II. As highlights, Perceiver IO outperforms a Transformer-based BERT baseline on the GLUE language benchmark despite removing input tokenization and achieves state-of-the-art performance on Sintel optical flow estimation with no explicit mechanisms for multiscale correspondence.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the added flexible querying mechanism can produce outputs of arbitrary sizes and semantics across domains without introducing hidden task-specific assumptions or requiring per-task architectural changes that undermine the generality claim.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Perceiver IO is a general architecture that processes arbitrary structured inputs and outputs with linear scaling and achieves strong results on GLUE, Sintel optical flow, multi-task reasoning, and StarCraft II without task-specific components.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Perceiver IO adds a flexible querying mechanism to the Perceiver so one architecture processes arbitrary structured inputs and produces outputs of any size or type while scaling linearly.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"96f5fabd157d02586b1607250c9aec59ae2e9087dbfbf6f42b6bf819c6df6cf1"},"source":{"id":"2107.14795","kind":"arxiv","version":3},"verdict":{"id":"c6c19aed-5b3e-443d-8004-adc57417d892","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T19:43:58.764181Z","strongest_claim":"The same architecture achieves strong results on tasks spanning natural language and visual understanding, multi-task and multi-modal reasoning, and StarCraft II. As highlights, Perceiver IO outperforms a Transformer-based BERT baseline on the GLUE language benchmark despite removing input tokenization and achieves state-of-the-art performance on Sintel optical flow estimation with no explicit mechanisms for multiscale correspondence.","one_line_summary":"Perceiver IO is a general architecture that processes arbitrary structured inputs and outputs with linear scaling and achieves strong results on GLUE, Sintel optical flow, multi-task reasoning, and StarCraft II without task-specific components.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the added flexible querying mechanism can produce outputs of arbitrary sizes and semantics across domains without introducing hidden task-specific assumptions or requiring per-task architectural changes that undermine the generality claim.","pith_extraction_headline":"Perceiver IO adds a flexible querying mechanism to the Perceiver so one architecture processes arbitrary structured inputs and produces outputs of any size or type while scaling linearly."},"references":{"count":103,"sample":[{"doi":"","year":2012,"title":"Imitating interactive intelligence, 2021, 2012.05672 http://arxiv.org/abs/2012.05672","work_id":"993f94c8-0c91-4ae3-9f40-68c03b5723d3","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"VATT : Transformers for multimodal self-supervised learning from raw video, audio and text","work_id":"40471e6e-c402-41c0-9da0-b94ca27bc316","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Self-supervised multimodal versatile networks","work_id":"222413c3-71e9-4e35-afe9-7d3cb141429e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"The D eep M ind JAX E cosystem, 2020","work_id":"c47be3c1-be1b-4120-8234-644c5355e25f","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2004,"title":"Longformer: The Long-Document Transformer","work_id":"abea7a44-6668-4de7-aab6-f53a6e5aa088","ref_index":5,"cited_arxiv_id":"2004.05150","is_internal_anchor":true}],"resolved_work":103,"snapshot_sha256":"bc2a8846c668d4d21559628741d95fc4c8692ae1ee00a33e534a8bb0e002d42b","internal_anchors":7},"formal_canon":{"evidence_count":1,"snapshot_sha256":"83adb9dc241655ce4a3529e67ad4544ec9ccf4b3fb6fa1c1ee015b7a34a21465"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}