Yufei Liu

Yichong Leng, Zhifang Guo, Kai Shen et al. · microsoft-research

Speech conveys more information than text, as the same word can be uttered in various voices to convey diverse information. Compared to traditional text-to-speech (TTS) methods relying on speech prompts (reference speech) for voice variability, using text prompts (descriptions) is more user-friendly since speech prompts can be hard to find or may not exist at all. TTS approaches based on the text prompt face two main challenges: 1) the one-to-many problem, where not all details about voice variability can be described in the text prompt, and 2) the limited availability of text prompt datasets, where vendors and large cost of data labeling are required to write text prompts for speech. In this work, we introduce PromptTTS 2 to address these challenges with a variation network to provide variability information of voice not captured by text prompts, and a prompt generation pipeline to utilize the large language models (LLM) to compose high quality text prompts. Specifically, the variation network predicts the representation extracted from the reference speech (which contains full information about voice variability) based on the text prompt representation. For the prompt generation pipeline, it generates text prompts for speech with a speech language understanding model to recognize voice attributes (e.g., gender, speed) from speech and a large language model to formulate text prompts based on the recognition results. Experiments on a large-scale (44K hours) speech dataset demonstrate that compared to the previous works, PromptTTS 2 generates voices more consistent with text prompts and supports the sampling of diverse voice variability, thereby offering users more choices on voice generation. Additionally, the prompt generation pipeline produces high-quality text prompts, eliminating the large labeling cost. The demo page of PromptTTS 2 is available online.

Yizhou Peng, Yufei Liu, Jicheng Zhang et al.

Internal Language Model Estimation (ILME) based language model (LM) fusion has been shown significantly improved recognition results over conventional shallow fusion in both intra-domain and cross-domain speech recognition tasks. In this paper, we attempt to apply our ILME method to cross-domain code-switching speech recognition (CSSR) work. Specifically, our curiosity comes from several aspects. First, we are curious about how effective the ILME-based LM fusion is for both intra-domain and cross-domain CSSR tasks. We verify this with or without merging two code-switching domains. More importantly, we train an end-to-end (E2E) speech recognition model by means of merging two monolingual data sets and observe the efficacy of the proposed ILME-based LM fusion for CSSR. Experimental results on SEAME that is from Southeast Asian and another Chinese Mainland CS data set demonstrate the effectiveness of the proposed ILME-based LM fusion method.

Yufei Liu, Jie Cao, Dechang Pi

Social influence prediction has permeated many domains, including marketing, behavior prediction, recommendation systems, and more. However, traditional methods of predicting social influence not only require domain expertise,they also rely on extracting user features, which can be very tedious. Additionally, graph convolutional networks (GCNs), which deals with graph data in non-Euclidean space, are not directly applicable to Euclidean space. To overcome these problems, we extended DeepInf such that it can predict the social influence of COVID-19 via the transition probability of the page rank domain. Furthermore, our implementation gives rise to a deep learning-based personalized propagation algorithm, called DeepPP. The resulting algorithm combines the personalized propagation of a neural prediction model with the approximate personalized propagation of a neural prediction model from page rank analysis. Four social networks from different domains as well as two COVID-19 datasets were used to demonstrate the efficiency and effectiveness of the proposed algorithm. Compared to other baseline methods, DeepPP provides more accurate social influence predictions. Further, experiments demonstrate that DeepPP can be applied to real-world prediction data for COVID-19.

Yu Qi, Fan Yang, Yousong Zhu et al.

Autoregressive language modeling (ALM) have been successfully used in self-supervised pre-training in Natural language processing (NLP). However, this paradigm has not achieved comparable results with other self-supervised approach in computer vision (e.g., contrastive learning, mask image modeling). In this paper, we try to find the reason why autoregressive modeling does not work well on vision tasks. To tackle this problem, we fully analyze the limitation of visual autoregressive methods and proposed a novel stochastic autoregressive image modeling (named SAIM) by the two simple designs. First, we employ stochastic permutation strategy to generate effective and robust image context which is critical for vision tasks. Second, we create a parallel encoder-decoder training process in which the encoder serves a similar role to the standard vision transformer focus on learning the whole contextual information, and meanwhile the decoder predicts the content of the current position, so that the encoder and decoder can reinforce each other. By introducing stochastic prediction and the parallel encoder-decoder, SAIM significantly improve the performance of autoregressive image modeling. Our method achieves the best accuracy (83.9%) on the vanilla ViT-Base model among methods using only ImageNet-1K data. Transfer performance in downstream tasks also show that our model achieves competitive performance.

Shuanglong Yao, Dechang Pi, Junfu Chen et al.

The task of link prediction aims to solve the problem of incomplete knowledge caused by the difficulty of collecting facts from the real world. GCNs-based models are widely applied to solve link prediction problems due to their sophistication, but GCNs-based models are suffering from two problems in the structure and training process. 1) The transformation methods of GCN layers become increasingly complex in GCN-based knowledge representation models; 2) Due to the incompleteness of the knowledge graph collection process, there are many uncollected true facts in the labeled negative samples. Therefore, this paper investigates the characteristic of the information aggregation coefficient (self-attention) of adjacent nodes and redesigns the self-attention mechanism of the GAT structure. Meanwhile, inspired by human thinking habits, we designed a semi-supervised self-training method over pre-trained models. Experimental results on the benchmark datasets FB15k-237 and WN18RR show that our proposed self-attention mechanism and semi-supervised self-training method can effectively improve the performance of the link prediction task. If you look at FB15k-237, for example, the proposed method improves Hits@1 by about 30%.

Yirong Zeng, Xiao Ding, Yufei Liu et al.

Tool use represents a critical capability for AI agents, with recent advances focusing on leveraging reinforcement learning (RL) to scale up the explicit reasoning process to achieve better performance. However, there are some key challenges for tool use in current RL-based scaling approaches: (a) direct RL training often struggles to scale up thinking length sufficiently to solve complex problems, and (b) scaled-up models tend to overthink simpler problems, resulting in substantial token inefficiency. To address these challenges, we propose a novel training paradigm that first employs warm-up supervised fine-tuning to help models distinguish between simple and complex problems, followed by RL that enable models to automatically determine appropriate reasoning trajectories. Furthermore, to tackle the issue of automatic thinking-length scaling, we discover that entropy-based optimization objectives effectively maintain model diversity while successfully unlocking the model's scaling capabilities. Based on this insight, we introduce an entropy-based long-short reasoning fusion RL strategy. Our experiments on three benchmarks demonstrate that model successfully achieves auto-scaling for efficient tool use, achieving significant 9.8\% accuracy improvements while reducing computational overhead by \textasciitilde81\%.

Yirong Zeng, Yufei Liu, Xiao Ding et al.

A central belief in scaling reinforcement learning with verifiable rewards for instruction following (IF) tasks is that, a diverse mixture of verifiable hard and unverifiable soft constraints is essential for generalizing to unseen instructions. In this work, we challenge this prevailing consensus through a systematic empirical investigation. Counter-intuitively, we find that models trained on hard-only constraints consistently outperform those trained on mixed datasets. Extensive experiments reveal that reward precision, rather than constraint diversity, is the primary driver of effective alignment. The LLM judge suffers from a low recall rate in detecting false response, which leads to severe reward hacking, thereby undermining the benefits of diversity. Furthermore, analysis of the attention mechanism reveals that high-precision rewards develop a transferable meta-skill for IF. Motivated by these insights, we propose a simple yet effective data-centric refinement strategy that prioritizes reward precision. Evaluated on five benchmarks, our approach outperforms competitive baselines by 13.4\% in performance while achieving a 58\% reduction in training time, maintaining strong generalization beyond instruction following. Our findings advocate for a paradigm shift: moving away from the indiscriminate pursuit of data diversity toward high-precision rewards.

Dongyang Liu, Shicheng Li, Yutong Liu et al.

Recent advancements have established Diffusion Transformers (DiTs) as a dominant framework in generative modeling. Building on this success, Lumina-Next achieves exceptional performance in the generation of photorealistic images with Next-DiT. However, its potential for video generation remains largely untapped, with significant challenges in modeling the spatiotemporal complexity inherent to video data. To address this, we introduce Lumina-Video, a framework that leverages the strengths of Next-DiT while introducing tailored solutions for video synthesis. Lumina-Video incorporates a Multi-scale Next-DiT architecture, which jointly learns multiple patchifications to enhance both efficiency and flexibility. By incorporating the motion score as an explicit condition, Lumina-Video also enables direct control of generated videos' dynamic degree. Combined with a progressive training scheme with increasingly higher resolution and FPS, and a multi-source training scheme with mixed natural and synthetic data, Lumina-Video achieves remarkable aesthetic quality and motion smoothness at high training and inference efficiency. We additionally propose Lumina-V2A, a video-to-audio model based on Next-DiT, to create synchronized sounds for generated videos. Codes are released at https://www.github.com/Alpha-VLLM/Lumina-Video.

Yufei Liu, Manzhou Li, Qin Ma

This study proposes a method based on lightweight convolutional neural networks (CNN) and generative adversarial networks (GAN) for apple ripeness and damage level detection tasks. Initially, a lightweight CNN model is designed by optimizing the model's depth and width, as well as employing advanced model compression techniques, successfully reducing the model's parameter and computational requirements, thus enhancing real-time performance in practical applications. Simultaneously, attention mechanisms are introduced, dynamically adjusting the importance of different feature layers to improve the performance in object detection tasks. To address the issues of sample imbalance and insufficient sample size, GANs are used to generate realistic apple images, expanding the training dataset and enhancing the model's recognition capability when faced with apples of varying ripeness and damage levels. Furthermore, by applying the object detection network for damage location annotation on damaged apples, the accuracy of damage level detection is improved, providing a more precise basis for decision-making. Experimental results show that in apple ripeness grading detection, the proposed model achieves 95.6\%, 93.8\%, 95.0\%, and 56.5 in precision, recall, accuracy, and FPS, respectively. In apple damage level detection, the proposed model reaches 95.3\%, 93.7\%, and 94.5\% in precision, recall, and mAP, respectively. In both tasks, the proposed method outperforms other mainstream models, demonstrating the excellent performance and high practical value of the proposed method in apple ripeness and damage level detection tasks.

Yirong Zeng, Shen You, Yufei Liu et al.

Equipping LLMs with external tools effectively addresses internal reasoning limitations. However, it introduces a critical yet under-explored phenomenon: tool overuse, the unnecessary tool-use during reasoning. In this paper, we first reveal this phenomenon is pervasive across diverse LLMs. We then experimentally elucidate its underlying mechanisms through two key lenses: (1) First, by analyzing tool-use behavior across different internal knowledge availability regions, we identify a \textit{knowledge epistemic illusion}: models misjudge internal knowledge boundaries and fail to accurately perceive their actual knowledge availability. To mitigate this, we propose a knowledge-aware epistemic boundary alignment strategy based on direct preference optimization, which reduces tool usage in by 82.8\% while yielding an accuracy improvement. (2) Second, we establish a causal link between reward structures and tool-use behavior by visualizing the tool-augmented training process. It reveals that \textit{outcome-only rewards} inadvertently encourage tool overuse by rewarding only final correctness, regardless of tool efficiency. To verify this, we balance reward signals during training rather than relying on outcome-only rewards, cutting unnecessary tool calls by 66.7\% (7B) and 60.7\% (32B) without sacrificing accuracy. Finally, we provide theoretical justification in this two lenses to understand tool overuse.

Xuying Zhang, Yutong Liu, Yangguang Li et al.

We present TAR3D, a novel framework that consists of a 3D-aware Vector Quantized-Variational AutoEncoder (VQ-VAE) and a Generative Pre-trained Transformer (GPT) to generate high-quality 3D assets. The core insight of this work is to migrate the multimodal unification and promising learning capabilities of the next-token prediction paradigm to conditional 3D object generation. To achieve this, the 3D VQ-VAE first encodes a wide range of 3D shapes into a compact triplane latent space and utilizes a set of discrete representations from a trainable codebook to reconstruct fine-grained geometries under the supervision of query point occupancy. Then, the 3D GPT, equipped with a custom triplane position embedding called TriPE, predicts the codebook index sequence with prefilling prompt tokens in an autoregressive manner so that the composition of 3D geometries can be modeled part by part. Extensive experiments on ShapeNet and Objaverse demonstrate that TAR3D can achieve superior generation quality over existing methods in text-to-3D and image-to-3D tasks

Qianxi Hua, Xinyue Li, Zheng Yan et al.

Embodied intelligence has advanced rapidly in recent years; however, bimanual manipulation-especially in contact-rich tasks remains challenging. This is largely due to the lack of datasets with rich physical interaction signals, systematic task organization, and sufficient scale. To address these limitations, we introduce the VTOUCH dataset. It leverages vision based tactile sensing to provide high-fidelity physical interaction signals, adopts a matrix-style task design to enable systematic learning, and employs automated data collection pipelines covering real-world, demand-driven scenarios to ensure scalability. To further validate the effectiveness of the dataset, we conduct extensive quantitative experiments on cross-modal retrieval as well as real-robot evaluation. Finally, we demonstrate real-world performance through generalizable inference across multiple robots, policies, and tasks.

Yanqing Liu, Ruiqing Xue, Chong Zhang et al.

While diffusion and autoregressive (AR) models have significantly advanced generative modeling, they each present distinct limitations. AR models, which rely on causal attention, cannot exploit future context and suffer from slow generation speeds. Conversely, diffusion models struggle with key-value (KV) caching. To overcome these challenges, we introduce Dragon-FM, a novel text-to-speech (TTS) design that unifies AR and flow-matching. This model processes 48 kHz audio codec tokens in chunks at a compact rate of 12.5 tokens per second. This design enables AR modeling across chunks, ensuring global coherence, while parallel flow-matching within chunks facilitates fast iterative denoising. Thus, the model leverages KV-cache across chunks and utilizes bidirectional context within each chunk. Furthermore, it bridges continuous and discrete feature modeling, demonstrating that continuous AR flow-matching can predict discrete tokens with finite scalar quantizers. This efficient codec and fast chunk-autoregressive architecture also make the model highly effective for generating long-form content, such as podcasts. Experiments on podcast datasets demonstrate its capability to efficiently generate high-quality zero-shot podcasts.

Yufei Liu, Xieyuanli Chen, Hainan Pan et al.

The prevailing paradigm of perceptive humanoid locomotion relies heavily on active depth sensors. However, this depth-centric approach fundamentally discards the rich semantic and dense appearance cues of the visual world, severing low-level control from the high-level reasoning essential for general embodied intelligence. While monocular RGB offers a ubiquitous, information-dense alternative, end-to-end reinforcement learning from raw 2D pixels suffers from extreme sample inefficiency and catastrophic sim-to-real collapse due to the inherent loss of geometric scale. To break this deadlock, we propose GeoLoco, a purely RGB-driven locomotion framework that conceptualizes monocular images as high-dimensional 3D latent representations by harnessing the powerful geometric priors of a frozen, scale-aware Visual Foundation Model (VFM). Rather than naive feature concatenation, we design a proprioceptive-query multi-head cross-attention mechanism that dynamically attends to task-critical topological features conditioned on the robot's real-time gait phase. Crucially, to prevent the policy from overfitting to superficial textures, we introduce a dual-head auxiliary learning scheme. This explicit regularization forces the high-dimensional latent space to strictly align with the physical terrain geometry, ensuring robust zero-shot sim-to-real transfer. Trained exclusively in simulation, GeoLoco achieves robust zero-shot transfer to the Unitree G1 humanoid and successfully negotiates challenging terrains.

Yuzhi Liu, Zixuan Chen, Zirui Zhang et al.

The Neural Tangent Kernel (NTK) offers a powerful tool to study the functional dynamics of neural networks. In the so-called lazy, or kernel regime, the NTK remains static during training and the network function is linear in the static neural tangents feature space. The evolution of the NTK during training is necessary for feature learning, a key driver of deep learning success. The study of the NTK dynamics has led to several critical discoveries in recent years, in generalization and scaling behaviours. However, this body of work has been limited to the single task setting, where the data distribution is assumed constant over time. In this work, we present a comprehensive empirical analysis of NTK dynamics in continual learning, where the data distribution shifts over time. Our findings highlight continual learning as a rich and underutilized testbed for probing the dynamics of neural training. At the same time, they challenge the validity of static-kernel approximations in theoretical treatments of continual learning, even at large scale.

Yufei Liu, Haoke Xiao, Jiaxing Chai et al.

The advent of Large Vision Models (LVMs) offers new opportunities for few-shot medical image segmentation. However, existing training-free methods based on LVMs fail to effectively utilize negative prompts, leading to poor performance on low-contrast medical images. To address this issue, we propose SynPo, a training-free few-shot method based on LVMs (e.g., SAM), with the core insight: improving the quality of negative prompts. To select point prompts in a more reliable confidence map, we design a novel Confidence Map Synergy Module by combining the strengths of DINOv2 and SAM. Based on the confidence map, we select the top-k pixels as the positive points set and choose the negative points set using a Gaussian distribution, followed by independent K-means clustering for both sets. Then, these selected points are leveraged as high-quality prompts for SAM to get the segmentation results. Extensive experiments demonstrate that SynPo achieves performance comparable to state-of-the-art training-based few-shot methods.

Mingzhuang Wang, Yvyang Li, Xiyang Zhang et al.

Coral reefs, crucial for sustaining marine biodiversity and ecological processes (e.g., nutrient cycling, habitat provision), face escalating threats, underscoring the need for efficient monitoring. Coral reef ecological monitoring faces dual challenges of low efficiency in manual analysis and insufficient segmentation accuracy in complex underwater scenarios. This study develops the YH-MINER system, establishing an intelligent framework centered on the Multimodal Large Model (MLLM) for "object detection-semantic segmentation-prior input". The system uses the object detection module (mAP@0.5=0.78) to generate spatial prior boxes for coral instances, driving the segment module to complete pixel-level segmentation in low-light and densely occluded scenarios. The segmentation masks and finetuned classification instructions are fed into the Qwen2-VL-based multimodal model as prior inputs, achieving a genus-level classification accuracy of 88% and simultaneously extracting core ecological metrics. Meanwhile, the system retains the scalability of the multimodal model through standardized interfaces, laying a foundation for future integration into multimodal agent-based underwater robots and supporting the full-process automation of "image acquisition-prior generation-real-time analysis".

Yufei Liu, Junshu Tang, Chu Zheng et al.

High-fidelity 3D garment synthesis from text is desirable yet challenging for digital avatar creation. Recent diffusion-based approaches via Score Distillation Sampling (SDS) have enabled new possibilities but either intricately couple with human body or struggle to reuse. We introduce ClotheDreamer, a 3D Gaussian-based method for generating wearable, production-ready 3D garment assets from text prompts. We propose a novel representation Disentangled Clothe Gaussian Splatting (DCGS) to enable separate optimization. DCGS represents clothed avatar as one Gaussian model but freezes body Gaussian splats. To enhance quality and completeness, we incorporate bidirectional SDS to supervise clothed avatar and garment RGBD renderings respectively with pose conditions and propose a new pruning strategy for loose clothing. Our approach can also support custom clothing templates as input. Benefiting from our design, the synthetic 3D garment can be easily applied to virtual try-on and support physically accurate animation. Extensive experiments showcase our method's superior and competitive performance. Our project page is at https://ggxxii.github.io/clothedreamer.

Yufei Liu, Junwei Zhu, Junshu Tang et al.

Texturing 3D humans with semantic UV maps remains a challenge due to the difficulty of acquiring reasonably unfolded UV. Despite recent text-to-3D advancements in supervising multi-view renderings using large text-to-image (T2I) models, issues persist with generation speed, text consistency, and texture quality, resulting in data scarcity among existing datasets. We present TexDreamer, the first zero-shot multimodal high-fidelity 3D human texture generation model. Utilizing an efficient texture adaptation finetuning strategy, we adapt large T2I model to a semantic UV structure while preserving its original generalization capability. Leveraging a novel feature translator module, the trained model is capable of generating high-fidelity 3D human textures from either text or image within seconds. Furthermore, we introduce ArTicuLated humAn textureS (ATLAS), the largest high-resolution (1024 X 1024) 3D human texture dataset which contains 50k high-fidelity textures with text descriptions.

Yufei Liu, Rao Ma, Haihua Xu et al.

An end-to-end (E2E) ASR model implicitly learns a prior Internal Language Model (ILM) from the training transcripts. To fuse an external LM using Bayes posterior theory, the log likelihood produced by the ILM has to be accurately estimated and subtracted. In this paper we propose two novel approaches to estimate the ILM based on Listen-Attend-Spell (LAS) framework. The first method is to replace the context vector of the LAS decoder at every time step with a vector that is learned with training transcripts. Furthermore, we propose another method that uses a lightweight feed-forward network to directly map query vector to context vector in a dynamic sense. Since the context vectors are learned by minimizing the perplexities on training transcripts, and their estimation is independent of encoder output, hence the ILMs are accurately learned for both methods. Experiments show that the ILMs achieve the lowest perplexity, indicating the efficacy of the proposed methods. In addition, they also significantly outperform the shallow fusion method, as well as two previously proposed ILM Estimation (ILME) approaches on several datasets.

Liang Zhang, Yufei Liu, Hang Xiao et al.

Scene text detection has received attention for years and achieved an impressive performance across various benchmarks. In this work, we propose an efficient and accurate approach to detect multioriented text in scene images. The proposed feature fusion mechanism allows us to use a shallower network to reduce the computational complexity. A self-attention mechanism is adopted to suppress false positive detections. Experiments on public benchmarks including ICDAR 2013, ICDAR 2015 and MSRA-TD500 show that our proposed approach can achieve better or comparable performances with fewer parameters and less computational cost.