arXiv subjects

Mingxiao Li

Publications and source records attributed to Mingxiao Li.

At least 37 records · Page 2Linked to original sources

arXiv2025

Efficient and wavelength-tunable second-harmonic generation towards the green gap

Achieving compact and efficient visible laser sources is crucial for a wide range of applications. However traditional semiconductor laser technology faces difficulties in producing high-brightness green light, leaving a green gap in wavelength coverage. Second-harmonic generation (SHG) offers a promising alternative by converting near-infrared sources to visible wavelengths with high efficiency and spectral purity. Here, we demonstrate efficient and tunable SHG within the green spectrum using a high-Q Si3N4 microresonator. A space-charge grating induced by the photogalvanic effect realizes reconfigurable grating numbers and flexible wavelength tuning. Additionally, grating formation dynamics and competition is observed. These findings underscore the potential of silicon nitride as a robust, integrative platform for on-chip, tunable green light sources.

physics.optics↗

arXiv2025

Towards More Accurate Personalized Image Generation: Addressing Overfitting and Evaluation Bias

Personalized image generation via text prompts has great potential to improve daily life and professional work by facilitating the creation of customized visual content. The aim of image personalization is to create images based on a user-provided subject while maintaining both consistency of the subject and flexibility to accommodate various textual descriptions of that subject. However, current methods face challenges in ensuring fidelity to the text prompt while not overfitting to the training data. In this work, we introduce a novel training pipeline that incorporates an attractor to filter out distractions in training images, allowing the model to focus on learning an effective representation of the personalized subject. Moreover, current evaluation methods struggle due to the lack of a dedicated test set. The evaluation set-up typically relies on the training data of the personalization task to compute text-image and image-image similarity scores, which, while useful, tend to overestimate performance. Although human evaluations are commonly used as an alternative, they often suffer from bias and inconsistency. To address these issues, we curate a diverse and high-quality test set with well-designed prompts. With this new benchmark, automatic evaluation metrics can reliably assess model performance

cs.CV↗

arXiv2025

On a Connection Between Imitation Learning and RLHF

This work studies the alignment of large language models with preference data from an imitation learning perspective. We establish a close theoretical connection between reinforcement learning from human feedback RLHF and imitation learning (IL), revealing that RLHF implicitly performs imitation learning on the preference data distribution. Building on this connection, we propose DIL, a principled framework that directly optimizes the imitation learning objective. DIL provides a unified imitation learning perspective on alignment, encompassing existing alignment algorithms as special cases while naturally introducing new variants. By bridging IL and RLHF, DIL offers new insights into alignment with RLHF. Extensive experiments demonstrate that DIL outperforms existing methods on various challenging benchmarks.

cs.LG↗

arXiv2025

SimPER: A Minimalist Approach to Preference Alignment without Hyperparameters

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.

cs.LG↗

arXiv2025

Step-Audio: Unified Understanding and Generation in Intelligent Speech Interaction

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.

cs.CL↗

arXiv2025

From Visuals to Vocabulary: Establishing Equivalence Between Image and Text Token Through Autoregressive Pre-training in MLLMs

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.

cs.CV↗

arXiv2025

Advancing General Multimodal Capability of Vision-language Models with Pyramid-descent Visual Position Encoding

Vision-language Models (VLMs) have shown remarkable capabilities in advancing general artificial intelligence, yet the irrational encoding of visual positions persists in inhibiting the models' comprehensive perception performance across different levels of granularity. In this work, we propose Pyramid-descent Visual Position Encoding (PyPE), a novel approach designed to enhance the perception of visual tokens within VLMs. By assigning visual position indexes from the periphery to the center and expanding the central receptive field incrementally, PyPE addresses the limitations of traditional raster-scan methods and mitigates the long-term decay effects induced by Rotary Position Embedding (RoPE). Our method reduces the relative distance between interrelated visual elements and instruction tokens, promoting a more rational allocation of attention weights and allowing for a multi-granularity perception of visual elements and countering the over-reliance on anchor tokens. Extensive experimental evaluations demonstrate that PyPE consistently improves the general capabilities of VLMs across various sizes. Code is available at https://github.com/SakuraTroyChen/PyPE.

cs.CV↗

arXiv2025

Action-based image editing guided by human instructions

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.

cs.CV↗

arXiv2025

A Semantic Approach to Successive Interference Cancellation for Multiple Access Networks

Differing from the conventional communication system paradigm that models information source as a sequence of (i.i.d. or stationary) random variables, the semantic approach aims at extracting and sending the high-level features of the content deeply contained in the source, thereby breaking the performance limits from the statistical information theory. As a pioneering work in this area, the deep learning-enabled semantic communication (DeepSC) constitutes a novel algorithmic framework based on the transformer--which is a deep learning tool widely used to process text numerically. The main goal of this work is to extend the DeepSC approach from the point-to-point link to the multi-user multiple access channel (MAC). The inter-user interference has long been identified as the bottleneck of the MAC. In the classic information theory, the successive interference cancellation (SIC) scheme is a common way to mitigate interference and achieve the channel capacity. Our main contribution is to incorporate the SIC scheme into the DeepSC. As opposed to the traditional SIC that removes interference in the digital symbol domain, the proposed semantic SIC works in the domain of the semantic word embedding vectors. Furthermore, to enhance the training efficiency, we propose a pretraining scheme and a partial retraining scheme that quickly adjust the neural network parameters when new users are added to the MAC. We also modify the existing loss function to facilitate training. Finally, we present numerical experiments to demonstrate the advantage of the proposed semantic approach as compared to the existing benchmark methods.

cs.IT↗

arXiv2024

Cal-DPO: Calibrated Direct Preference Optimization for Language Model Alignment

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.

cs.LG↗

arXiv2024

A photonic integrated circuit for heterogeneous second harmonic generation

Heterogeneous integration of GaAs-based lasers with frequency doubling waveguides presents a clear path to scalable coherent sources in the so-called green gap, yet frequency doubling systems have so far relied on separately manufactured lasers to deliver enough power for second harmonic generation. In this work, we propose a photonic integrated circuit (PIC) which alleviates the performance requirements for integrated frequency doublers. Two gain sections are connected by waveguides, with a frequency converter and a wavelength separator in between. The fundamental light circulates between the gain sections until it is converted and emitted through the wavelength separator. Variants of this separated gain PIC are discussed, and the PIC is implemented with thin film lithium niobate and directly bonded GaAs-based lasers, coupled by on-chip facets and adiabatic tapers, realizing visible light generation in the 515-595 nm range.

physics.optics↗

arXiv2024

TS-LLaVA: Constructing Visual Tokens through Thumbnail-and-Sampling for Training-Free Video Large Language Models

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.

cs.CV↗

arXiv2024

How to Leverage Demonstration Data in Alignment for Large Language Model? A Self-Imitation Learning Perspective

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).

cs.CL↗

arXiv2024

Pockels Laser Directly Driving Ultrafast Optical Metrology

The invention of the laser unleashed the potential of optical metrology, leading to numerous advancements in modern science and technology. This reliance on lasers, however, also sets a bottleneck for precision optical metrology which is complicated by sophisticated photonic infrastructure required for delicate laser-wave control, leading to limited metrology performance and significant system complexity. Here we make a key step towards resolving this challenge, by demonstrating a Pockels laser with multi-functional capability that advances the optical metrology to a new level. The chip-scale laser exhibits a narrow intrinsic linewidth down to 167 Hz and a broad mode-hop-free tuning range up to 24 GHz. In particular, it offers an unprecedented frequency chirping rate up to 20 EHz/s, and an enormous modulation bandwidth >10 GHz, both orders of magnitude larger than any existing lasers. With this laser, we are able to successfully achieve velocimetry of 40 m/s at a short distance of 0.4 m, with a measurable velocity up to the first cosmic velocity at 1 m away, that is inaccessible by conventional ranging approaches, and distance metrology with a ranging resolution of <2 cm. Moreover, for the first time to the best of our knowledge, we are able to realize a dramatically simplified architecture for laser frequency stabilization, by direct locking the laser to an external reference gas cell without any extra external light control. We successfully achieve a long-term laser stability with a frequency fluctuation of only $\pm$ 6.5 MHz over 60 minutes. The demonstrated Pockels laser combines elegantly high laser coherence with ultrafast frequency reconfigurability and superior multifunctional capability that we envision to have profound impacts on many areas including communication, sensing, autonomous driving, quantum information processing, and beyond.

physics.optics↗

arXiv2024

Animate Your Motion: Turning Still Images into Dynamic Videos

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.

cs.CV↗

arXiv2024

Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps

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.

cs.CV↗

arXiv2024

Investigation of Q degradation in low-loss Si3N4 from heterogeneous laser integration

High-performance, high-volume-manufacturing Si3N4 photonics requires extremely low waveguide losses augmented with heterogeneously integrated lasers for applications beyond traditional markets of high-capacity interconnects. State-of-the-art quality factors (Q) over 200 million at 1550 nm have been shown previously; however, maintaining high Qs throughout laser fabrication has not been shown. Here, Si3N4 resonator intrinsic Qs over 100 million are demonstrated on a fully integrated heterogeneous laser platform. Qi is measured throughout laser processing steps, showing degradation down to 50 million from dry etching, metal evaporation, and ion implant steps, and controllable recovery to over 100 million from annealing at 250C - 350C.

physics.optics↗

arXiv2024

NeuroCine: Decoding Vivid Video Sequences from Human Brain Activties

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.

cs.CV↗