Mingxiao Li

Mang Ning, Mingxiao Li, Jianlin Su et al.

Diffusion models have demonstrated impressive generative capabilities, but their \textit{exposure bias} problem, described as the input mismatch between training and sampling, lacks in-depth exploration. In this paper, we systematically investigate the exposure bias problem in diffusion models by first analytically modelling the sampling distribution, based on which we then attribute the prediction error at each sampling step as the root cause of the exposure bias issue. Furthermore, we discuss potential solutions to this issue and propose an intuitive metric for it. Along with the elucidation of exposure bias, we propose a simple, yet effective, training-free method called Epsilon Scaling to alleviate the exposure bias. We show that Epsilon Scaling explicitly moves the sampling trajectory closer to the vector field learned in the training phase by scaling down the network output, mitigating the input mismatch between training and sampling. Experiments on various diffusion frameworks (ADM, DDIM, EDM, LDM, DiT, PFGM++) verify the effectiveness of our method. Remarkably, our ADM-ES, as a state-of-the-art stochastic sampler, obtains 2.17 FID on CIFAR-10 under 100-step unconditional generation. The code is available at \url{https://github.com/forever208/ADM-ES} and \url{https://github.com/forever208/EDM-ES}.

Mingxiao Li, Zehao Wang, Tinne Tuytelaars et al.

In this work, we study the problem of Embodied Referring Expression Grounding, where an agent needs to navigate in a previously unseen environment and localize a remote object described by a concise high-level natural language instruction. When facing such a situation, a human tends to imagine what the destination may look like and to explore the environment based on prior knowledge of the environmental layout, such as the fact that a bathroom is more likely to be found near a bedroom than a kitchen. We have designed an autonomous agent called Layout-aware Dreamer (LAD), including two novel modules, that is, the Layout Learner and the Goal Dreamer to mimic this cognitive decision process. The Layout Learner learns to infer the room category distribution of neighboring unexplored areas along the path for coarse layout estimation, which effectively introduces layout common sense of room-to-room transitions to our agent. To learn an effective exploration of the environment, the Goal Dreamer imagines the destination beforehand. Our agent achieves new state-of-the-art performance on the public leaderboard of the REVERIE dataset in challenging unseen test environments with improvement in navigation success (SR) by 4.02% and remote grounding success (RGS) by 3.43% compared to the previous state-of-the-art. The code is released at https://github.com/zehao-wang/LAD

Mingxiao Li, Marie-Francine Moens

Knowledge-based visual question answering (VQA) is a vision-language task that requires an agent to correctly answer image-related questions using knowledge that is not presented in the given image. It is not only a more challenging task than regular VQA but also a vital step towards building a general VQA system. Most existing knowledge-based VQA systems process knowledge and image information similarly and ignore the fact that the knowledge base (KB) contains complete information about a triplet, while the extracted image information might be incomplete as the relations between two objects are missing or wrongly detected. In this paper, we propose a novel model named dynamic knowledge memory enhanced multi-step graph reasoning (DMMGR), which performs explicit and implicit reasoning over a key-value knowledge memory module and a spatial-aware image graph, respectively. Specifically, the memory module learns a dynamic knowledge representation and generates a knowledge-aware question representation at each reasoning step. Then, this representation is used to guide a graph attention operator over the spatial-aware image graph. Our model achieves new state-of-the-art accuracy on the KRVQR and FVQA datasets. We also conduct ablation experiments to prove the effectiveness of each component of the proposed model.

Zehao Wang, Mingxiao Li, Minye Wu et al.

In this paper, we introduce the map-language navigation task where an agent executes natural language instructions and moves to the target position based only on a given 3D semantic map. To tackle the task, we design the instruction-aware Path Proposal and Discrimination model (iPPD). Our approach leverages map information to provide instruction-aware path proposals, i.e., it selects all potential instruction-aligned candidate paths to reduce the solution space. Next, to represent the map observations along a path for a better modality alignment, a novel Path Feature Encoding scheme tailored for semantic maps is proposed. An attention-based Language Driven Discriminator is designed to evaluate path candidates and determine the best path as the final result. Our method can naturally avoid error accumulation compared with single-step greedy decision methods. Comparing to a single-step imitation learning approach, iPPD has performance gains above 17% on navigation success and 0.18 on path matching measurement nDTW in challenging unseen environments.

Mingxiao Li, Marie-Francine Moens

Visual dialog is a vision-language task where an agent needs to answer a series of questions grounded in an image based on the understanding of the dialog history and the image. The occurrences of coreference relations in the dialog makes it a more challenging task than visual question-answering. Most previous works have focused on learning better multi-modal representations or on exploring different ways of fusing visual and language features, while the coreferences in the dialog are mainly ignored. In this paper, based on linguistic knowledge and discourse features of human dialog we propose two soft constraints that can improve the model's ability of resolving coreferences in dialog in an unsupervised way. Experimental results on the VisDial v1.0 dataset shows that our model, which integrates two novel and linguistically inspired soft constraints in a deep transformer neural architecture, obtains new state-of-the-art performance in terms of recall at 1 and other evaluation metrics compared to current existing models and this without pretraining on other vision-language datasets. Our qualitative results also demonstrate the effectiveness of the method that we propose.

Jingyuan Sun, Mingxiao Li, Marie-Francine Moens

Reconstructing visual stimuli from human brain activities provides a promising opportunity to advance our understanding of the brain's visual system and its connection with computer vision models. Although deep generative models have been employed for this task, the challenge of generating high-quality images with accurate semantics persists due to the intricate underlying representations of brain signals and the limited availability of parallel data. In this paper, we propose a two-phase framework named Contrast and Diffuse (CnD) to decode realistic images from functional magnetic resonance imaging (fMRI) recordings. In the first phase, we acquire representations of fMRI data through self-supervised contrastive learning. In the second phase, the encoded fMRI representations condition the diffusion model to reconstruct visual stimulus through our proposed concept-aware conditioning method. Experimental results show that CnD reconstructs highly plausible images on challenging benchmarks. We also provide a quantitative interpretation of the connection between the latent diffusion model (LDM) components and the human brain's visual system. In summary, we present an effective approach for reconstructing visual stimuli based on human brain activity and offer a novel framework to understand the relationship between the diffusion model and the human brain visual system.

Wei Sun, Mingxiao Li, Damien Sileo et al.

Medical Question Answering~(medical QA) systems play an essential role in assisting healthcare workers in finding answers to their questions. However, it is not sufficient to merely provide answers by medical QA systems because users might want explanations, that is, more analytic statements in natural language that describe the elements and context that support the answer. To do so, we propose a novel approach for generating natural language explanations for answers predicted by medical QA systems. As high-quality medical explanations require additional medical knowledge, so that our system extract knowledge from medical textbooks to enhance the quality of explanations during the explanation generation process. Concretely, we designed an expectation-maximization approach that makes inferences about the evidence found in these texts, offering an efficient way to focus attention on lengthy evidence passages. Experimental results, conducted on two datasets MQAE-diag and MQAE, demonstrate the effectiveness of our framework for reasoning with textual evidence. Our approach outperforms state-of-the-art models, achieving a significant improvement of \textbf{6.86} and \textbf{9.43} percentage points on the Rouge-1 score; \textbf{8.23} and \textbf{7.82} percentage points on the Bleu-4 score on the respective datasets.

Ailin Huang, Boyong Wu, Bruce Wang et al.

Real-time speech interaction, serving as a fundamental interface for human-machine collaboration, holds immense potential. However, current open-source models face limitations such as high costs in voice data collection, weakness in dynamic control, and limited intelligence. To address these challenges, this paper introduces Step-Audio, the first production-ready open-source solution. Key contributions include: 1) a 130B-parameter unified speech-text multi-modal model that achieves unified understanding and generation, with the Step-Audio-Chat version open-sourced; 2) a generative speech data engine that establishes an affordable voice cloning framework and produces the open-sourced lightweight Step-Audio-TTS-3B model through distillation; 3) an instruction-driven fine control system enabling dynamic adjustments across dialects, emotions, singing, and RAP; 4) an enhanced cognitive architecture augmented with tool calling and role-playing abilities to manage complex tasks effectively. Based on our new StepEval-Audio-360 evaluation benchmark, Step-Audio achieves state-of-the-art performance in human evaluations, especially in terms of instruction following. On open-source benchmarks like LLaMA Question, shows 9.3% average performance improvement, demonstrating our commitment to advancing the development of open-source multi-modal language technologies. Our code and models are available at https://github.com/stepfun-ai/Step-Audio.

Mingxiao Li, Rui Jin, Liyao Xiang et al.

The traditional methods for data compression are typically based on the symbol-level statistics, with the information source modeled as a long sequence of i.i.d. random variables or a stochastic process, thus establishing the fundamental limit as entropy for lossless compression and as mutual information for lossy compression. However, the source (including text, music, and speech) in the real world is often statistically ill-defined because of its close connection to human perception, and thus the model-driven approach can be quite suboptimal. This study places careful emphasis on English text and exploits its semantic aspect to enhance the compression efficiency further. The main idea stems from the puzzle crossword, observing that the hidden words can still be precisely reconstructed so long as some key letters are provided. The proposed masking-based strategy resembles the above game. In a nutshell, the encoder evaluates the semantic importance of each word according to the semantic loss and then masks the minor ones, while the decoder aims to recover the masked words from the semantic context by means of the Transformer. Our experiments show that the proposed semantic approach can achieve much higher compression efficiency than the traditional methods such as Huffman code and UTF-8 code, while preserving the meaning in the target text to a great extent.

Boyong Wu, Chao Yan, Chen Hu et al.

This paper presents Step-Audio 2, an end-to-end multi-modal large language model designed for industry-strength audio understanding and speech conversation. By integrating a latent audio encoder and reasoning-centric reinforcement learning (RL), Step-Audio 2 achieves promising performance in automatic speech recognition (ASR) and audio understanding. To facilitate genuine end-to-end speech conversation, Step-Audio 2 incorporates the generation of discrete audio tokens into language modeling, significantly enhancing its responsiveness to paralinguistic information such as speaking styles and emotions. To effectively leverage the rich textual and acoustic knowledge in real-world data, Step-Audio 2 integrates retrieval-augmented generation (RAG) and is able to call external tools such as web search to mitigate hallucination and audio search to switch timbres. Trained on millions of hours of speech and audio data, Step-Audio 2 delivers intelligence and expressiveness across diverse conversational scenarios. Evaluation results demonstrate that Step-Audio 2 achieves state-of-the-art performance on various audio understanding and conversational benchmarks compared to other open-source and commercial solutions. Please visit https://github.com/stepfun-ai/Step-Audio2 for more information.

Wenhai Lai, Mingxiao Li, Kaiming Shen et al.

Deploying metasurfaces (MTSs) to eliminate wireless blind spots requires jointly determining the physical placement of MTSs and the meta-atom phase shifts. Existing methods typically rely on explicit channel estimation, which incurs prohibitive overhead and is often intractable in real-world networks. To sidestep this bottleneck, we propose RFZero, a channel-state-information (CSI)-free deployment paradigm. Instead of estimating channels, RFZero extracts macro-environmental features from visual photos to guide MTS placement, and leverages reference signal received power (RSRP) feedback for dynamic phase-shift optimization. Most importantly, RFZero operates independently of base stations, thereby enabling seamless plug-and-play implementation. Real-world field tests confirm that RFZero completely eliminates all blind spots in a $100\text{ m}^2$ indoor area using just a pair of $1.5\text{ m}\times 0.9\text{ m}$ MTSs.

Teng Xiao, Yige Yuan, Zhengyu Chen et al.

Existing preference optimization objectives for language model alignment require additional hyperparameters that must be extensively tuned to achieve optimal performance, increasing both the complexity and time required for fine-tuning large language models. In this paper, we propose a simple yet effective hyperparameter-free preference optimization algorithm for alignment. We observe that promising performance can be achieved simply by optimizing inverse perplexity, which is calculated as the inverse of the exponentiated average log-likelihood of the chosen and rejected responses in the preference dataset. The resulting simple learning objective, SimPER, is easy to implement and eliminates the need for expensive hyperparameter tuning and a reference model, making it both computationally and memory efficient. Extensive experiments on widely used real-world benchmarks, including MT-Bench, AlpacaEval 2, and 10 key benchmarks of the Open LLM Leaderboard with 5 base models, demonstrate that SimPER consistently and significantly outperforms existing approaches-even without any hyperparameters or a reference model . For example, despite its simplicity, SimPER outperforms state-of-the-art methods by up to 5.7 points on AlpacaEval 2 and achieves the highest average ranking across 10 benchmarks on the Open LLM Leaderboard. The source code for SimPER is publicly available at: https://github.com/tengxiao1/SimPER.

Tingyu Qu, Mingxiao Li, Tinne Tuytelaars et al.

Recent advances in multimodal Large Language Models (LLMs) have shown great success in understanding multi-modal contents. For video understanding tasks, training-based video LLMs are difficult to build due to the scarcity of high-quality, curated video-text paired data. In contrast, paired image-text data are much easier to obtain, and there is substantial similarity between images and videos. Consequently, extending image LLMs for video understanding tasks presents an appealing alternative. Developing effective strategies for compressing visual tokens from multiple frames is a promising way to leverage the powerful pre-trained image LLM. In this work, we explore the limitations of the existing compression strategies for building a training-free video LLM. The findings lead to our method TS-LLaVA, which constructs visual tokens through a Thumbnail-and-Sampling strategy. Given a video, we select few equidistant frames from all input frames to construct a Thumbnail image as a detailed visual cue, complemented by Sampled visual tokens from all input frames. Our method establishes the new state-of-the-art performance among training-free video LLMs on various benchmarks. Notably, our 34B model outperforms GPT-4V on the MVBench benchmark, and achieves performance comparable to the 72B training-based video LLM, Video-LLaMA2, on the challenging MLVU benchmark. Code is available at https://github.com/tingyu215/TS-LLaVA.

Mingxiao Li, Mang Ning, Marie-Francine Moens

Text-to-image generation models have made significant progress in producing high-quality images from textual descriptions, yet they continue to struggle with maintaining subject consistency across multiple images, a fundamental requirement for visual storytelling. Existing methods attempt to address this by either fine-tuning models on large-scale story visualization datasets, which is resource-intensive, or by using training-free techniques that share information across generations, which still yield limited success. In this paper, we introduce a novel training-free sampling strategy called Zigzag Sampling with Asymmetric Prompts and Visual Sharing to enhance subject consistency in visual story generation. Our approach proposes a zigzag sampling mechanism that alternates between asymmetric prompting to retain subject characteristics, while a visual sharing module transfers visual cues across generated images to %further enforce consistency. Experimental results, based on both quantitative metrics and qualitative evaluations, demonstrate that our method significantly outperforms previous approaches in generating coherent and consistent visual stories. The code is available at https://github.com/Mingxiao-Li/Asymmetry-Zigzag-StoryDiffusion.

Mingxiao Li, Tingyu Qu, Ruicong Yao et al.

Diffusion Probabilistic Models (DPM) have shown remarkable efficacy in the synthesis of high-quality images. However, their inference process characteristically requires numerous, potentially hundreds, of iterative steps, which could exaggerate the problem of exposure bias due to the training and inference discrepancy. Previous work has attempted to mitigate this issue by perturbing inputs during training, which consequently mandates the retraining of the DPM. In this work, we conduct a systematic study of exposure bias in DPM and, intriguingly, we find that the exposure bias could be alleviated with a novel sampling method that we propose, without retraining the model. We empirically and theoretically show that, during inference, for each backward time step $t$ and corresponding state $\hat{x}_t$, there might exist another time step $t_s$ which exhibits superior coupling with $\hat{x}_t$. Based on this finding, we introduce a sampling method named Time-Shift Sampler. Our framework can be seamlessly integrated to existing sampling algorithms, such as DDPM, DDIM and other high-order solvers, inducing merely minimal additional computations. Experimental results show our method brings significant and consistent improvements in FID scores on different datasets and sampling methods. For example, integrating Time-Shift Sampler to F-PNDM yields a FID=3.88, achieving 44.49\% improvements as compared to F-PNDM, on CIFAR-10 with 10 sampling steps, which is more performant than the vanilla DDIM with 100 sampling steps. Our code is available at https://github.com/Mingxiao-Li/TS-DPM.

Teng Xiao, Yige Yuan, Huaisheng Zhu et al.

We study the problem of aligning large language models (LLMs) with human preference data. Contrastive preference optimization has shown promising results in aligning LLMs with available preference data by optimizing the implicit reward associated with the policy. However, the contrastive objective focuses mainly on the relative values of implicit rewards associated with two responses while ignoring their actual values, resulting in suboptimal alignment with human preferences. To address this limitation, we propose calibrated direct preference optimization (Cal-DPO), a simple yet effective algorithm. We show that substantial improvement in alignment with the given preferences can be achieved simply by calibrating the implicit reward to ensure that the learned implicit rewards are comparable in scale to the ground-truth rewards. We demonstrate the theoretical advantages of Cal-DPO over existing approaches. The results of our experiments on a variety of standard benchmarks show that Cal-DPO remarkably improves off-the-shelf methods.

Teng Xiao, Mingxiao Li, Yige Yuan et al.

This paper introduces a novel generalized self-imitation learning ($\textbf{GSIL}$) framework, which effectively and efficiently aligns large language models with offline demonstration data. We develop $\textbf{GSIL}$ by deriving a surrogate objective of imitation learning with density ratio estimates, facilitating the use of self-generated data and optimizing the imitation learning objective with simple classification losses. $\textbf{GSIL}$ eliminates the need for complex adversarial training in standard imitation learning, achieving lightweight and efficient fine-tuning for large language models. In addition, $\textbf{GSIL}$ encompasses a family of offline losses parameterized by a general class of convex functions for density ratio estimation and enables a unified view for alignment with demonstration data. Extensive experiments show that $\textbf{GSIL}$ consistently and significantly outperforms baselines in many challenging benchmarks, such as coding (HuamnEval), mathematical reasoning (GSM8K) and instruction-following benchmark (MT-Bench).

Jingyuan Sun, Mingxiao Li, Zijiao Chen et al.

In the pursuit to understand the intricacies of human brain's visual processing, reconstructing dynamic visual experiences from brain activities emerges as a challenging yet fascinating endeavor. While recent advancements have achieved success in reconstructing static images from non-invasive brain recordings, the domain of translating continuous brain activities into video format remains underexplored. In this work, we introduce NeuroCine, a novel dual-phase framework to targeting the inherent challenges of decoding fMRI data, such as noises, spatial redundancy and temporal lags. This framework proposes spatial masking and temporal interpolation-based augmentation for contrastive learning fMRI representations and a diffusion model enhanced by dependent prior noise for video generation. Tested on a publicly available fMRI dataset, our method shows promising results, outperforming the previous state-of-the-art models by a notable margin of ${20.97\%}$, ${31.00\%}$ and ${12.30\%}$ respectively on decoding the brain activities of three subjects in the fMRI dataset, as measured by SSIM. Additionally, our attention analysis suggests that the model aligns with existing brain structures and functions, indicating its biological plausibility and interpretability.

Mingxiao Li, Bo Wan, Marie-Francine Moens et al.

In recent years, diffusion models have made remarkable strides in text-to-video generation, sparking a quest for enhanced control over video outputs to more accurately reflect user intentions. Traditional efforts predominantly focus on employing either semantic cues, like images or depth maps, or motion-based conditions, like moving sketches or object bounding boxes. Semantic inputs offer a rich scene context but lack detailed motion specificity; conversely, motion inputs provide precise trajectory information but miss the broader semantic narrative. For the first time, we integrate both semantic and motion cues within a diffusion model for video generation, as demonstrated in Fig 1. To this end, we introduce the Scene and Motion Conditional Diffusion (SMCD), a novel methodology for managing multimodal inputs. It incorporates a recognized motion conditioning module and investigates various approaches to integrate scene conditions, promoting synergy between different modalities. For model training, we separate the conditions for the two modalities, introducing a two-stage training pipeline. Experimental results demonstrate that our design significantly enhances video quality, motion precision, and semantic coherence.

Mingxiao Li, Fang Qu, Zhanpeng Chen et al.

While MLLMs perform well on perceptual tasks, they lack precise multimodal alignment, limiting performance. To address this challenge, we propose Vision Dynamic Embedding-Guided Pretraining (VDEP), a hybrid autoregressive training paradigm for MLLMs. Utilizing dynamic embeddings from the MLP following the visual encoder, this approach supervises image hidden states and integrates image tokens into autoregressive training. Existing MLLMs primarily focused on recovering information from textual inputs, often neglecting the effective processing of image data. In contrast, the key improvement of this work is the reinterpretation of multimodal alignment as a process of recovering information from input data, with particular emphasis on reconstructing detailed visual features.The proposed method seamlessly integrates into standard models without architectural changes. Experiments on 13 benchmarks show VDEP outperforms baselines, surpassing existing methods.

Maria Mihaela Trusca, Mingxiao Li, Marie-Francine Moens

Text-based image editing is typically approached as a static task that involves operations such as inserting, deleting, or modifying elements of an input image based on human instructions. Given the static nature of this task, in this paper, we aim to make this task dynamic by incorporating actions. By doing this, we intend to modify the positions or postures of objects in the image to depict different actions while maintaining the visual properties of the objects. To implement this challenging task, we propose a new model that is sensitive to action text instructions by learning to recognize contrastive action discrepancies. The model training is done on new datasets defined by extracting frames from videos that show the visual scenes before and after an action. We show substantial improvements in image editing using action-based text instructions and high reasoning capabilities that allow our model to use the input image as a starting scene for an action while generating a new image that shows the final scene of the action.

Jaron Maene, Mingxiao Li, Marie-Francine Moens

The lottery ticket hypothesis states that sparse subnetworks exist in randomly initialized dense networks that can be trained to the same accuracy as the dense network they reside in. However, the subsequent work has failed to replicate this on large-scale models and required rewinding to an early stable state instead of initialization. We show that by using a training method that is stable with respect to linear mode connectivity, large networks can also be entirely rewound to initialization. Our subsequent experiments on common vision tasks give strong credence to the hypothesis in Evci et al. (2020b) that lottery tickets simply retrain to the same regions (although not necessarily to the same basin). These results imply that existing lottery tickets could not have been found without the preceding dense training by iterative magnitude pruning, raising doubts about the use of the lottery ticket hypothesis.