AI Paper Atlas

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

NADS: Navigator-Guided Data Selection for Mitigating Catastrophic Forgetting in Fine-Tuning

JIANHAO ZHANG, Ou Wu, Yi Du

NeurIPS 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes NADS, a navigator-guided data selection method that picks synthetic samples to mitigate catastrophic forgetting when fine-tuning LLMs on a new task without original pretraining data.

Problem

Catastrophic forgetting during single-stage fine-tuning of LLMs on a new task when original pretraining data are unavailable; synthetic candidate pools are noisy and redundant, and existing selection methods rely on indirect proxy signals poorly aligned with actual training benefit.

Approach

Selects synthetic training data using a navigator signal intended to directly track actual training benefit rather than indirect proxies, to reduce forgetting while learning the new task.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use navigator-guided constraint selection with KD instead of the strongest replay baselineLLaMA3.2-3B on Magicoder · Magicoder overalloverall accuracy: 42.01 -> 43.46Improved
remove KL-discrepancy factor from the utility (gate-only utility) (ablation)NADS · Magicoder, LLaMA3.2-3Bgeneral capability: 42.52 -> 37.64Hurt
add diversity-aware selection term (ablation)NADS · Magicoder, LLaMA3.2-3Bgeneral capability: 40.21 -> 42.52Improved
add the pretrained-confidence entropy gate to the utility (ablation)NADS · Magicoder, LLaMA3.2-3Boverall accuracy: 42.22 -> 43.46Improved
select constraints by navigator utility instead of random, perplexity or standard DPP selection (ablation)NADS · Magicoder, LLaMA3.2-3Boverall accuracy: 39.94 / 40.39 / 39.66 -> 43.46Improved
use random constraints under a fixed KD objective (ablation)baseline methods with KD · Magicoder, LLaMA3.2general capabilityNo clear effect
use navigator-selected constraints under a fixed KD objective (ablation)baseline methods with KD · Magicoder, LLaMA3.2general capabilityImproved
train the navigator unconstrained on the new taskLLaMA3.2-3B navigator · MagicoderHumanEval vs MMLU: HumanEval 28.16 -> 45.50; MMLU 56.92 -> 50.60Mixed
use a navigator checkpoint from a middle training stage (M_nav-2) rather than early or late (ablation)NADS · Magicoder, LLaMA3.2-3Btarget and general trade-offImproved
use forward KL instead of JS divergence as utility signal (ablation)NADS · Magicoder, LLaMA3.2-3BperformanceImproved
increase distillation strength lambda (ablation)NADS · Magicodergeneral vs targetMixed
use the NADS iterative optimizer instead of Fast Greedy or Lazy-Fast Greedy (ablation)NADS data selection · Magicoder candidate poolselection timeImproved

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