arXiv subjects

Sheng Liu

Publications and source records attributed to Sheng Liu.

At least 91 records · Page 5Linked to original sources

arXiv2024

TFG: Unified Training-Free Guidance for Diffusion Models

Given an unconditional diffusion model and a predictor for a target property of interest (e.g., a classifier), the goal of training-free guidance is to generate samples with desirable target properties without additional training. Existing methods, though effective in various individual applications, often lack theoretical grounding and rigorous testing on extensive benchmarks. As a result, they could even fail on simple tasks, and applying them to a new problem becomes unavoidably difficult. This paper introduces a novel algorithmic framework encompassing existing methods as special cases, unifying the study of training-free guidance into the analysis of an algorithm-agnostic design space. Via theoretical and empirical investigation, we propose an efficient and effective hyper-parameter searching strategy that can be readily applied to any downstream task. We systematically benchmark across 7 diffusion models on 16 tasks with 40 targets, and improve performance by 8.5% on average. Our framework and benchmark offer a solid foundation for conditional generation in a training-free manner.

cs.LG↗

arXiv2024

Multimodal Instruction Tuning with Hybrid State Space Models

Handling lengthy context is crucial for enhancing the recognition and understanding capabilities of multimodal large language models (MLLMs) in applications such as processing high-resolution images or high frame rate videos. The rise in image resolution and frame rate substantially increases computational demands due to the increased number of input tokens. This challenge is further exacerbated by the quadratic complexity with respect to sequence length of the self-attention mechanism. Most prior works either pre-train models with long contexts, overlooking the efficiency problem, or attempt to reduce the context length via downsampling (e.g., identify the key image patches or frames) to decrease the context length, which may result in information loss. To circumvent this issue while keeping the remarkable effectiveness of MLLMs, we propose a novel approach using a hybrid transformer-MAMBA model to efficiently handle long contexts in multimodal applications. Our multimodal model can effectively process long context input exceeding 100k tokens, outperforming existing models across various benchmarks. Remarkably, our model enhances inference efficiency for high-resolution images and high-frame-rate videos by about 4 times compared to current models, with efficiency gains increasing as image resolution or video frames rise. Furthermore, our model is the first to be trained on low-resolution images or low-frame-rate videos while being capable of inference on high-resolution images and high-frame-rate videos, offering flexibility for inference in diverse scenarios.

cs.CV↗

arXiv2024

Reducing Hallucinations in Vision-Language Models via Latent Space Steering

Hallucination poses a challenge to the deployment of large vision-language models (LVLMs) in applications. Unlike in large language models (LLMs), hallucination in LVLMs often arises from misalignments between visual inputs and textual outputs. This paper investigates the underlying mechanisms of hallucination, focusing on the unique structure of LVLMs that distinguishes them from large language models (LLMs). We identify that hallucinations often arise from the sensitivity of text decoders to vision inputs, a natural phenomenon when image encoders and text decoders are pre-trained separately. Inspired by this, we introduce Visual and Textual Intervention (VTI), a novel technique designed to reduce hallucinations by steering latent space representations during inference to enhance the stability of vision features. As a task-agnostic test-time intervention, VTI can be easily applied to any problem without additional cost. Extensive experiments demonstrate that it can effectively reduce hallucinations and outperform baseline methods across multiple metrics, highlighting the critical role of vision feature stability in LVLMs.

cs.CV↗

arXiv2024

The robustness of skyrmion numbers of structured optical fields in atmospheric turbulence

The development of vector optical fields has brought forth numerous applications. Among these optical fields, a particular class of vector vortex beams has emerged, leading to the emergence of intriguing optical skyrmion fields characterized by skyrmion numbers. The optical skyrmion fields are well-defined by their effective magnetization and possess topologically protected configurations. It is anticipated that this type of optical structure can be exploited for encoding information in optical communication, even under perturbations such as turbulent air, optical fibers, and even general random media. In this study, we numerically demonstrate that the skyrmion numbers of optical skyrmion fields exhibit a certain degree of robustness to atmospheric turbulence, even though their intensity, phase and polarization patterns are distorted. Intriguingly, it is also observed that a larger difference between the absolute values of two azimuthal indices of the vectorial structured light field can lead to a superior level of resilience. These properties not only enhance the versatility of skyrmion fields and their numbers, but also open up new possibilities for their use in various applications across noisy channels.

physics.optics↗

arXiv2024

Orientation-dependent surface radiation damage in $β$-Ga2O3 explored by multiscale atomic simulations

Ultrawide bandgap semiconductor $β$-Ga2O3 holds extensive potential for applications in high-radiation environments. One of the primary challenges in its practical application is unveiling the mechanisms of surface irradiation damage under extreme conditions. In this study, we investigate the orientation-dependent mechanisms of radiation damage on four experimentally relevant $β$-Ga2O3 surface facets, namely, (100), (010), (001), and (-201), at various temperatures. We employ a multiscale atomic simulation approach, combining machine-learning-driven molecular dynamics (ML-MD) simulations and density functional theory (DFT) calculations. The results reveal that Ga vacancies and O interstitials are the predominant defects across all four surfaces, with the formation of many antisite defects Ga_O and few O_Ga observed. Among the two Ga sites and three O sites, the vacancy found in the O2 site is dominant, while the interstitials at the Ga1 and O1 sites are more significant. Interestingly, the (010) surface exhibits the lowest defect density, owing to its more profound channeling effect leading to a broader spread of defects. The influence of temperature on surface irradiation damage of $β$-Ga2O3 should be evaluated based on the unique crystal surface characteristics. Moreover, the formation energy and defect concentration calculated by DFT corroborate the results of the MD simulations. Comprehending surface radiation damage at the atomic level is crucial for assessing the radiation tolerance and predicting the performance changes of $β$-Ga2O3-based device in high-radiation environments.

cond-mat.mtrl-sci↗

arXiv2024

Understanding and Improving Transfer Learning of Deep Models via Neural Collapse

With the ever-increasing complexity of large-scale pre-trained models coupled with a shortage of labeled data for downstream training, transfer learning has become the primary approach in many fields, including natural language processing, computer vision, and multi-modal learning. Despite recent progress, the fine-tuning process for large-scale pre-trained models in vision still mostly relies on trial and error. This work investigates the relationship between neural collapse (NC) and transfer learning for classification problems. NC is an intriguing while prevalent phenomenon that has been recently discovered in terms of the final-layer features and linear classifiers of trained neural networks. Specifically, during the terminal phase of training, NC implies that the variability of the features within each class diminishes to zero, while the means of features between classes are maximally and equally distanced. In this work, we examine the NC attributes of pre-trained models on both downstream and source data for transfer learning, and we find strong correlation between feature collapse and downstream performance. In particular, we discovered a systematic pattern that emerges when linear probing pre-trained models on downstream training data: the more feature collapse of pre-trained models on downstream training data, the higher the transfer accuracy. Additionally, we also studied the relationship between NC and transfer accuracy on the source data. Moreover, these findings allow us to develop a principled, parameter-efficient fine-tuning method that employs skip-connection to induce the last-layer feature collapse on downstream data. Our proposed fine-tuning methods deliver good performances while reducing fine-tuning parameters by at least 90% and mitigating overfitting in situations especially when the downstream data is scarce.

cs.LG↗

arXiv2024

TextGrad: Automatic "Differentiation" via Text

AI is undergoing a paradigm shift, with breakthroughs achieved by systems orchestrating multiple large language models (LLMs) and other complex components. As a result, developing principled and automated optimization methods for compound AI systems is one of the most important new challenges. Neural networks faced a similar challenge in its early days until backpropagation and automatic differentiation transformed the field by making optimization turn-key. Inspired by this, we introduce TextGrad, a powerful framework performing automatic ``differentiation'' via text. TextGrad backpropagates textual feedback provided by LLMs to improve individual components of a compound AI system. In our framework, LLMs provide rich, general, natural language suggestions to optimize variables in computation graphs, ranging from code snippets to molecular structures. TextGrad follows PyTorch's syntax and abstraction and is flexible and easy-to-use. It works out-of-the-box for a variety of tasks, where the users only provide the objective function without tuning components or prompts of the framework. We showcase TextGrad's effectiveness and generality across a diverse range of applications, from question answering and molecule optimization to radiotherapy treatment planning. Without modifying the framework, TextGrad improves the zero-shot accuracy of GPT-4o in Google-Proof Question Answering from $51\%$ to $55\%$, yields $20\%$ relative performance gain in optimizing LeetCode-Hard coding problem solutions, improves prompts for reasoning, designs new druglike small molecules with desirable in silico binding, and designs radiation oncology treatment plans with high specificity. TextGrad lays a foundation to accelerate the development of the next-generation of AI systems.

cs.CL↗

arXiv2024

Dual Dynamic Threshold Adjustment Strategy for Deep Metric Learning

Loss functions and sample mining strategies are essential components in deep metric learning algorithms. However, the existing loss function or mining strategy often necessitate the incorporation of additional hyperparameters, notably the threshold, which defines whether the sample pair is informative. The threshold provides a stable numerical standard for determining whether to retain the pairs. It is a vital parameter to reduce the redundant sample pairs participating in training. Nonetheless, finding the optimal threshold can be a time-consuming endeavor, often requiring extensive grid searches. Because the threshold cannot be dynamically adjusted in the training stage, we should conduct plenty of repeated experiments to determine the threshold. Therefore, we introduce a novel approach for adjusting the thresholds associated with both the loss function and the sample mining strategy. We design a static Asymmetric Sample Mining Strategy (ASMS) and its dynamic version Adaptive Tolerance ASMS (AT-ASMS), tailored for sample mining methods. ASMS utilizes differentiated thresholds to address the problems (too few positive pairs and too many redundant negative pairs) caused by only applying a single threshold to filter samples. AT-ASMS can adaptively regulate the ratio of positive and negative pairs during training according to the ratio of the currently mined positive and negative pairs. This meta-learning-based threshold generation algorithm utilizes a single-step gradient descent to obtain new thresholds. We combine these two threshold adjustment algorithms to form the Dual Dynamic Threshold Adjustment Strategy (DDTAS). Experimental results show that our algorithm achieves competitive performance on CUB200, Cars196, and SOP datasets.

cs.MM↗

arXiv2024

GNSS Measurement-Based Context Recognition for Vehicle Navigation using Gated Recurrent Unit

Recent years, people have put forward higher and higher requirements for context-adaptive navigation (CAN). CAN system realizes seamless navigation in complex environments by recognizing the ambient surroundings of vehicles, and it is crucial to develop a fast, reliable, and robust navigational context recognition (NCR) method to enable CAN systems to operate effectively. Environmental context recognition based on Global Navigation Satellite System (GNSS) measurements has attracted widespread attention due to its low cost because it does not require additional infrastructure. The performance and application value of NCR methods depend on three main factors: context categorization, feature extraction, and classification models. In this paper, a fine-grained context categorization framework comprising seven environment categories (open sky, tree-lined avenue, semi-outdoor, urban canyon, viaduct-down, shallow indoor, and deep indoor) is proposed, which currently represents the most elaborate context categorization framework known in this research domain. To improve discrimination between categories, a new feature called the C/N0-weighted azimuth distribution factor, is designed. Then, to ensure real-time performance, a lightweight gated recurrent unit (GRU) network is adopted for its excellent sequence data processing capabilities. A dataset containing 59,996 samples is created and made publicly available to researchers in the NCR community on Github. Extensive experiments have been conducted on the dataset, and the results show that the proposed method achieves an overall recognition accuracy of 99.41\% for isolated scenarios and 94.95\% for transition scenarios, with an average transition delay of 2.14 seconds.

eess.SP↗

arXiv2024

Mapping the Increasing Use of LLMs in Scientific Papers

Scientific publishing lays the foundation of science by disseminating research findings, fostering collaboration, encouraging reproducibility, and ensuring that scientific knowledge is accessible, verifiable, and built upon over time. Recently, there has been immense speculation about how many people are using large language models (LLMs) like ChatGPT in their academic writing, and to what extent this tool might have an effect on global scientific practices. However, we lack a precise measure of the proportion of academic writing substantially modified or produced by LLMs. To address this gap, we conduct the first systematic, large-scale analysis across 950,965 papers published between January 2020 and February 2024 on the arXiv, bioRxiv, and Nature portfolio journals, using a population-level statistical framework to measure the prevalence of LLM-modified content over time. Our statistical estimation operates on the corpus level and is more robust than inference on individual instances. Our findings reveal a steady increase in LLM usage, with the largest and fastest growth observed in Computer Science papers (up to 17.5%). In comparison, Mathematics papers and the Nature portfolio showed the least LLM modification (up to 6.3%). Moreover, at an aggregate level, our analysis reveals that higher levels of LLM-modification are associated with papers whose first authors post preprints more frequently, papers in more crowded research areas, and papers of shorter lengths. Our findings suggests that LLMs are being broadly used in scientific writings.

cs.CL↗

arXiv2024

Beyond Gradient and Priors in Privacy Attacks: Leveraging Pooler Layer Inputs of Language Models in Federated Learning

Language models trained via federated learning (FL) demonstrate impressive capabilities in handling complex tasks while protecting user privacy. Recent studies indicate that leveraging gradient information and prior knowledge can potentially reveal training samples within FL setting. However, these investigations have overlooked the potential privacy risks tied to the intrinsic architecture of the models. This paper presents a two-stage privacy attack strategy that targets the vulnerabilities in the architecture of contemporary language models, significantly enhancing attack performance by initially recovering certain feature directions as additional supervisory signals. Our comparative experiments demonstrate superior attack performance across various datasets and scenarios, highlighting the privacy leakage risk associated with the increasingly complex architectures of language models. We call for the community to recognize and address these potential privacy risks in designing large language models.

cs.LG↗

arXiv2024

Quantum Light Generation based on GaN Microring towards Fully On-chip Source

Integrated quantum light source is increasingly desirable in large-scale quantum information processing.~Despite recent remarkable advances, new material platform is constantly being explored for the fully on-chip integration of quantum light generation, active and passive manipulation, and detection. Here, for the first time, we demonstrate a gallium nitride (GaN) microring based quantum light generation in the telecom C-band, which has potential towards the monolithic integration of quantum light source.~In our demonstration, the GaN microring has a free spectral range of 330 GHz and a near-zero anomalous dispersion region of over 100 nm. The generation of energy-time entangled photon pair is demonstrated with a typical raw two-photon interference visibility of 95.5$\pm$6.5%, which is further configured to generate heralded single photon with a typical heralded second-order auto-correlation $g^{(2)}_{H}(0)$ of 0.045$\pm$0.001. Our results pave the way for developing chip-scale quantum photonic circuit.

quant-ph↗

arXiv2024

In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering

Large language models (LLMs) demonstrate emergent in-context learning capabilities, where they adapt to new tasks based on example demonstrations. However, in-context learning has seen limited effectiveness in many settings, is difficult to quantitatively control and takes up context window space. To overcome these limitations, we propose an alternative approach that recasts in-context learning as in-context vectors (ICV). Using ICV has two steps. We first use a forward pass on demonstration examples to create the in-context vector from the latent embedding of the LLM. This vector captures essential information about the intended task. On a new query, instead of adding demonstrations to the prompt, we shift the latent states of the LLM using the ICV. The ICV approach has several benefits: 1) it enables the LLM to more effectively follow the demonstration examples; 2) it's easy to control by adjusting the magnitude of the ICV; 3) it reduces the length of the prompt by removing the in-context demonstrations; 4) ICV is computationally much more efficient than fine-tuning. We demonstrate that ICV achieves better performance compared to standard in-context learning and fine-tuning on diverse tasks including safety, style transfer, role-playing and formatting. Moreover, we show that we can flexibly teach LLM to simultaneously follow different types of instructions by simple vector arithmetics on the corresponding ICVs.

cs.LG↗

arXiv2024

Recovering Sign Bits of DCT Coefficients in Digital Images as an Optimization Problem

Recovering unknown, missing, damaged, distorted, or lost information in DCT coefficients is a common task in multiple applications of digital image processing, including image compression, selective image encryption, and image communication. This paper investigates the recovery of sign bits in DCT coefficients of digital images, by proposing two different approximation methods to solve a mixed integer linear programming (MILP) problem, which is NP-hard in general. One method is a relaxation of the MILP problem to a linear programming (LP) problem, and the other splits the original MILP problem into some smaller MILP problems and an LP problem. We considered how the proposed methods can be applied to JPEG-encoded images and conducted extensive experiments to validate their performances. The experimental results showed that the proposed methods outperformed other existing methods by a substantial margin, both according to objective quality metrics and our subjective evaluation.

cs.CV↗

arXiv2023

The Security Analysis of Continuous-Variable Quantum Key Distribution under Limited Eavesdropping with Practical Fiber

Research on optimal eavesdropping models under practical conditions will help to evaluate realistic risk when employing quantum key distribution (QKD) system for secure information transmission. Intuitively, fiber loss will lead to the optical energy leaking to the environment, rather than harvested by the eavesdropper, which also limits the eavesdropping ability while improving the QKD system performance in practical use. However, defining the optimal eavesdropping model in the presence of lossy fiber is difficult because the channel is beyond the control of legitimate partners and the leaked signal is undetectable. Here we investigate how the fiber loss influences the eavesdropping ability based on a teleportation-based collective attack model which requires two distant stations and a shared entanglement source. We find that if the distributed entanglement is limited due to the practical loss, the optimal attack occurs when the two teleportation stations are merged to one and placed close to the transmitter site, which performs similar to the entangling-cloning attack but with a reduced wiretapping ratio. Assuming Eve uses the best available hollow-core fiber, the secret key rate in the practical environment can be 20%~40% higher than that under ideal eavesdropping. While if the entanglement distillation technology is mature enough to provide high quality of distributed entanglement, the two teleportation stations should be distantly separated for better eavesdropping performance, where the eavesdropping can even approach the optimal collective attack. Under the current level of entanglement purification technology, the unavoidable fiber loss can still greatly limit the eavesdropping ability as well as enhance the secret key rate and transmission distance of the realistic system, which promotes the development of QKD systems in practical application scenarios.

quant-ph↗

arXiv2023

Making Self-supervised Learning Robust to Spurious Correlation via Learning-speed Aware Sampling

Self-supervised learning (SSL) has emerged as a powerful technique for learning rich representations from unlabeled data. The data representations are able to capture many underlying attributes of data, and be useful in downstream prediction tasks. In real-world settings, spurious correlations between some attributes (e.g. race, gender and age) and labels for downstream tasks often exist, e.g. cancer is usually more prevalent among elderly patients. In this paper, we investigate SSL in the presence of spurious correlations and show that the SSL training loss can be minimized by capturing only a subset of the conspicuous features relevant to those sensitive attributes, despite the presence of other important predictive features for the downstream tasks. To address this issue, we investigate the learning dynamics of SSL and observe that the learning is slower for samples that conflict with such correlations (e.g. elder patients without cancer). Motivated by these findings, we propose a learning-speed aware SSL (LA-SSL) approach, in which we sample each training data with a probability that is inversely related to its learning speed. We evaluate LA-SSL on three datasets that exhibit spurious correlations between different attributes, demonstrating that it improves the robustness of pretrained representations on downstream classification tasks.

cs.LG↗

arXiv2023

Channel-Wise Contrastive Learning for Learning with Noisy Labels

In real-world datasets, noisy labels are pervasive. The challenge of learning with noisy labels (LNL) is to train a classifier that discerns the actual classes from given instances. For this, the model must identify features indicative of the authentic labels. While research indicates that genuine label information is embedded in the learned features of even inaccurately labeled data, it's often intertwined with noise, complicating its direct application. Addressing this, we introduce channel-wise contrastive learning (CWCL). This method distinguishes authentic label information from noise by undertaking contrastive learning across diverse channels. Unlike conventional instance-wise contrastive learning (IWCL), CWCL tends to yield more nuanced and resilient features aligned with the authentic labels. Our strategy is twofold: firstly, using CWCL to extract pertinent features to identify cleanly labeled samples, and secondly, progressively fine-tuning using these samples. Evaluations on several benchmark datasets validate our method's superiority over existing approaches.

cs.LG↗

arXiv2023

Multiple Instance Learning via Iterative Self-Paced Supervised Contrastive Learning

Learning representations for individual instances when only bag-level labels are available is a fundamental challenge in multiple instance learning (MIL). Recent works have shown promising results using contrastive self-supervised learning (CSSL), which learns to push apart representations corresponding to two different randomly-selected instances. Unfortunately, in real-world applications such as medical image classification, there is often class imbalance, so randomly-selected instances mostly belong to the same majority class, which precludes CSSL from learning inter-class differences. To address this issue, we propose a novel framework, Iterative Self-paced Supervised Contrastive Learning for MIL Representations (ItS2CLR), which improves the learned representation by exploiting instance-level pseudo labels derived from the bag-level labels. The framework employs a novel self-paced sampling strategy to ensure the accuracy of pseudo labels. We evaluate ItS2CLR on three medical datasets, showing that it improves the quality of instance-level pseudo labels and representations, and outperforms existing MIL methods in terms of both bag and instance level accuracy. Code is available at https://github.com/Kangningthu/ItS2CLR

cs.CV↗