Soft Preference Optimization: Aligning Language Models to Expert Distributions

• URL: http://arxiv.org/abs/2405.00747v3
• Date: Mon, 27 May 2024 19:59:00 GMT
• Title: Soft Preference Optimization: Aligning Language Models to Expert Distributions
• Authors: Arsalan Sharifnassab, Sina Ghiassian, Saber Salehkaleybar, Surya Kanoria, Dale Schuurmans,
• Abstract summary: SPO is a method for aligning generative models, such as Large Language Models (LLMs), with human preferences. SPO integrates preference loss with a regularization term across the model's entire output distribution. We showcase SPO's methodology, its theoretical foundation, and its comparative advantages in simplicity, computational efficiency, and alignment precision.
• Score: 40.84391304598521
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We propose Soft Preference Optimization (SPO), a method for aligning generative models, such as Large Language Models (LLMs), with human preferences, without the need for a reward model. SPO optimizes model outputs directly over a preference dataset through a natural loss function that integrates preference loss with a regularization term across the model's entire output distribution rather than limiting it to the preference dataset. Although SPO does not require the assumption of an existing underlying reward model, we demonstrate that, under the Bradley-Terry (BT) model assumption, it converges to a softmax of scaled rewards, with the distribution's "softness" adjustable via the softmax exponent, an algorithm parameter. We showcase SPO's methodology, its theoretical foundation, and its comparative advantages in simplicity, computational efficiency, and alignment precision.

Related papers

• Step-level Value Preference Optimization for Mathematical Reasoning [6.318873143509028]
  We introduce a novel algorithm called Step-level Value Preference Optimization (SVPO) Our method achieves state-of-the-art performance on both in-domain and out-of-domain mathematical reasoning benchmarks.
  arXiv  Detail & Related papers  (2024-06-16T09:06:17Z)
• Robust Preference Optimization through Reward Model Distillation [68.65844394615702]
  Language model (LM) post-training involves maximizing a reward function that is derived from preference annotations. DPO is a popular offline alignment method that trains a policy directly on preference data without the need to train a reward model or apply reinforcement learning. We analyze this phenomenon and propose distillation to get a better proxy for the true preference distribution over generation pairs.
  arXiv  Detail & Related papers  (2024-05-29T17:39:48Z)
• Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer [52.09480867526656]
  We identify the source of misalignment as a form of distributional shift and uncertainty in learning human preferences. To mitigate overoptimization, we first propose a theoretical algorithm that chooses the best policy for an adversarially chosen reward model. Using the equivalence between reward models and the corresponding optimal policy, the algorithm features a simple objective that combines a preference optimization loss and a supervised learning loss.
  arXiv  Detail & Related papers  (2024-05-26T05:38:50Z)
• Self-Supervised Dataset Distillation for Transfer Learning [77.4714995131992]
  We propose a novel problem of distilling an unlabeled dataset into a set of small synthetic samples for efficient self-supervised learning (SSL) We first prove that a gradient of synthetic samples with respect to a SSL objective in naive bilevel optimization is textitbiased due to randomness originating from data augmentations or masking. We empirically validate the effectiveness of our method on various applications involving transfer learning.
  arXiv  Detail & Related papers  (2023-10-10T10:48:52Z)
• Modeling the Second Player in Distributionally Robust Optimization [90.25995710696425]
  We argue for the use of neural generative models to characterize the worst-case distribution. This approach poses a number of implementation and optimization challenges. We find that the proposed approach yields models that are more robust than comparable baselines.
  arXiv  Detail & Related papers  (2021-03-18T14:26:26Z)
• Autoregressive Score Matching [113.4502004812927]
  We propose autoregressive conditional score models (AR-CSM) where we parameterize the joint distribution in terms of the derivatives of univariable log-conditionals (scores) For AR-CSM models, this divergence between data and model distributions can be computed and optimized efficiently, requiring no expensive sampling or adversarial training. We show with extensive experimental results that it can be applied to density estimation on synthetic data, image generation, image denoising, and training latent variable models with implicit encoders.
  arXiv  Detail & Related papers  (2020-10-24T07:01:24Z)
• Maximum Entropy Model Rollouts: Fast Model Based Policy Optimization without Compounding Errors [10.906666680425754]
  We propose a Dyna-style model-based reinforcement learning algorithm, which we called Maximum Entropy Model Rollouts (MEMR) To eliminate the compounding errors, we only use our model to generate single-step rollouts.
  arXiv  Detail & Related papers  (2020-06-08T21:38:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.