RL Reasoning Prompt-Purification and Rollout-Calibration Diagram
Overview
A two-stage diagram of reinforcement learning for reasoning with prompt purification. It is an editable PowerPoint diagram adapted from the original paper figure under CC BY 4.0.
What the diagram shows
Stage I, identify and purify interference tokens: a low-success prompt xi (a row of prompt-token cells) → a bar chart of interference scores with a dashed threshold γ, where the two bars above the line are the candidate interference tokens → a denoised prompt x′i in which those tokens become hatched interference cells; a legend explains prompt tokens, interference tokens, correct rollouts and wrong rollouts. Stage II, calibrated rollouts policy optimisation: a Rollout bar produces rows of token cells with ✓ / ✗ verdicts, a dashed Low-quality Rollouts block is replaced when acc(x′) > acc(x) by a dashed High-quality Rollouts block from the Re-rollout bar (Calibrating Rollouts), and both lead to verdict cards that merge through a bracket into a tall RL bar, with the final group 𝒢i labelled below. Every box, arrow, label and math term is a native shape you can recolor, resize and relabel for your own slides, posters or papers.
File format
The download contains .pptx (1 slide, 12.82 × 6.52 in; opens in PowerPoint, Keynote, WPS and LibreOffice Impress; Google Slides works too, but complex freeforms may shift slightly), .svg and .drawio. Includes preview.png (full resolution) and a 1200 px web preview.
Shapes and editing
The file has 292 native editable shapes: 222 auto-shapes, 37 text boxes, 10 freeforms, 23 lines/connectors. No embedded images. Everything is vector. You can edit all text and math (xi, x′i, γ, acc(x′) > acc(x), 𝒢i), every prompt-token and rollout-token cell, the hatched interference cells, the interference-score bars, axes and dashed threshold, the ✓ / ✗ verdict badges, the dashed rollout blocks, the Rollout / Re-rollout / RL bars, the verdict cards, the bracket, the block arrows, the legend and both dashed stage panels.
Embedded images and fonts
This file has no embedded images; everything is vector. Body text uses Calibri and formulas use Times New Roman. Calibri ships with Microsoft Office and Times New Roman is preinstalled on Windows and macOS (LibreOffice substitutes the metric-compatible Carlito and Liberation Serif). The script 𝒢 uses Cambria Math, which ships with Microsoft Office; other apps substitute a math font. Sub- and superscripts are real formatted text runs, not equation objects. The SVG keeps text as text; the draw.io file is a shape-by-shape conversion and may need small nudges. This is a faithful redraw for reuse, not the authors' original source file.
Source & license Adapted from Figure 3 of Less Noise, More Voice: Reinforcement Learning for Reasoning via Instruction Purification by Yiju Guo, Tianyi Hu, Zexu Sun, Yankai Lin (arXiv:2601.21244, https://arxiv.org/abs/2601.21244). Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Redrawn as editable shapes; changes made by 叫我Alonzo就好了/alonzolab. Not endorsed by the original authors.
Questions or feedback? Email [email protected]