research/Paper

5,072 records · yaml · accepts undeclared fields

Fields

Rules

• paperArxivAttestationRule
• paperAuthorLinksRule
• paperUrlRule
• paperTrimmedStringsRule

Defects

Checked 200 of 5,072 records (the per-store cap is 200), so this is a sample, not a whole-store verdict.

No defects in the 200 record(s) checked.

Example record

/Users/eshao/.config/mnt/data/research-reader/papers/abas-2026-tur-dpo.yaml

slug: abas-2026-tur-dpo
stub: true
renderings:
  - published: https://w.sh/paper/abas-2026-tur-dpo
    kind: paper page
provenance:
  via: ICML 2026 Agent Reproduction Challenge index
  caveats: Stub minted from the ICML 2026 challenge index for Repro linking; title, authors, year, arXiv id, and abstract are index values; Zotero and structured summary omitted rather than guessed.
name: 'TUR-DPO: Topology- and Uncertainty-Aware Direct Preference Optimization'
ids:
  arxiv: '2605.00224'
year: 2026
authors:
  - name: Abdulhady Abas Abdullah
    openalexId: A5038294530
    slug: openalex-a5038294530
  - name: Fatemeh Daneshfar
    openalexId: A5003861036
    slug: openalex-a5003861036
  - name: Seyedali Mirjalili
    openalexId: A5135381857
    slug: openalex-a5135381857
  - name: Mourad Oussalah
    openalexId: A5135337281
    slug: openalex-a5135337281
abstract: Aligning large language models (LLMs) with human preferences is commonly done via reinforcement learning from human feedback (RLHF) with Proximal Policy Optimization (PPO) or, more simply, via Direct Preference Optimization (DPO). While DPO is stable and RL-free, it treats preferences as flat winner vs. loser signals and is sensitive to noisy or brittle preferences arising from fragile chains of thought. We propose TUR-DPO, a topology- and uncertainty-aware variant of DPO that rewards how answers are derived, not only what they say, by eliciting lightweight reasoning topologies and combining semantic faithfulness, utility, and topology quality into a calibrated uncertainty signal. A small learnable reward is factorized over these signals and incorporated into an uncertainty-weighted DPO objective that remains RL-free and relies only on a fixed or moving reference policy. Empirically, across open 7-8B models and benchmarks spanning mathematical reasoning, factual question answering, summarization, and helpful/harmless dialogue, TUR-DPO improves judge win-rates, faithfulness, and calibration relative to DPO while preserving training simplicity and avoiding online rollouts. We further observe consistent gains in multimodal and long-context settings, and show that TUR-DPO matches or exceeds PPO on reasoning-centric tasks while maintaining operational simplicity.
pdf:
  via: query-arxiv
  id: '2605.00224'
sessions: []
venue:
  id: venue/icml
  track: main
  presentation: poster
  provenance:
    via: conf-icml26 program scrape
    checkedOn: '2026-08-11'

Where it lives