Search

Reinforcement learning from human feedback (RLHF) is a promising solution to align large language models (LLMs) more closely with human values. Off-policy preference optimization, where the preference data is obtained from other models, is widely adopted due to its cost efficiency and scalability. However, off-policy preference optimization often suffers from a distributional gap between the policy used for data collection and the target policy, leading to suboptimal optimization. In this paper, we propose a novel strategy to mitigate this problem by simulating on-policy learning with off-policy preference data. Our Weighted Preference Optimization (WPO) method adapts off-policy data to resemble on-policy data more closely by reweighting preference pairs according to their probability under the current policy. This method not only addresses the distributional gap problem but also enhances the optimization process without incurring additional costs. We validate our method on instruction following benchmarks including Alpaca Eval 2 and MT-bench. WPO not only outperforms Direct Preference Optimization (DPO) by up to 5.6% on Alpaca Eval 2 but also establishes a remarkable length-controlled winning rate against GPT-4-turbo of 48.6% based on Llama-3-8B-Instruct, making it the strongest 8B model on the leaderboard. We will release the code and models at https://github.com/wzhouad/WPO.

The cosine router in sparse Mixture of Experts (MoE) has recently emerged as an attractive alternative to the conventional linear router. Indeed, the cosine router demonstrates favorable performance in image and language tasks and exhibits better ability to mitigate the representation collapse issue, which often leads to parameter redundancy and limited representation potentials. Despite its empirical success, a comprehensive analysis of the cosine router in sparse MoE has been lacking. Considering the least square estimation of the cosine routing sparse MoE, we demonstrate that due to the intrinsic interaction of the model parameters in the cosine router via some partial differential equations, regardless of the structures of the experts, the estimation rates of experts and model parameters can be as slow as $\mathcal{O}(1/\log^{\tau}(n))$ where $\tau > 0$ is some constant and $n$ is the sample size. Surprisingly, these pessimistic non-polynomial convergence rates can be circumvented by the widely used technique in practice to stabilize the cosine router -- simply adding noises to the $\mathbb{L}_{2}$ norms in the cosine router, which we refer to as \textit{perturbed cosine router}. Under the strongly identifiable settings of the expert functions, we prove that the estimation rates for both the experts and model parameters under the perturbed cosine routing sparse MoE are significantly improved to polynomial rates. Finally, we conduct extensive simulation studies in both synthetic and real data settings to empirically validate our theoretical results., Comment: 44 pages, 2 figures

Auto-regressive generation models achieve competitive performance across many different NLP tasks such as summarization, question answering, and classifications. However, they are also known for being slow in inference, which makes them challenging to deploy in real-time applications. We propose a switchable decision to accelerate inference by dynamically assigning computation resources for each data instance. Automatically making decisions on where to skip and how to balance quality and computation cost with constrained optimization, our dynamic neural generation networks enforce the efficient inference path and determine the optimized trade-off. Experiments across question answering, summarization, and classification benchmarks show that our method benefits from less computation cost during inference while keeping the same accuracy. Extensive experiments and ablation studies demonstrate that our method can be general, effective, and beneficial for many NLP tasks., Comment: Accepted to ICML 2024

Recent works have demonstrated success in controlling sentence attributes ($e.g.$, sentiment) and structure ($e.g.$, syntactic structure) based on the diffusion language model. A key component that drives theimpressive performance for generating high-quality samples from noise is iteratively denoise for thousands of steps. While beneficial, the complexity of starting from the noise and the learning steps has limited its implementation to many NLP real-world applications. This paper proposes Language Rectified Flow ({\ours}). Our method is based on the reformulation of the standard probabilistic flow models. Language rectified flow learns (neural) ordinary differential equation models to transport between the source distribution and the target distribution, hence providing a unified and effective solution to generative modeling and domain transfer. From the source distribution, our language rectified flow yields fast simulation and effectively decreases the inference time. Experiments on three challenging fine-grained control tasks and multiple high-quality text editing show that our method consistently outperforms its baselines. Extensive experiments and ablation studies demonstrate that our method can be general, effective, and beneficial for many NLP tasks., Comment: Accepted to NAACL 2024

Slicing distribution selection has been used as an effective technique to improve the performance of parameter estimators based on minimizing sliced Wasserstein distance in applications. Previous works either utilize expensive optimization to select the slicing distribution or use slicing distributions that require expensive sampling methods. In this work, we propose an optimization-free slicing distribution that provides a fast sampling for the Monte Carlo estimation of expectation. In particular, we introduce the random-path projecting direction (RPD) which is constructed by leveraging the normalized difference between two random vectors following the two input measures. From the RPD, we derive the random-path slicing distribution (RPSD) and two variants of sliced Wasserstein, i.e., the Random-Path Projection Sliced Wasserstein (RPSW) and the Importance Weighted Random-Path Projection Sliced Wasserstein (IWRPSW). We then discuss the topological, statistical, and computational properties of RPSW and IWRPSW. Finally, we showcase the favorable performance of RPSW and IWRPSW in gradient flow and the training of denoising diffusion generative models on images., Comment: Accepted to ICML 2024, 21 pages, 5 figures, 2 tables

Aligning language models (LMs) with preferences is an important problem in natural language generation. A key challenge is that preferences are typically provided at the *sequence level* while LM training and generation both occur at the *token level*. There is, therefore, a *granularity mismatch* between the preference and the LM training losses, which may complicate the learning problem. In this paper, we address this issue by developing an alternate training process, where we iterate between grounding the sequence-level preference into token-level training guidance, and improving the LM with the learned guidance. For guidance learning, we design a framework that extends the pairwise-preference learning in imitation learning to both variable-length LM generation and the utilization of the preference among multiple generations. For LM training, based on the amount of supervised data, we present two *minimalist* learning objectives that utilize the learned guidance. In experiments, our method performs competitively on two distinct representative LM tasks -- discrete-prompt generation and text summarization., Comment: 37th Conference on Neural Information Processing Systems (NeurIPS 2023)

AI tasks encompass a wide range of domains and fields. While numerous AI models have been designed for specific tasks and applications, they often require considerable human efforts in finding the right model architecture, optimization algorithm, and hyperparameters. Recent advances in large language models (LLMs) like ChatGPT show remarkable capabilities in various aspects of reasoning, comprehension, and interaction. Consequently, we propose developing task-oriented prompts and automatically utilizing LLMs to automate the training pipeline. To implement this concept, we present the AutoML-GPT, which employs GPT as the bridge to diverse AI models and dynamically trains models with optimized hyperparameters. AutoML-GPT dynamically takes user requests from the model and data cards and composes the corresponding prompt paragraph. Ultimately, with this prompt paragraph, AutoML-GPT will automatically conduct the experiments from data processing to model architecture, hyperparameter tuning, and predicted training log. By leveraging {\ours}'s robust language capabilities and the available AI models, AutoML-GPT can tackle numerous intricate AI tasks across various tasks and datasets. This approach achieves remarkable results in computer vision, natural language processing, and other challenging areas. Extensive experiments and ablation studies demonstrate that our method can be general, effective, and beneficial for many AI tasks.

Through prompting, large-scale pre-trained models have become more expressive and powerful, gaining significant attention in recent years. Though these big models have zero-shot capabilities, in general, labeled data are still required to adapt them to downstream tasks. To overcome this critical limitation, we propose an unsupervised fine-tuning framework to directly fine-tune the model or prompt on the unlabeled target data. We demonstrate how to apply our method to both language-augmented vision and masked-language models by aligning the discrete distributions extracted from the prompts and target data. To verify our approach's applicability, we conduct extensive experiments on image classification, sentiment analysis, and natural language inference tasks. Across 13 image-related tasks and 15 language-related ones, the proposed approach achieves consistent improvements over the baselines., Comment: ICML 2023; PyTorch code is available at https://github.com/korawat-tanwisuth/POUF

When learning task-oriented dialogue (ToD) agents, reinforcement learning (RL) techniques can naturally be utilized to train dialogue strategies to achieve user-specific goals. Prior works mainly focus on adopting advanced RL techniques to train the ToD agents, while the design of the reward function is not well studied. This paper aims at answering the question of how to efficiently learn and leverage a reward function for training end-to-end (E2E) ToD agents. Specifically, we introduce two generalized objectives for reward-function learning, inspired by the classical learning-to-rank literature. Further, we utilize the learned reward function to guide the training of the E2E ToD agent. With the proposed techniques, we achieve competitive results on the E2E response-generation task on the Multiwoz 2.0 dataset. Source code and checkpoints are publicly released at https://github.com/Shentao-YANG/Fantastic_Reward_ICLR2023., Comment: Published as a conference paper at ICLR 2023

Unsupervised clustering under domain shift (UCDS) studies how to transfer the knowledge from abundant unlabeled data from multiple source domains to learn the representation of the unlabeled data in a target domain. In this paper, we introduce Prototype-oriented Clustering with Distillation (PCD) to not only improve the performance and applicability of existing methods for UCDS, but also address the concerns on protecting the privacy of both the data and model of the source domains. PCD first constructs a source clustering model by aligning the distributions of prototypes and data. It then distills the knowledge to the target model through cluster labels provided by the source model while simultaneously clustering the target data. Finally, it refines the target model on the target domain data without guidance from the source model. Experiments across multiple benchmarks show the effectiveness and generalizability of our source-private clustering method.

Retriever-reader models achieve competitive performance across many different NLP tasks such as open question answering and dialogue conversations. In this work, we notice these models easily overfit the top-rank retrieval passages and standard training fails to reason over the entire retrieval passages. We introduce a learnable passage mask mechanism which desensitizes the impact from the top-rank retrieval passages and prevents the model from overfitting. Controlling the gradient variance with fewer mask candidates and selecting the mask candidates with one-shot bi-level optimization, our learnable regularization strategy enforces the answer generation to focus on the entire retrieval passages. Experiments on different tasks across open question answering, dialogue conversation, and fact verification show that our method consistently outperforms its baselines. Extensive experiments and ablation studies demonstrate that our method can be general, effective, and beneficial for many NLP tasks., Comment: EMNLP 2022

In offline model-based reinforcement learning (offline MBRL), we learn a dynamic model from historically collected data, and subsequently utilize the learned model and fixed datasets for policy learning, without further interacting with the environment. Offline MBRL algorithms can improve the efficiency and stability of policy learning over the model-free algorithms. However, in most of the existing offline MBRL algorithms, the learning objectives for the dynamic models and the policies are isolated from each other. Such an objective mismatch may lead to inferior performance of the learned agents. In this paper, we address this issue by developing an iterative offline MBRL framework, where we maximize a lower bound of the true expected return, by alternating between dynamic-model training and policy learning. With the proposed unified model-policy learning framework, we achieve competitive performance on a wide range of continuous-control offline reinforcement learning datasets. Source code is publicly released., Comment: 36th Conference on Neural Information Processing Systems (NeurIPS 2022)

Offline reinforcement learning (RL) extends the paradigm of classical RL algorithms to purely learning from static datasets, without interacting with the underlying environment during the learning process. A key challenge of offline RL is the instability of policy training, caused by the mismatch between the distribution of the offline data and the undiscounted stationary state-action distribution of the learned policy. To avoid the detrimental impact of distribution mismatch, we regularize the undiscounted stationary distribution of the current policy towards the offline data during the policy optimization process. Further, we train a dynamics model to both implement this regularization and better estimate the stationary distribution of the current policy, reducing the error induced by distribution mismatch. On a wide range of continuous-control offline RL datasets, our method indicates competitive performance, which validates our algorithm. The code is publicly available., Comment: International Conference on Machine Learning (ICML) 2022

Active learning, which effectively collects informative unlabeled data for annotation, reduces the demand for labeled data. In this work, we propose to retrieve unlabeled samples with a local sensitivity and hardness-aware acquisition function. The proposed method generates data copies through local perturbations and selects data points whose predictive likelihoods diverge the most from their copies. We further empower our acquisition function by injecting the select-worst case perturbation. Our method achieves consistent gains over the commonly used active learning strategies in various classification tasks. Furthermore, we observe consistent improvements over the baselines on the study of prompt selection in prompt-based few-shot learning. These experiments demonstrate that our acquisition guided by local sensitivity and hardness can be effective and beneficial for many NLP tasks., Comment: NAACL 2022 (finding); Our code is publicly available at https://github.com/szhang42/allsh

Generating images from natural language instructions is an intriguing yet highly challenging task. We approach text-to-image generation by combining the power of the retrained CLIP representation with an off-the-shelf image generator (GANs), optimizing in the latent space of GAN to find images that achieve maximum CLIP score with the given input text. Compared to traditional methods that train generative models from text to image starting from scratch, the CLIP+GAN approach is training-free, zero shot and can be easily customized with different generators. However, optimizing CLIP score in the GAN space casts a highly challenging optimization problem and off-the-shelf optimizers such as Adam fail to yield satisfying results. In this work, we propose a FuseDream pipeline, which improves the CLIP+GAN approach with three key techniques: 1) an AugCLIP score which robustifies the CLIP objective by introducing random augmentation on image. 2) a novel initialization and over-parameterization strategy for optimization which allows us to efficiently navigate the non-convex landscape in GAN space. 3) a composed generation technique which, by leveraging a novel bi-level optimization formulation, can compose multiple images to extend the GAN space and overcome the data-bias. When promoted by different input text, FuseDream can generate high-quality images with varying objects, backgrounds, artistic styles, even novel counterfactual concepts that do not appear in the training data of the GAN we use. Quantitatively, the images generated by FuseDream yield top-level Inception score and FID score on MS COCO dataset, without additional architecture design or training. Our code is publicly available at \url{https://github.com/gnobitab/FuseDream}.

The neural attention mechanism has been incorporated into deep neural networks to achieve state-of-the-art performance in various domains. Most such models use multi-head self-attention which is appealing for the ability to attend to information from different perspectives. This paper introduces alignment attention that explicitly encourages self-attention to match the distributions of the key and query within each head. The resulting alignment attention networks can be optimized as an unsupervised regularization in the existing attention framework. It is simple to convert any models with self-attention, including pre-trained ones, to the proposed alignment attention. On a variety of language understanding tasks, we show the effectiveness of our method in accuracy, uncertainty estimation, generalization across domains, and robustness to adversarial attacks. We further demonstrate the general applicability of our approach on graph attention and visual question answering, showing the great potential of incorporating our alignment method into various attention-related tasks., Comment: NeurIPS 2021; Our code is publicly available at https://github.com/szhang42/alignment_attention

Existing methods for unsupervised domain adaptation often rely on minimizing some statistical distance between the source and target samples in the latent space. To avoid the sampling variability, class imbalance, and data-privacy concerns that often plague these methods, we instead provide a memory and computation-efficient probabilistic framework to extract class prototypes and align the target features with them. We demonstrate the general applicability of our method on a wide range of scenarios, including single-source, multi-source, class-imbalance, and source-private domain adaptation. Requiring no additional model parameters and having a moderate increase in computation over the source model alone, the proposed method achieves competitive performance with state-of-the-art methods., Comment: NeurIPS 2021

Training NLP systems typically assumes access to annotated data that has a single human label per example. Given imperfect labeling from annotators and inherent ambiguity of language, we hypothesize that single label is not sufficient to learn the spectrum of language interpretation. We explore new annotation distribution schemes, assigning multiple labels per example for a small subset of training examples. Introducing such multi label examples at the cost of annotating fewer examples brings clear gains on natural language inference task and entity typing task, even when we simply first train with a single label data and then fine tune with multi label examples. Extending a MixUp data augmentation framework, we propose a learning algorithm that can learn from training examples with different amount of annotation (with zero, one, or multiple labels). This algorithm efficiently combines signals from uneven training data and brings additional gains in low annotation budget and cross domain settings. Together, our method achieves consistent gains in two tasks, suggesting distributing labels unevenly among training examples can be beneficial for many NLP tasks., Comment: EMNLP 2021; Our code is publicly available at https://github.com/szhang42/Uneven_training_data

Attention-based neural networks have achieved state-of-the-art results on a wide range of tasks. Most such models use deterministic attention while stochastic attention is less explored due to the optimization difficulties or complicated model design. This paper introduces Bayesian attention belief networks, which construct a decoder network by modeling unnormalized attention weights with a hierarchy of gamma distributions, and an encoder network by stacking Weibull distributions with a deterministic-upward-stochastic-downward structure to approximate the posterior. The resulting auto-encoding networks can be optimized in a differentiable way with a variational lower bound. It is simple to convert any models with deterministic attention, including pretrained ones, to the proposed Bayesian attention belief networks. On a variety of language understanding tasks, we show that our method outperforms deterministic attention and state-of-the-art stochastic attention in accuracy, uncertainty estimation, generalization across domains, and robustness to adversarial attacks. We further demonstrate the general applicability of our method on neural machine translation and visual question answering, showing great potential of incorporating our method into various attention-related tasks., Comment: ICML 2021

We study calibration in question answering, estimating whether model correctly predicts answer for each question. Unlike prior work which mainly rely on the model's confidence score, our calibrator incorporates information about the input example (e.g., question and the evidence context). Together with data augmentation via back translation, our simple approach achieves 5-10% gains in calibration accuracy on reading comprehension benchmarks. Furthermore, we present the first calibration study in the open retrieval setting, comparing the calibration accuracy of retrieval-based span prediction models and answer generation models. Here again, our approach shows consistent gains over calibrators relying on the model confidence. Our simple and efficient calibrator can be easily adapted to many tasks and model architectures, showing robust gains in all settings., Comment: ACL 2021 (finding)

Dropout has been demonstrated as a simple and effective module to not only regularize the training process of deep neural networks, but also provide the uncertainty estimation for prediction. However, the quality of uncertainty estimation is highly dependent on the dropout probabilities. Most current models use the same dropout distributions across all data samples due to its simplicity. Despite the potential gains in the flexibility of modeling uncertainty, sample-dependent dropout, on the other hand, is less explored as it often encounters scalability issues or involves non-trivial model changes. In this paper, we propose contextual dropout with an efficient structural design as a simple and scalable sample-dependent dropout module, which can be applied to a wide range of models at the expense of only slightly increased memory and computational cost. We learn the dropout probabilities with a variational objective, compatible with both Bernoulli dropout and Gaussian dropout. We apply the contextual dropout module to various models with applications to image classification and visual question answering and demonstrate the scalability of the method with large-scale datasets, such as ImageNet and VQA 2.0. Our experimental results show that the proposed method outperforms baseline methods in terms of both accuracy and quality of uncertainty estimation.

We study estimating inherent human disagreement (annotation label distribution) in natural language inference task. Post-hoc smoothing of the predicted label distribution to match the expected label entropy is very effective. Such simple manipulation can reduce KL divergence by almost half, yet will not improve majority label prediction accuracy or learn label distributions. To this end, we introduce a small amount of examples with multiple references into training. We depart from the standard practice of collecting a single reference per each training example, and find that collecting multiple references can achieve better accuracy under the fixed annotation budget. Lastly, we provide rich analyses comparing these two methods for improving label distribution estimation.

Attention modules, as simple and effective tools, have not only enabled deep neural networks to achieve state-of-the-art results in many domains, but also enhanced their interpretability. Most current models use deterministic attention modules due to their simplicity and ease of optimization. Stochastic counterparts, on the other hand, are less popular despite their potential benefits. The main reason is that stochastic attention often introduces optimization issues or requires significant model changes. In this paper, we propose a scalable stochastic version of attention that is easy to implement and optimize. We construct simplex-constrained attention distributions by normalizing reparameterizable distributions, making the training process differentiable. We learn their parameters in a Bayesian framework where a data-dependent prior is introduced for regularization. We apply the proposed stochastic attention modules to various attention-based models, with applications to graph node classification, visual question answering, image captioning, machine translation, and language understanding. Our experiments show the proposed method brings consistent improvements over the corresponding baselines.