Evolving Alignment via Asymmetric Self-Play

Current RLHF frameworks for aligning large language models (LLMs) typically assume a fixed prompt distribution, which is sub-optimal and limits the scalability of alignment and generalizability of models. To address this, we introduce a general open-ended RLHF framework that casts alignment as an asymmetric game between two players: (i) a creator that generates increasingly informative prompt distributions using the reward model, and (ii) a solver that learns to produce more preferred responses on prompts produced by the creator. This framework of Evolving Alignment via Asymmetric Self-Play (eva), results in a simple and efficient approach that can utilize any existing RLHF algorithm for scalable alignment. eva outperforms state-of-the-art methods on widely-used benchmarks, without the need of any additional human crafted prompts. Specifically, eva improves the win rate of Gemma-2-9B-it on Arena-Hard from 51.6% to 60.1% with DPO, from 55.7% to 58.9% with SPPO, from 52.3% to 60.7% with SimPO, and from 54.8% to 60.3% with ORPO, surpassing its 27B version and matching claude-3-opus. This improvement is persistent even when new human crafted prompts are introduced. Finally, we show eva is effective and robust under various ablation settings.

Vision Transformer based Random Walk for Group Re-Identification

Group re-identification (re-ID) aims to match groups with the same people under different cameras, mainly involves the challenges of group members and layout changes well. Most existing methods usually use the k-nearest neighbor algorithm to update node features to consider changes in group membership, but these methods cannot solve the problem of group layout changes. To this end, we propose a novel vision transformer based random walk framework for group re-ID. Specifically, we design a vision transformer based on a monocular depth estimation algorithm to construct a graph through the average depth value of pedestrian features to fully consider the impact of camera distance on group members relationships. In addition, we propose a random walk module to reconstruct the graph by calculating affinity scores between target and gallery images to remove pedestrians who do not belong to the current group. Experimental results show that our framework is superior to most methods.

DreamMover: Leveraging the Prior of Diffusion Models for Image Interpolation with Large Motion

We study the problem of generating intermediate images from image pairs with large motion while maintaining semantic consistency. Due to the large motion, the intermediate semantic information may be absent in input images. Existing methods either limit to small motion or focus on topologically similar objects, leading to artifacts and inconsistency in the interpolation results. To overcome this challenge, we delve into pre-trained image diffusion models for their capabilities in semantic cognition and representations, ensuring consistent expression of the absent intermediate semantic representations with the input. To this end, we propose DreamMover, a novel image interpolation framework with three main components: 1) A natural flow estimator based on the diffusion model that can implicitly reason about the semantic correspondence between two images. 2) To avoid the loss of detailed information during fusion, our key insight is to fuse information in two parts, high-level space and low-level space. 3) To enhance the consistency between the generated images and input, we propose the self-attention concatenation and replacement approach. Lastly, we present a challenging benchmark dataset InterpBench to evaluate the semantic consistency of generated results. Extensive experiments demonstrate the effectiveness of our method. Our project is available at https://dreamm0ver.github.io .

Building Math Agents with Multi-Turn Iterative Preference Learning

Recent studies have shown that large language models' (LLMs) mathematical problem-solving capabilities can be enhanced by integrating external tools, such as code interpreters, and employing multi-turn Chain-of-Thought (CoT) reasoning. While current methods focus on synthetic data generation and Supervised Fine-Tuning (SFT), this paper studies the complementary direct preference learning approach to further improve model performance. However, existing direct preference learning algorithms are originally designed for the single-turn chat task, and do not fully address the complexities of multi-turn reasoning and external tool integration required for tool-integrated mathematical reasoning tasks. To fill in this gap, we introduce a multi-turn direct preference learning framework, tailored for this context, that leverages feedback from code interpreters and optimizes trajectory-level preferences. This framework includes multi-turn DPO and multi-turn KTO as specific implementations. The effectiveness of our framework is validated through training of various language models using an augmented prompt set from the GSM8K and MATH datasets. Our results demonstrate substantial improvements: a supervised fine-tuned Gemma-1.1-it-7B model's performance increased from 77.5% to 83.9% on GSM8K and from 46.1% to 51.2% on MATH. Similarly, a Gemma-2-it-9B model improved from 84.1% to 86.3% on GSM8K and from 51.0% to 54.5% on MATH.

LAMPO: Large Language Models as Preference Machines for Few-shot Ordinal Classification

We introduce LAMPO, a novel paradigm that leverages Large Language Models (LLMs) for solving few-shot multi-class ordinal classification tasks. Unlike conventional methods, which concatenate all demonstration examples with the test instance and prompt LLMs to produce the pointwise prediction, our framework uses the LLM as a preference machine that makes a relative comparative decision between the test instance and each demonstration. A self-supervised method is then introduced to aggregate these binary comparisons into the final ordinal decision. LAMPO addresses several limitations inherent in previous methods, including context length constraints, ordering biases, and challenges associated with absolute point-wise estimation. Extensive experiments on seven public datasets demonstrate LAMPO's remarkably competitive performance across a diverse spectrum of applications (e.g., movie review analysis and hate speech detection). Notably, in certain applications, the improvement can be substantial, exceeding 20% in an absolute term. Moreover, we believe LAMPO represents an interesting addition to the non-parametric application layered on top of LLMs, as it supports black-box LLMs without necessitating the outputting of LLM's internal states (e.g., embeddings), as seen in previous approaches.

Boosting Reward Model with Preference-Conditional Multi-Aspect Synthetic Data Generation

Reward models (RMs) are crucial for aligning large language models (LLMs) with human preferences. They are trained using preference datasets where each example consists of one input prompt, two responses, and a preference label. As curating a high-quality human labeled preference dataset is both time-consuming and expensive, people often rely on existing powerful LLMs for preference label generation. This can potentially introduce noise and impede RM training. In this work, we present RMBoost, a novel synthetic preference data generation paradigm to boost reward model quality. Unlike traditional methods, which generate two responses before obtaining the preference label, RMBoost first generates one response and selects a preference label, followed by generating the second more (or less) preferred response conditioned on the pre-selected preference label and the first response. This approach offers two main advantages. First, RMBoost reduces labeling noise since preference pairs are constructed intentionally. Second, RMBoost facilitates the creation of more diverse responses by incorporating various quality aspects (e.g., helpfulness, relevance, completeness) into the prompts. We conduct extensive experiments across three diverse datasets and demonstrate that RMBoost outperforms other synthetic preference data generation techniques and significantly boosts the performance of four distinct reward models.

Multilingual Fine-Grained News Headline Hallucination Detection

The popularity of automated news headline generation has surged with advancements in pre-trained language models. However, these models often suffer from the ``hallucination'' problem, where the generated headline is not fully supported by its source article. Efforts to address this issue have predominantly focused on English, using over-simplistic classification schemes that overlook nuanced hallucination types. In this study, we introduce the first multilingual, fine-grained news headline hallucination detection dataset that contains over 11 thousand pairs in 5 languages, each annotated with detailed hallucination types by experts. We conduct extensive experiments on this dataset under two settings. First, we implement several supervised fine-tuning approaches as preparatory solutions and demonstrate this dataset's challenges and utilities. Second, we test various large language models' in-context learning abilities and propose two novel techniques, language-dependent demonstration selection and coarse-to-fine prompting, to boost the few-shot hallucination detection performance in terms of the example-F1 metric. We release this dataset to foster further research in multilingual, fine-grained headline hallucination detection.

WildAvatar: Web-scale In-the-wild Video Dataset for 3D Avatar Creation

Existing human datasets for avatar creation are typically limited to laboratory environments, wherein high-quality annotations (e.g., SMPL estimation from 3D scans or multi-view images) can be ideally provided. However, their annotating requirements are impractical for real-world images or videos, posing challenges toward real-world applications on current avatar creation methods. To this end, we propose the WildAvatar dataset, a web-scale in-the-wild human avatar creation dataset extracted from YouTube, with $10,000+$ different human subjects and scenes. WildAvatar is at least $10\times$ richer than previous datasets for 3D human avatar creation. We evaluate several state-of-the-art avatar creation methods on our dataset, highlighting the unexplored challenges in real-world applications on avatar creation. We also demonstrate the potential for generalizability of avatar creation methods, when provided with data at scale. We will publicly release our data source links and annotations, to push forward 3D human avatar creation and other related fields for real-world applications.

PLaD: Preference-based Large Language Model Distillation with Pseudo-Preference Pairs

Large Language Models (LLMs) have exhibited impressive capabilities in various tasks, yet their vast parameter sizes restrict their applicability in resource-constrained settings. Knowledge distillation (KD) offers a viable solution by transferring expertise from large teacher models to compact student models. However, traditional KD techniques face specific challenges when applied to LLMs, including restricted access to LLM outputs, significant teacher-student capacity gaps, and the inherited mis-calibration issue. In this work, we present PLaD, a novel preference-based LLM distillation framework. PLaD exploits the teacher-student capacity discrepancy to generate pseudo-preference pairs where teacher outputs are preferred over student outputs. Then, PLaD leverages a ranking loss to re-calibrate student's estimation of sequence likelihood, which steers the student's focus towards understanding the relative quality of outputs instead of simply imitating the teacher. PLaD bypasses the need for access to teacher LLM's internal states, tackles the student's expressivity limitations, and mitigates the student mis-calibration issue. Through extensive experiments on two sequence generation tasks and with various LLMs, we demonstrate the effectiveness of our proposed PLaD framework.

Video Coding with Cross-Component Sample Offset

Beyond the exploration of traditional spatial, temporal and subjective visual signal redundancy in image and video compression, recent research has focused on leveraging cross-color component redundancy to enhance coding efficiency. Cross-component coding approaches are motivated by the statistical correlations among different color components, such as those in the Y'CbCr color space, where luma (Y) color component typically exhibits finer details than chroma (Cb/Cr) color components. Inspired by previous cross-component coding algorithms, this paper introduces a novel in-loop filtering approach named Cross-Component Sample Offset (CCSO). CCSO utilizes co-located and neighboring luma samples to generate correction signals for both luma and chroma reconstructed samples. It is a multiplication-free, non-linear mapping process implemented using a look-up-table. The input to the mapping is a group of reconstructed luma samples, and the output is an offset value applied on the center luma or co-located chroma sample. Experimental results demonstrate that the proposed CCSO can be applied to both image and video coding, resulting in improved coding efficiency and visual quality. The method has been adopted into an experimental next-generation video codec beyond AV1 developed by the Alliance for Open Media (AOMedia), achieving significant objective coding gains up to 3.5\,\% and 1.8\,\% for PSNR and VMAF quality metrics, respectively, under random access configuration. Additionally, CCSO notably improves the subjective visual quality.