Ruoyu Wang

Ruoyu Wang, Jiachen Sun, Shaowei Hua et al.

Direct Preference Optimization (DPO) is a method for enhancing model performance by directly optimizing for the preferences or rankings of outcomes, instead of traditional loss functions. This approach has proven effective in aligning Large Language Models (LLMs) with human preferences. Despite its widespread use across various tasks, DPO has been criticized for its sensitivity to the effectiveness of Supervised Fine-Tuning (SFT) and its limitations in enabling models to learn human-preferred responses, leading to less satisfactory performance. To address these limitations, we propose Aligned Supervised Fine-Tuning (ASFT), an effective approach that better aligns LLMs with pair-wise datasets by optimizing absolute likelihood for each response, rather than using the Bradley-Terry model, and eliminates the need for a reference model. Through theoretical gradient analysis, we demonstrate that ASFT mitigates the issue where the DPO loss function decreases the probability of generating human-dispreferred data at a faster rate than it increases the probability of producing preferred data. Additionally, we compare ASFT to DPO and its latest variants, such as the single-step approach ORPO, using the latest instruction-tuned model Llama3, which has been fine-tuned on UltraFeedback and HH-RLHF. We evaluated performance on instruction-following benchmarks like MT-Bench and traditional text generation metrics such as BLEU-4 and ROUGE-L. Extensive experiments demonstrate that ASFT is an effective alignment approach, consistently outperforming existing methods.

Ruoyu Wang, Xiaoxuan Li, Lina Yao

Large Language Models (LLMs) have demonstrated remarkable efficiency in tackling various tasks based on human instructions, but studies reveal that they often struggle with tasks requiring reasoning, such as math or physics. This limitation raises questions about whether LLMs truly comprehend embedded knowledge or merely learn to replicate the token distribution without a true understanding of the content. In this paper, we delve into this problem and aim to enhance the reasoning capabilities of LLMs. First, we investigate if the model has genuine reasoning capabilities by visualizing the text generation process at the attention and representation level. Then, we formulate the reasoning process of LLMs into a causal framework, which provides a formal explanation of the problems observed in the visualization. Finally, building upon this causal framework, we propose Deconfounded Causal Adaptation (DCA), a novel parameter-efficient fine-tuning (PEFT) method to enhance the model's reasoning capabilities by encouraging the model to extract the general problem-solving skills and apply these skills to different questions. Experiments show that our method outperforms the baseline consistently across multiple benchmarks, and with only 1.2M tunable parameters, we achieve better or comparable results to other fine-tuning methods. This demonstrates the effectiveness and efficiency of our method in improving the overall accuracy and reliability of LLMs.

Ruoyu Wang, Lina Yao

Disentangled Representation Learning aims to improve the explainability of deep learning methods by training a data encoder that identifies semantically meaningful latent variables in the data generation process. Nevertheless, there is no consensus regarding a universally accepted definition for the objective of disentangled representation learning. In particular, there is a considerable amount of discourse regarding whether should the latent variables be mutually independent or not. In this paper, we first investigate these arguments on the interrelationships between latent variables by establishing a conceptual bridge between Epistemology and Disentangled Representation Learning. Then, inspired by these interdisciplinary concepts, we introduce a two-level latent space framework to provide a general solution to the prior arguments on this issue. Finally, we propose a novel method for disentangled representation learning by employing an integration of mutual information constraint and independence constraint within the Generative Adversarial Network (GAN) framework. Experimental results demonstrate that our proposed method consistently outperforms baseline approaches in both quantitative and qualitative evaluations. The method exhibits strong performance across multiple commonly used metrics and demonstrates a great capability in disentangling various semantic factors, leading to an improved quality of controllable generation, which consequently benefits the explainability of the algorithm.

Yuhan Pei, Ruoyu Wang, Yongqi Yang et al.

Originating from the diffusion phenomenon in physics, which describes the random movement and collisions of particles, diffusion generative models simulate a random walk in the data space along the denoising trajectory. This allows information to diffuse across regions, yielding harmonious outcomes. However, the chaotic and disordered nature of information diffusion in diffusion models often results in undesired interference between image regions, causing degraded detail preservation and contextual inconsistency. In this work, we address these challenges by reframing disordered diffusion as a powerful tool for text-vision-to-image generation (TV2I) tasks, achieving pixel-level condition fidelity while maintaining visual and semantic coherence throughout the image. We first introduce Cyclic One-Way Diffusion (COW), which provides an efficient unidirectional diffusion framework for precise information transfer while minimizing disruptive interference. Building on COW, we further propose Selective One-Way Diffusion (SOW), which utilizes Multimodal Large Language Models (MLLMs) to clarify the semantic and spatial relationships within the image. Based on these insights, SOW combines attention mechanisms to dynamically regulate the direction and intensity of diffusion according to contextual relationships. Extensive experiments demonstrate the untapped potential of controlled information diffusion, offering a path to more adaptive and versatile generative models in a learning-free manner.

Chen Cai, Tianyi Liu, Jianjun Gao et al.

Recent Multimodal Large Language Models (MLLMs) exhibit strong zero-shot abilities but struggle with complex Grounded Situation Recognition (GSR) and are resource-intensive for edge device deployment. Meanwhile, conventional GSR models often lack generalization ability, falling short in recognizing unseen and rare situations. In this paper, we exploit transferring knowledge from a teacher MLLM to a small GSR model to enhance its generalization and zero-shot abilities, thereby introducing the task of Open-vocabulary Grounded Situation Recognition (Ov-GSR). To achieve this, we propose Multimodal Interactive Prompt Distillation (MIPD), a novel framework that distills enriched multimodal knowledge from the foundation model, enabling the student Ov-GSR model to recognize unseen situations and be better aware of rare situations. Specifically, the MIPD framework first leverages the LLM-based Judgmental Rationales Generator (JRG) to construct positive and negative glimpse and gaze rationales enriched with contextual semantic information. The proposed scene-aware and instance-perception prompts are then introduced to align rationales with visual information from the MLLM teacher via the Negative-Guided Multimodal Prompting Alignment (NMPA) module, effectively capturing holistic and perceptual multimodal knowledge. Finally, the aligned multimodal knowledge is distilled into the student Ov-GSR model, providing a stronger foundation for generalization that enhances situation understanding, bridges the gap between seen and unseen scenarios, and mitigates prediction bias in rare cases. We evaluate MIPD on the refined Ov-SWiG dataset, achieving superior performance on seen, rare, and unseen situations, and further demonstrate improved unseen detection on the HICO-DET dataset.

Ruoyu Wang

Active learning (AL) is a machine learning algorithm that can achieve greater accuracy with fewer labeled training instances, for having the ability to ask oracles to label the most valuable unlabeled data chosen iteratively and heuristically by query strategies. Scientific experiments nowadays, though becoming increasingly automated, are still suffering from human involvement in the designing process and the exhaustive search in the experimental space. This article performs a retrospective study on a drug response dataset using the proposed AL scheme comprised of the matrix factorization method of alternating least square (ALS) and deep neural networks (DNN). This article also proposes an AL query strategy based on expected loss minimization. As a result, the retrospective study demonstrates that scientific experimental design, instead of being manually set, can be optimized by AL, and the proposed query strategy ELM sampling shows better experimental performance than other ones such as random sampling and uncertainty sampling.