{"paper":{"title":"Diffusion-Inspired Reconfiguration of Transformers for Uncertainty Calibration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Modeling each transformer block as a probabilistic mapping creates a diffusion-like path that propagates representation uncertainty without changing predictions.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bryan Kian Hsiang Low, Manh Cuong Dao, Phi Le Nguyen, Quang Hung Pham, Thao Nguyen Truong, Trong Nghia Hoang","submitted_at":"2026-02-09T17:24:47Z","abstract_excerpt":"Uncertainty calibration in pre-trained transformers is critical for their reliable deployment in risk-sensitive applications. Yet, most existing pre-trained transformers do not have a principled mechanism for uncertainty propagation through their feature transformation stack. In this work, we propose a diffusion-inspired reconfiguration of transformers in which each feature transformation block is modeled as a probabilistic mapping. Composing these probabilistic mappings reveals a probability path that mimics the structure of a diffusion process, transporting data mass from the input distribut"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Composing these probabilistic mappings reveals a probability path that mimics the structure of a diffusion process, transporting data mass from the input distribution to the pre-trained feature distribution. This probability path can then be recompiled on a diffusion process with a unified transition model to enable principled propagation of representation uncertainty throughout the pre-trained model's architecture while maintaining its original predictive performance.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That modeling each feature transformation block as a probabilistic mapping accurately captures and propagates representation uncertainty without introducing systematic biases or changing the model's learned behavior in unintended ways.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Diffusion-inspired probabilistic mappings enable principled uncertainty propagation through pre-trained transformer layers without degrading accuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Modeling each transformer block as a probabilistic mapping creates a diffusion-like path that propagates representation uncertainty without changing predictions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6161b6dd8c8fddb1cf746345ccdb39cc289ed57445c1a34a517c424c7b44105f"},"source":{"id":"2602.08920","kind":"arxiv","version":2},"verdict":{"id":"4adc9c85-68b9-4c0d-9495-ac42867f84bf","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T05:26:55.547267Z","strongest_claim":"Composing these probabilistic mappings reveals a probability path that mimics the structure of a diffusion process, transporting data mass from the input distribution to the pre-trained feature distribution. This probability path can then be recompiled on a diffusion process with a unified transition model to enable principled propagation of representation uncertainty throughout the pre-trained model's architecture while maintaining its original predictive performance.","one_line_summary":"Diffusion-inspired probabilistic mappings enable principled uncertainty propagation through pre-trained transformer layers without degrading accuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That modeling each feature transformation block as a probabilistic mapping accurately captures and propagates representation uncertainty without introducing systematic biases or changing the model's learned behavior in unintended ways.","pith_extraction_headline":"Modeling each transformer block as a probabilistic mapping creates a diffusion-like path that propagates representation uncertainty without changing predictions."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"0c33b46ca9602e2637640ab2ffa2bb2450aae2f8e4bee5ed7771a040d80460cc"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}