AI Paper Atlas

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

DUET: Optimizing LLM Training Data Mixtures via Noisy Feedback from Unseen, Downstream Evaluation Tasks

Zhiliang Chen, Gregory Kang Ruey Lau, Chuan-Sheng Foo, Bryan Kian Hsiang Low

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
optimize data mixing ratio with BO from task feedback (no data selection) (ablation)DUET · Unseen tasks (Llama-3-8b-Instruct, Fig. 4d)task accuracyImproved
add influence-function data selection to BO (ablation)DUET · Unseen tasks (Llama-3-8b-Instruct, Fig. 4d)task accuracyImproved
use IF-weighted sampling instead of LESS, remove-harmful or log-det selection (ablation)DUET · Unseen tasks (Llama-3-8b-Instruct, Fig. 4d)task accuracyImproved
increase sampling size k in inner loop (ablation)DUET · Unseen tasks (Llama-3-8b-Instruct, Fig. 4d)task accuracyImproved
use DUET instead of DoReMi and LESSDUET · In-domain and out-of-domain unseen taskstask accuracyImproved
use IF-weighted sampling instead of uniform random sampling for inner estimator (ablation)DUET · Two-domain mixture, ratio [0.5,0.5]estimator bias and varianceImproved
use deterministic removal of lowest-IF points instead of IF-weighted sampling (ablation)DUET · Unseen taskstask accuracyHurt

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