pith. sign in

over-trained

Then, due to the PL condition 1, Smoothness Assumption 2, we know there exists a w∗ γ (the projection ofw ∗ toS ∗ γ(α)) satisfies ∥w∗ −w ∗ γ∥ ≤ 1 µ ∥∇wLtrain(α,w ∗ γ)∥ ≤ 1 γµ ∥∇wLval(w∗ γ) +γ∇ wLtrain(w∗ γ)∥+∥∇ wLval(w∗ γ)∥ = ∥∇wLval(w∗ · 2048 · arXiv 7289.7587

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it
read on arXiv browse 1 citing papers

fields

cs.LG 1

years

2026 1

verdicts

UNVERDICTED 1

representative citing papers

TANDEM: Bi-Level Data Mixture Optimization with Twin Networks

cs.LG · 2026-06-03 · unverdicted · novelty 5.0

TANDEM solves bi-level data mixture optimization for LLMs via twin proxy and reference networks that measure domain efficacy by model difference and up-weight beneficial domains, with claimed theoretical guarantees and gains in restricted-data and SFT settings.

citing papers explorer

Showing 1 of 1 citing paper.

  • TANDEM: Bi-Level Data Mixture Optimization with Twin Networks cs.LG · 2026-06-03 · unverdicted · none · ref 48

    TANDEM solves bi-level data mixture optimization for LLMs via twin proxy and reference networks that measure domain efficacy by model difference and up-weight beneficial domains, with claimed theoretical guarantees and gains in restricted-data and SFT settings.