AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Influence-Preserving Proxies for Gradient-Based Data Selection in LLM FineTuning

Sirui Chen, Yunzhe Qi, Mengting Ai, Yifan Sun, Ruizhong Qiu, Jiaru Zou, Jingrui He

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression

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
build proxy with IProX instead of using off-the-shelf smaller modelQwen3-4B target · Dolly, avg of MMLU/BBH/TyDiQAaverage score: 63.81 (Qwen3-1.7B) -> 64.18 (IProX 1.5B)Improved
build proxy with IProX (rho=0.3) instead of Llama3.2-1B off-the-shelfLlama3.2-3B target · Dolly, avg of MMLU/BBH/TyDiQAaverage score: 47.35 -> 48.97Improved
replace standard SVD with influence-preserving SVD (IPSVD) (ablation)IProX · Llama3.2-3B, TyDiQA rho=0.7exact match: 32.73 -> 39.04Improved
add gradient-alignment stage (ablation)IProX · Llama3.2-3B, TyDiQA rho=0.7exact match: 35.32 -> 39.79Improved
add KL anchoring of logits (ablation)IProX · Llama3.2-3B, TyDiQA rho=0.7exact match: 37.66 -> 39.79Improved
use IProX instead of Layer Extraction and Influence Scorer baselinesLlama3.2-3B, Gemma3-4B · Dolly, avgaverage score: +2.81 and +1.99 average over strongest baselineImproved
use IProX with Influence Function instead of TracInLlama3.2-3B, Gemma3-4B · Dolly, avgaverage score: 46.81 -> 49.56 (Llama3.2, rho=0.3)Improved
increase proxy size (lower sparsity rho) (ablation)IProX · four model familiesaverage score: Llama3.2-3B: 47.54 (rho=0.7) -> 48.97 (rho=0.3)Improved
increase probe set size beyond 3x default (ablation)IProX · Llama3.2-3B rho=0.7average score: 47.85 (3x) vs 47.55 (5x)No clear effect
reduce probe set diversity via 30% SMOTE redundancy (ablation)IProX · Llama3.2-3B rho=0.7average score: 47.54 -> 46.48Hurt
set KL coefficient to 0.1 vs 0.5/0.01/0.001 (ablation)IProX · Llama3.2-3B rho=0.7average score: 46.90/46.59/46.36 -> 47.54Improved
compress to extreme sparsity rho=0.9IProX vs Layer Extraction · Llama3.2-3B, Qwen2-7Baverage score: 44.64 -> 46.67 (Llama3.2); 57.38 -> 59.64 (Qwen2-7B)Improved

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