AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Inverse Reinforcement Learning with Dynamic Reward Scaling for LLM Alignment

Ruoxi Cheng, Hao-Xuan Ma, Weixin Wang, Ranjie Duan, Jiexi Liu, Xiaoshuang Jia, Simeng Qin, Xiaochun Cao, Yang Liu, Xiaojun Jia

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 22 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add dynamic hardness scaling on top of IRL reward models (ablation)GRPO with IRL reward (Llama-3.1-8B) · StrongRejectgoodness score: 0.8917 -> 0.9361Improved
add dynamic hardness scaling on top of IRL reward models (ablation)GRPO with IRL reward (Qwen-2-7B) · StrongRejectgoodness score: 0.5960 -> 0.8798Improved
replace static reward with IRL shadow reward models (ablation)GRPO (Llama-3.1-8B) · WildChatrefusal rate: 55.61% -> 67.54%Improved
replace static reward with IRL shadow reward models (ablation)GRPO (Qwen-2-7B) · StrongRejectgoodness score: 0.5155 -> 0.5960Improved
use product of data and model hardness instead of equal-weight sum (ablation)DR-IRL (Llama-3.1-8B) · StrongRejectgoodness score: 0.9139 -> 0.9361Improved
use product instead of tuned weighted sum (ablation)DR-IRL (Qwen-2-7B) · WildChatrefusal rate: 78.13% -> 81.53%Improved
use seven per-category reward models instead of a single reward model (ablation)DR-IRL (Llama) · StrongRejectgoodness score: 0.9182 -> 0.9361Improved
use seven per-category reward models instead of a single reward model (ablation)DR-IRL (Llama) · Stereotyperefusal rate: 96.25% -> 98.87%Improved
include both data and model hardness coefficients (vs none) (ablation)DR-IRL (Llama-3.1-8B) · StrongRejectgoodness score: roughly 4 percentage pointsImproved
replace CoT with CoD prompting for data generation (ablation)Qwen-2-7B · CoD vs CoT accuracyaccuracy: 91.9% -> 94.1%Improved
replace CoT with CoD prompting for data generation (ablation)Llama-3.1-8B · CoD vs CoT accuracyaccuracy: 86.2% -> 87.9%Improved
apply hardness coefficient to DPODPO (Llama-3.1-8B) · StrongRejectgoodness score: 0.5054 -> 0.5826Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy