Meta-Learning Objectives for Preference Optimization

• URL: http://arxiv.org/abs/2411.06568v2
• Date: Tue, 04 Feb 2025 22:02:01 GMT
• Title: Meta-Learning Objectives for Preference Optimization
• Authors: Carlo Alfano, Silvia Sapora, Jakob Nicolaus Foerster, Patrick Rebeschini, Yee Whye Teh,
• Abstract summary: We show that it is possible to gain insights on the efficacy of preference optimization algorithms on simpler benchmarks.<n>We propose a novel family of PO algorithms based on mirror descent, which we call Mirror Preference Optimization (MPO)
• Score: 39.15940594751445
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Evaluating preference optimization (PO) algorithms on LLM alignment is a challenging task that presents prohibitive costs, noise, and several variables like model size and hyper-parameters. In this work, we show that it is possible to gain insights on the efficacy of PO algorithm on much simpler benchmarks. We design a diagnostic suite of MuJoCo tasks and datasets, which we use to systematically evaluate PO algorithms, establishing a more controlled and cheaper benchmark. We then propose a novel family of PO algorithms based on mirror descent, which we call Mirror Preference Optimization (MPO). Through evolutionary strategies, we search this class to discover algorithms specialized to specific properties of preference datasets, such as mixed-quality or noisy data. We demonstrate that our discovered PO algorithms outperform all known algorithms in the targeted MuJoCo settings. Finally, based on the insights gained from our MuJoCo experiments, we design a novel PO algorithm that significantly outperforms existing baselines in an LLM alignment task.

Related papers

• Combinatorial Optimization for All: Using LLMs to Aid Non-Experts in Improving Optimization Algorithms [0.9668407688201361]
  Large Language Models (LLMs) have shown notable potential in code generation for optimization algorithms. This paper examines how LLMs, rather than creating algorithms from scratch, can improve existing ones without the need for specialized expertise.
  arXiv  Detail & Related papers  (2025-03-14T00:26:00Z)
• Visualising Policy-Reward Interplay to Inform Zeroth-Order Preference Optimisation of Large Language Models [0.36326779753373206]
  Zeroth-Order (ZO) optimisation, using function evaluations instead of gradients, reduces memory usage but suffers from slow convergence in high-dimensional models. We introduce ZOPrO, a novel ZO algorithm designed for Preference optimisation in LLMs. We demonstrate that our method consistently enhances reward signals while achieving convergence times comparable to first-order methods.
  arXiv  Detail & Related papers  (2025-03-05T12:49:48Z)
• Reward-Augmented Data Enhances Direct Preference Alignment of LLMs [56.24431208419858]
  We introduce reward-conditioned Large Language Models (LLMs) that learn from the entire spectrum of response quality within the dataset. We propose an effective yet simple data relabeling method that conditions the preference pairs on quality scores to construct a reward-augmented dataset.
  arXiv  Detail & Related papers  (2024-10-10T16:01:51Z)
• EVOLvE: Evaluating and Optimizing LLMs For Exploration [76.66831821738927]
  Large language models (LLMs) remain under-studied in scenarios requiring optimal decision-making under uncertainty. We measure LLMs' (in)ability to make optimal decisions in bandits, a state-less reinforcement learning setting relevant to many applications. Motivated by the existence of optimal exploration algorithms, we propose efficient ways to integrate this algorithmic knowledge into LLMs.
  arXiv  Detail & Related papers  (2024-10-08T17:54:03Z)
• Self-supervised Preference Optimization: Enhance Your Language Model with Preference Degree Awareness [27.43137305486112]
  We propose a novel Self-supervised Preference Optimization (SPO) framework, which constructs a self-supervised preference degree loss combined with the alignment loss. The results demonstrate that SPO can be seamlessly integrated with existing preference optimization methods to achieve state-of-the-art performance.
  arXiv  Detail & Related papers  (2024-09-26T12:37:26Z)
• Revisiting BPR: A Replicability Study of a Common Recommender System Baseline [78.00363373925758]
  We study the features of the BPR model, indicating their impact on its performance, and investigate open-source BPR implementations. Our analysis reveals inconsistencies between these implementations and the original BPR paper, leading to a significant decrease in performance of up to 50% for specific implementations. We show that the BPR model can achieve performance levels close to state-of-the-art methods on the top-n recommendation tasks and even outperform them on specific datasets.
  arXiv  Detail & Related papers  (2024-09-21T18:39:53Z)
• Bridging and Modeling Correlations in Pairwise Data for Direct Preference Optimization [75.1240295759264]
  We propose an effective framework for Bridging and Modeling Correlations in pairwise data, named BMC. We increase the consistency and informativeness of the pairwise preference signals through targeted modifications. We identify that DPO alone is insufficient to model these correlations and capture nuanced variations.
  arXiv  Detail & Related papers  (2024-08-14T11:29:47Z)
• Discovering Preference Optimization Algorithms with and for Large Language Models [50.843710797024805]
  offline preference optimization is a key method for enhancing and controlling the quality of Large Language Model (LLM) outputs. We perform objective discovery to automatically discover new state-of-the-art preference optimization algorithms without (expert) human intervention. Experiments demonstrate the state-of-the-art performance of DiscoPOP, a novel algorithm that adaptively blends logistic and exponential losses.
  arXiv  Detail & Related papers  (2024-06-12T16:58:41Z)
• 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)
• Beyond Single-Model Views for Deep Learning: Optimization versus Generalizability of Stochastic Optimization Algorithms [13.134564730161983]
  This paper adopts a novel approach to deep learning optimization, focusing on gradient descent (SGD) and its variants. We show that SGD and its variants demonstrate performance on par with flat-minimas like SAM, albeit with half the gradient evaluations. Our study uncovers several key findings regarding the relationship between training loss and hold-out accuracy, as well as the comparable performance of SGD and noise-enabled variants.
  arXiv  Detail & Related papers  (2024-03-01T14:55:22Z)
• Unleashing the Potential of Large Language Models as Prompt Optimizers: Analogical Analysis with Gradient-based Model Optimizers [108.72225067368592]
  We propose a novel perspective to investigate the design of large language models (LLMs)-based prompts. We identify two pivotal factors in model parameter learning: update direction and update method. We develop a capable Gradient-inspired Prompt-based GPO.
  arXiv  Detail & Related papers  (2024-02-27T15:05:32Z)
• Efficient Non-Parametric Optimizer Search for Diverse Tasks [93.64739408827604]
  We present the first efficient scalable and general framework that can directly search on the tasks of interest. Inspired by the innate tree structure of the underlying math expressions, we re-arrange the spaces into a super-tree. We adopt an adaptation of the Monte Carlo method to tree search, equipped with rejection sampling and equivalent- form detection.
  arXiv  Detail & Related papers  (2022-09-27T17:51:31Z)
• Explainable Landscape-Aware Optimization Performance Prediction [0.0]
  We are investigating explainable landscape-aware regression models. The contribution of each landscape feature to the prediction of the optimization algorithm performance is estimated on a global and local level. The results show a proof of concept that different set of features are important for different problem instances.
  arXiv  Detail & Related papers  (2021-10-22T07:46:33Z)
• Meta Learning Black-Box Population-Based Optimizers [0.0]
  We propose the use of meta-learning to infer population-based blackbox generalizations. We show that the meta-loss function encourages a learned algorithm to alter its search behavior so that it can easily fit into a new context.
  arXiv  Detail & Related papers  (2021-03-05T08:13:25Z)
• Mixed Strategies for Robust Optimization of Unknown Objectives [93.8672371143881]
  We consider robust optimization problems, where the goal is to optimize an unknown objective function against the worst-case realization of an uncertain parameter. We design a novel sample-efficient algorithm GP-MRO, which sequentially learns about the unknown objective from noisy point evaluations. GP-MRO seeks to discover a robust and randomized mixed strategy, that maximizes the worst-case expected objective value.
  arXiv  Detail & Related papers  (2020-02-28T09:28:17Z)

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.