Pith. sign in

REVIEW 6 cited by

Aligning Language Models with Preferences through f-divergence Minimization

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2302.08215 v2 pith:RHV2BKAK submitted 2023-02-16 cs.CL cs.LGstat.ML

classification cs.CLcs.LGstat.ML
keywords distributiontargetdivergenceobjectivesalgorithmaligningapproximatedifferent
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

Aligning language models with preferences can be posed as approximating a target distribution representing some desired behavior. Existing approaches differ both in the functional form of the target distribution and the algorithm used to approximate it. For instance, Reinforcement Learning from Human Feedback (RLHF) corresponds to minimizing a reverse KL from an implicit target distribution arising from a KL penalty in the objective. On the other hand, Generative Distributional Control (GDC) has an explicit target distribution and minimizes a forward KL from it using the Distributional Policy Gradient (DPG) algorithm. In this paper, we propose a new approach, f-DPG, which allows the use of any f-divergence to approximate any target distribution that can be evaluated. f-DPG unifies both frameworks (RLHF, GDC) and the approximation methods (DPG, RL with KL penalties). We show the practical benefits of various choices of divergence objectives and demonstrate that there is no universally optimal objective but that different divergences present different alignment and diversity trade-offs. We show that Jensen-Shannon divergence strikes a good balance between these objectives, and frequently outperforms forward KL divergence by a wide margin, leading to significant improvements over prior work. These distinguishing characteristics between divergences persist as the model size increases, highlighting the importance of selecting appropriate divergence objectives.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Beyond Post-Hoc Temperature Scaling: Bilevel Optimization for LLM Calibration

    cs.LG 2026-08 conditional novelty 6.0 of 10

    CALM uses bilevel optimization to tune per-vocabulary temperature-like logit adjustments during LLM fine-tuning, and reports improved out-of-domain calibration for aligned language models.

  2. Feedback-Driven Vision-Language Alignment with Minimal Human Supervision

    cs.CV 2025-01 conditional novelty 6.0 of 10

    SVP uses self-captioning with grounding feedback and a scoring filter to adapt vision-language models with about 1,000 images, improving captioning, referring, object recall, and hallucination control.

  3. Frictional Agent Alignment Framework: Slow Down and Don't Break Things

    cs.CL 2025-05 reject novelty 5.0 of 10

    FAAF aligns an LLM with a two-part loss, one part conditioned on a detected belief-misalignment state, to generate reflection-prompting interventions, and reports higher win rates than DPO, IPO, and PPO on collaborati...

  4. RED: Unleashing Token-Level Rewards from Holistic Feedback via Reward Redistribution

    cs.CL 2024-11 conditional novelty 5.0 of 10

    RED assigns token-level rewards by taking the difference of a sequence reward model's scores on adjacent prefixes, improving RLHF training without additional reward-model training.

  5. Multi-level Value Alignment in Agentic AI Systems: Survey and Perspectives

    cs.AI 2025-06 conditional novelty 4.0 of 10

    A survey proposes a macro-meso-micro value framework for agentic AI alignment and maps applications, methods, and benchmarks onto it.

  6. Multi-Response Preference Optimization with Augmented Ranking Dataset

    cs.CL 2024-12 reject novelty 4.0 of 10

    The paper proposes Multi-DPO, a weighted multi-response preference loss, and a synthetic data augmentation pipeline, but the derivation is mathematically unsound and experimental support is weak.

Pith tools