arXiv subjects

Feng Lin

Publications and source records attributed to Feng Lin.

At least 73 records · Page 4Linked to original sources

arXiv2024

Fast exact simulation of the first passage of a tempered stable subordinator across a non-increasing function

We construct a fast exact algorithm for the simulation of the first-passage time, jointly with the undershoot and overshoot, of a tempered stable subordinator over an arbitrary non-increasing absolutely continuous function. We prove that the running time of our algorithm has finite exponential moments and provide bounds on its expected running time with explicit dependence on the characteristics of the process and the initial value of the function. The expected running time grows at most cubically in the stability parameter (as it approaches either $0$ or $1$) and is linear in the tempering parameter and the initial value of the function. Numerical performance, based on the implementation in the dedicated GitHub repository, exhibits a good agreement with our theoretical bounds. We provide numerical examples to illustrate the performance of our algorithm in Monte Carlo estimation.

math.PR↗

arXiv2024

Exposing the Deception: Uncovering More Forgery Clues for Deepfake Detection

Deepfake technology has given rise to a spectrum of novel and compelling applications. Unfortunately, the widespread proliferation of high-fidelity fake videos has led to pervasive confusion and deception, shattering our faith that seeing is believing. One aspect that has been overlooked so far is that current deepfake detection approaches may easily fall into the trap of overfitting, focusing only on forgery clues within one or a few local regions. Moreover, existing works heavily rely on neural networks to extract forgery features, lacking theoretical constraints guaranteeing that sufficient forgery clues are extracted and superfluous features are eliminated. These deficiencies culminate in unsatisfactory accuracy and limited generalizability in real-life scenarios. In this paper, we try to tackle these challenges through three designs: (1) We present a novel framework to capture broader forgery clues by extracting multiple non-overlapping local representations and fusing them into a global semantic-rich feature. (2) Based on the information bottleneck theory, we derive Local Information Loss to guarantee the orthogonality of local representations while preserving comprehensive task-relevant information. (3) Further, to fuse the local representations and remove task-irrelevant information, we arrive at a Global Information Loss through the theoretical analysis of mutual information. Empirically, our method achieves state-of-the-art performance on five benchmark datasets.Our code is available at \url{https://github.com/QingyuLiu/Exposing-the-Deception}, hoping to inspire researchers.

cs.CV↗

arXiv2024

CLIPose: Category-Level Object Pose Estimation with Pre-trained Vision-Language Knowledge

Most of existing category-level object pose estimation methods devote to learning the object category information from point cloud modality. However, the scale of 3D datasets is limited due to the high cost of 3D data collection and annotation. Consequently, the category features extracted from these limited point cloud samples may not be comprehensive. This motivates us to investigate whether we can draw on knowledge of other modalities to obtain category information. Inspired by this motivation, we propose CLIPose, a novel 6D pose framework that employs the pre-trained vision-language model to develop better learning of object category information, which can fully leverage abundant semantic knowledge in image and text modalities. To make the 3D encoder learn category-specific features more efficiently, we align representations of three modalities in feature space via multi-modal contrastive learning. In addition to exploiting the pre-trained knowledge of the CLIP's model, we also expect it to be more sensitive with pose parameters. Therefore, we introduce a prompt tuning approach to fine-tune image encoder while we incorporate rotations and translations information in the text descriptions. CLIPose achieves state-of-the-art performance on two mainstream benchmark datasets, REAL275 and CAMERA25, and runs in real-time during inference (40FPS).

cs.CV↗

arXiv2024

Effect of particle size on the phase transformation behavior and equation of state of Si under hydrostatic loading

High-pressure synchrotron X-ray diffraction (XRD) studies have been conducted on three types of Si particles (micron, 100 nm, and 30 nm). The pressure for initiation of Si-I->Si-II phase transformation (PT) essentially increases with a reduction in particle size. For 30 nm Si particles, Si-I directly transforms to Si-XI by skipping the intermediate Si-II phase, which appears during the pressure release. The evolution of phase fractions of Si particles under hydrostatic compression is studied. The equation of state (EOS) of Si-I, Si-II, Si-V, and Si-XI for all three particle sizes is determined, and the results are compared with other studies. A simple iterative procedure is suggested to extract the EOS of Si-XI and Si-II from the data for a mixture of two and three phases with different pressures in each phase. Using previous atomistic simulations, EOS for Si-II is extended to ambient pressure, which is important for plastic strain-induced phase transformations. Surprisingly, the EOS of micron and 30 nm Si are identical, but different from 100 nm particles. In particular, the Si-I phase of 100 nm Si is less compressible than that of micron and 30 nm Si. The reverse Si-V->Si-I PT is observed for the first time after complete pressure release to the ambient for 100 nm particles.

cond-mat.mtrl-sci↗

arXiv2024

Phoneme-Based Proactive Anti-Eavesdropping with Controlled Recording Privilege

The widespread smart devices raise people's concerns of being eavesdropped on. To enhance voice privacy, recent studies exploit the nonlinearity in microphone to jam audio recorders with inaudible ultrasound. However, existing solutions solely rely on energetic masking. Their simple-form noise leads to several problems, such as high energy requirements and being easily removed by speech enhancement techniques. Besides, most of these solutions do not support authorized recording, which restricts their usage scenarios. In this paper, we design an efficient yet robust system that can jam microphones while preserving authorized recording. Specifically, we propose a novel phoneme-based noise with the idea of informational masking, which can distract both machines and humans and is resistant to denoising techniques. Besides, we optimize the noise transmission strategy for broader coverage and implement a hardware prototype of our system. Experimental results show that our system can reduce the recognition accuracy of recordings to below 50\% under all tested speech recognition systems, which is much better than existing solutions.

cs.CR↗

arXiv2024

BiTA: Bi-Directional Tuning for Lossless Acceleration in Large Language Models

Large language models (LLMs) commonly employ autoregressive generation during inference, leading to high memory bandwidth demand and consequently extended latency. To mitigate this inefficiency, we present Bi-directional Tuning for lossless Acceleration (BiTA), an innovative method expediting LLMs via streamlined semi-autoregressive generation and draft verification. Inspired by the concept of prompt tuning, we enhance LLMs with a parameter-efficient design called bi-directional tuning for the capability in semi-autoregressive generation. Employing efficient tree-based decoding, the models perform draft candidate generation and verification in parallel, ensuring outputs identical to their autoregressive counterparts under greedy sampling. BiTA serves as a lightweight plug-in module, seamlessly boosting the inference efficiency of existing LLMs without requiring additional assistance models or incurring significant extra memory costs. Applying the proposed BiTA, LLaMA-2-70B-Chat achieves a 2.7$\times$ speedup on the MT-Bench benchmark. Extensive experiments confirm our method surpasses state-of-the-art acceleration techniques.

cs.CL↗

arXiv2023

SoybeanNet: Transformer-Based Convolutional Neural Network for Soybean Pod Counting from Unmanned Aerial Vehicle (UAV) Images

Soybeans are a critical source of food, protein and oil, and thus have received extensive research aimed at enhancing their yield, refining cultivation practices, and advancing soybean breeding techniques. Within this context, soybean pod counting plays an essential role in understanding and optimizing production. Despite recent advancements, the development of a robust pod-counting algorithm capable of performing effectively in real-field conditions remains a significant challenge This paper presents a pioneering work of accurate soybean pod counting utilizing unmanned aerial vehicle (UAV) images captured from actual soybean fields in Michigan, USA. Specifically, this paper presents SoybeanNet, a novel point-based counting network that harnesses powerful transformer backbones for simultaneous soybean pod counting and localization with high accuracy. In addition, a new dataset of UAV-acquired images for soybean pod counting was created and open-sourced, consisting of 113 drone images with more than 260k manually annotated soybean pods captured under natural lighting conditions. Through comprehensive evaluations, SoybeanNet demonstrated superior performance over five state-of-the-art approaches when tested on the collected images. Remarkably, SoybeanNet achieved a counting accuracy of $84.51\%$ when tested on the testing dataset, attesting to its efficacy in real-world scenarios. The publication also provides both the source code (\url{https://github.com/JiajiaLi04/Soybean-Pod-Counting-from-UAV-Images}) and the labeled soybean dataset (\url{https://www.kaggle.com/datasets/jiajiali/uav-based-soybean-pod-images}), offering a valuable resource for future research endeavors in soybean pod counting and related fields.

cs.CV↗

arXiv2023

FLTracer: Accurate Poisoning Attack Provenance in Federated Learning

Federated Learning (FL) is a promising distributed learning approach that enables multiple clients to collaboratively train a shared global model. However, recent studies show that FL is vulnerable to various poisoning attacks, which can degrade the performance of global models or introduce backdoors into them. In this paper, we first conduct a comprehensive study on prior FL attacks and detection methods. The results show that all existing detection methods are only effective against limited and specific attacks. Most detection methods suffer from high false positives, which lead to significant performance degradation, especially in not independent and identically distributed (non-IID) settings. To address these issues, we propose FLTracer, the first FL attack provenance framework to accurately detect various attacks and trace the attack time, objective, type, and poisoned location of updates. Different from existing methodologies that rely solely on cross-client anomaly detection, we propose a Kalman filter-based cross-round detection to identify adversaries by seeking the behavior changes before and after the attack. Thus, this makes it resilient to data heterogeneity and is effective even in non-IID settings. To further improve the accuracy of our detection method, we employ four novel features and capture their anomalies with the joint decisions. Extensive evaluations show that FLTracer achieves an average true positive rate of over $96.88\%$ at an average false positive rate of less than $2.67\%$, significantly outperforming SOTA detection methods. \footnote{Code is available at \url{https://github.com/Eyr3/FLTracer}.}

cs.CR↗

arXiv2023

Universal Object Detection with Large Vision Model

Over the past few years, there has been growing interest in developing a broad, universal, and general-purpose computer vision system. Such systems have the potential to address a wide range of vision tasks simultaneously, without being limited to specific problems or data domains. This universality is crucial for practical, real-world computer vision applications. In this study, our focus is on a specific challenge: the large-scale, multi-domain universal object detection problem, which contributes to the broader goal of achieving a universal vision system. This problem presents several intricate challenges, including cross-dataset category label duplication, label conflicts, and the necessity to handle hierarchical taxonomies. To address these challenges, we introduce our approach to label handling, hierarchy-aware loss design, and resource-efficient model training utilizing a pre-trained large vision model. Our method has demonstrated remarkable performance, securing a prestigious second-place ranking in the object detection track of the Robust Vision Challenge 2022 (RVC 2022) on a million-scale cross-dataset object detection benchmark. We believe that our comprehensive study will serve as a valuable reference and offer an alternative approach for addressing similar challenges within the computer vision community. The source code for our work is openly available at https://github.com/linfeng93/Large-UniDet.

cs.CV↗

arXiv2023

Locate and Verify: A Two-Stream Network for Improved Deepfake Detection

Deepfake has taken the world by storm, triggering a trust crisis. Current deepfake detection methods are typically inadequate in generalizability, with a tendency to overfit to image contents such as the background, which are frequently occurring but relatively unimportant in the training dataset. Furthermore, current methods heavily rely on a few dominant forgery regions and may ignore other equally important regions, leading to inadequate uncovering of forgery cues. In this paper, we strive to address these shortcomings from three aspects: (1) We propose an innovative two-stream network that effectively enlarges the potential regions from which the model extracts forgery evidence. (2) We devise three functional modules to handle the multi-stream and multi-scale features in a collaborative learning scheme. (3) Confronted with the challenge of obtaining forgery annotations, we propose a Semi-supervised Patch Similarity Learning strategy to estimate patch-level forged location annotations. Empirically, our method demonstrates significantly improved robustness and generalizability, outperforming previous methods on six benchmarks, and improving the frame-level AUC on Deepfake Detection Challenge preview dataset from 0.797 to 0.835 and video-level AUC on CelebDF$\_$v1 dataset from 0.811 to 0.847. Our implementation is available at https://github.com/sccsok/Locate-and-Verify.

cs.CV↗

arXiv2023

DFIL: Deepfake Incremental Learning by Exploiting Domain-invariant Forgery Clues

The malicious use and widespread dissemination of deepfake pose a significant crisis of trust. Current deepfake detection models can generally recognize forgery images by training on a large dataset. However, the accuracy of detection models degrades significantly on images generated by new deepfake methods due to the difference in data distribution. To tackle this issue, we present a novel incremental learning framework that improves the generalization of deepfake detection models by continual learning from a small number of new samples. To cope with different data distributions, we propose to learn a domain-invariant representation based on supervised contrastive learning, preventing overfit to the insufficient new data. To mitigate catastrophic forgetting, we regularize our model in both feature-level and label-level based on a multi-perspective knowledge distillation approach. Finally, we propose to select both central and hard representative samples to update the replay set, which is beneficial for both domain-invariant representation learning and rehearsal-based knowledge preserving. We conduct extensive experiments on four benchmark datasets, obtaining the new state-of-the-art average forgetting rate of 7.01 and average accuracy of 85.49 on FF++, DFDC-P, DFD, and CDF2. Our code is released at https://github.com/DeepFakeIL/DFIL.

cs.CV↗

arXiv2023

Learning in Dynamic Systems and Its Application to Adaptive PID Control

Deep learning using neural networks has revolutionized machine learning and put artificial intelligence into everyday life. In order to introduce self-learning to dynamic systems other than neural networks, we extend the Brandt-Lin learning algorithm of neural networks to a large class of dynamic systems. This extension is possible because the Brandt-Lin algorithm does not require a dedicated step to back-propagate the errors in neural networks. To this end, we first generalize signal-flow graphs so that they can be used to model nonlinear systems as well as linear systems. We then derive the extended Brandt-Lin algorithm that can be used to adapt the weights of branches in generalized signal-flow graphs. We show the applications of the new algorithm by applying it to adaptive PID control. In particular, we derive a new adaptation law for PID controllers. We verify the effectiveness of the method using simulations for linear and nonlinear plants, stable as well as unstable plants.

eess.SY↗

arXiv2023

Plastic strain-induced olivine-ringwoodite phase transformation at room temperature: main rules and the mechanism of the deep-focus earthquake

Deep-focus earthquakes that occur at 350-660 km are theorized to be caused by strain-induced olivine-spinel phase transformation (PT). We introduce and apply dynamic rotational diamond anvil cell with rough diamond anvils to deform San Carlos olivine. While olivine was never transformed to spinel at any pressure at room temperature, we obtained olivine-ringwoodite PT under severe plastic shear at 15-28 GPa within seconds. This is conceptual proof of the difference between pressure- and plastic strain-induced PTs and that plastic straining can accelerate this PT from million years to timescales relevant for the earthquake. The PT pressure linearly reduces with increasing plastic strain, corresponding increasing dislocation density and decreasing crystallite size. The main rules of the coupled severe plastic flow, PT, and microstructure evolution are found.

physics.geo-ph↗

arXiv2023

Idealizing Tauc Plot for Accurate Bandgap Determination of Semiconductor with UV-Vis: A Case Study for Cubic Boron Arsenide

The Tauc plot method is widely used to determine the bandgap of semiconductors via UV-visible optical spectroscopy due to its simplicity and perceived accuracy. However, the actual Tauc plot often exhibits significant baseline absorption below the expected bandgap, leading to discrepancies in the calculated bandgap depending on whether the linear fit is extrapolated to zero or non-zero baseline. In this study, we show that both extrapolation methods can produce significant errors by simulating Tauc plots with varying levels of baseline absorption. To address this issue, we propose a new method that involves idealizing the absorption spectrum by removing its baseline before constructing the Tauc plot. Experimental verification of this method using a gallium phosphide (GaP) wafer with intentionally introduced baseline absorptions shows promising results. Furthermore, we apply this new method to cubic boron arsenide (c-BAs) and resolve discrepancies in c-BAs bandgap values reported by different groups, obtaining a converging bandgap of 1.835 eV based on both previous and new transmission spectra. The method is applicable to both indirect and direct bandgap semiconductors, regardless of whether the absorption spectrum is measured via transmission or diffuse reflectance, will become essential to obtain accurate values of their bandgaps.

cond-mat.mtrl-sci↗

arXiv2023

Fast exact simulation of the first-passage event of a subordinator

This paper provides an exact simulation algorithm for the sampling from the joint law of the first-passage time, the undershoot and the overshoot of a subordinator crossing a non-increasing boundary. We prove that the running time of this algorithm has finite moments of all positive orders and give an explicit bound on the expected running time in terms of the Lévy measure of the subordinator. This bound provides performance guarantees that make our algorithm suitable for Monte Carlo estimation. We provide a GitHub repository with an implementation of the algorithm in Python and Julia.

math.PR↗

arXiv2023

In-situ study of rules of nanostructure evolution, severe plastic deformations, and friction under high pressure

Severe plastic deformations under high pressure are used to produce nanostructured materials but were studied ex-situ. We introduce rough diamond anvils to reach maximum friction equal to yield strength in shear and perform the first in-situ study of the evolution of the pressure-dependent yield strength and nanostructural parameters for severely pre-deformed Zr. ω-Zr behaves like perfectly plastic, isotropic, and strain-path-independent. This is related to reaching steady values of the crystallite size and dislocation density, which are pressure-, strain- and strain-path-independent. However, steady states for α-Zr obtained with smooth and rough anvils are different, which causes major challenge in plasticity theory.

cond-mat.mtrl-sci↗

arXiv2023

Rules of plastic strain-induced phase transformations and nanostructure evolution under high-pressure and severe plastic flow

Rough diamond anvils (rough-DA) are introduced to intensify all occurring processes during an in-situ study of heterogeneous compression of strongly pre-deformed Zr in diamond anvil cell (DAC). Crystallite size and dislocation density of Zr are getting pressure-, plastic strain tensor- and strain-path-independent during α-ω phase transformation (PT) and depend solely on the volume fraction of ω-Zr. Rough-DA produce a steady nanostructure in α-Zr with lower crystallite size and larger dislocation density than smooth-DA, leading to a two-time reduction in a minimum pressure for α-ω PT to a record value 0.67 GPa. The kinetics of strain-induced PT unexpectedly depends on time.

cond-mat.mtrl-sci↗

arXiv2023

Plastic strain-induced phase transformations in silicon: drastic reduction of transformation pressures, change in transformation sequence, and particle size effect

Pressure-induced phase transformations (PTs) between numerous phases of Si, the most important electronic material, have been studied for decades. This is not the case for plastic strain-induced PTs. Here, we revealed in-situ various unexpected plastic strain-induced PT phenomena. Thus, for 100 nm Si, strain-induced PT Si-I to Si-II (and Si-I to Si-III) initiates at 0.4 GPa (0.6 GPa) versus 16.2 GPa ($\infty$, since it does not occur) under hydrostatic conditions; for 30 nm Si, it is 6.1 GPa versus $\infty$. The predicted theoretical correlation between the direct and inverse Hall-Petch effect of the grain size on the yield strength and the minimum pressure for strain-induced PT is confirmed for the appearance of Si-II. Retaining Si-II at ambient pressure and obtaining reverse Si-II to Si-I PT are achieved, demonstrating the possibilities of manipulating different synthetic paths.

cond-mat.mtrl-sci↗