arXiv subjects

Yu Fu

Publications and source records attributed to Yu Fu.

At least 91 records · Page 5Linked to original sources

arXiv2024

Inducing superconductivity in quantum anomalous Hall regime

Interfacing the quantum anomalous Hall insulator with a conventional superconductor is known to be a promising manner for realizing a topological superconductor, which has been continuously pursued for years. Such a proximity route depends to a great extent on the control of the delicate interfacial coupling of the two constituents. However, a recent experiment reported the failure to reproduce such a topological superconductor, which is ascribed to the negligence of the electrical short by the superconductor in the theoretical proposal. Here, we reproduce this topological superconductor with attention to the interface control. The resulted conductance matrix under a wide magnetic field range agrees with the fingerprint of this topological superconductor. This allows us to develop a phase diagram that unveils three regions parameterized by various coupling limits, which not only supports the feasibility to fabricate the topological superconductor by proximity but also fully explains the origin of the previous debate. The present work provides a comprehensible guide on fabricating the topological superconductor.

cond-mat.mes-hall↗

arXiv2024

Algebraic Independence of Special Points on Shimura Varieties

Given a correspondence $V$ between a connected Shimura variety $S$, a commutative connected algebraic group $G$, and $n \in \mathbb{N}$, we prove that the $V$-images of any $n$ special points on $S$ outside a proper Zariski closed subset are algebraically independent. Our result unifies previous unlikely intersection results on multiplicative independence and linear independence. We prove multiplicative independence of differences of singular moduli, generalizing previous results by Pila-Tsimerman, and Aslanlyan-Eterović-Fowler. We also give an application to abelian varieties by proving that the special points of $S$ whose $V$-images lie in a finite-rank subgroup of $T$ are contained in a finite union of proper special subvarieties of $S$, only dependent on the rank of the subgroup. In this way, our result is a generalization of the works of Pila-Tsimerman and Buium-Poonen.

math.NT↗

arXiv2024

Safety Alignment in NLP Tasks: Weakly Aligned Summarization as an In-Context Attack

Recent developments in balancing the usefulness and safety of Large Language Models (LLMs) have raised a critical question: Are mainstream NLP tasks adequately aligned with safety consideration? Our study, focusing on safety-sensitive documents obtained through adversarial attacks, reveals significant disparities in the safety alignment of various NLP tasks. For instance, LLMs can effectively summarize malicious long documents but often refuse to translate them. This discrepancy highlights a previously unidentified vulnerability: attacks exploiting tasks with weaker safety alignment, like summarization, can potentially compromise the integrity of tasks traditionally deemed more robust, such as translation and question-answering (QA). Moreover, the concurrent use of multiple NLP tasks with lesser safety alignment increases the risk of LLMs inadvertently processing harmful content. We demonstrate these vulnerabilities in various safety-aligned LLMs, particularly Llama2 models, Gemini and GPT-4, indicating an urgent need for strengthening safety alignments across a broad spectrum of NLP tasks.

cs.CL↗

arXiv2024

Cross-Task Defense: Instruction-Tuning LLMs for Content Safety

Recent studies reveal that Large Language Models (LLMs) face challenges in balancing safety with utility, particularly when processing long texts for NLP tasks like summarization and translation. Despite defenses against malicious short questions, the ability of LLMs to safely handle dangerous long content, such as manuals teaching illicit activities, remains unclear. Our work aims to develop robust defenses for LLMs in processing malicious documents alongside benign NLP task queries. We introduce a defense dataset comprised of safety-related examples and propose single-task and mixed-task losses for instruction tuning. Our empirical results demonstrate that LLMs can significantly enhance their capacity to safely manage dangerous content with appropriate instruction tuning. Additionally, strengthening the defenses of tasks most susceptible to misuse is effective in protecting LLMs against processing harmful information. We also observe that trade-offs between utility and safety exist in defense strategies, where Llama2, utilizing our proposed approach, displays a significantly better balance compared to Llama1.

cs.CL↗

arXiv2024

Phasic Diversity Optimization for Population-Based Reinforcement Learning

Reviewing the previous work of diversity Rein-forcement Learning,diversity is often obtained via an augmented loss function,which requires a balance between reward and diversity.Generally,diversity optimization algorithms use Multi-armed Bandits algorithms to select the coefficient in the pre-defined space. However, the dynamic distribution of reward signals for MABs or the conflict between quality and diversity limits the performance of these methods. We introduce the Phasic Diversity Optimization (PDO) algorithm, a Population-Based Training framework that separates reward and diversity training into distinct phases instead of optimizing a multi-objective function. In the auxiliary phase, agents with poor performance diversified via determinants will not replace the better agents in the archive. The decoupling of reward and diversity allows us to use an aggressive diversity optimization in the auxiliary phase without performance degradation. Furthermore, we construct a dogfight scenario for aerial agents to demonstrate the practicality of the PDO algorithm. We introduce two implementations of PDO archive and conduct tests in the newly proposed adversarial dogfight and MuJoCo simulations. The results show that our proposed algorithm achieves better performance than baselines.

cs.LG↗

arXiv2024

Molecular Dynamics Simulations of Anisotropic Particles Accelerated by Neural-Net Predicted Interactions

Rigid bodies, made of smaller composite beads, are commonly used to simulate anisotropic particles with molecular dynamics or Monte Carlo methods. To accurately represent the particle shape and to obtain smooth and realistic effective pair interactions between two rigid bodies, each body may need to contain hundreds of spherical beads. Given an interacting pair of particles, traditional MD methods calculate the inter-body distances between the beads of the rigid bodies within a certain distance. For a system containing many anisotropic particles, distance calculations are computationally costly and limit the attainable system size and simulation time. However, the effective interaction between two rigid particles only depends on the distance between their center of masses and their relative orientation. Therefore, a function directly mapping the center of mass distance and orientation to the interaction energy between the two rigid bodies, would completely bypass inter-bead distance calculations. It is challenging to derive such a general function analytically for most non-spherical rigid bodies. We have trained neural nets, powerful tools to fit nonlinear functions to complex datasets, to achieve this task. The pair configuration is taken as input and the energy, forces and torques between two rigid particles are predicted directly. We show that MD simulations of cubes and cylinders performed with forces and torques obtained from the gradients of the energy neural-nets quantitatively match traditional simulations that uses composite rigid bodies. Both structural quantities and dynamic measures are in agreement, while achieving up to 23 times speed up over traditional molecular dynamics, depending on hardware and system size. The method can, in principle, be applied to any irregular shape with any pair interaction, provided that sufficient training data can be obtained.

cond-mat.soft↗

arXiv2024

Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark Remedy

To mitigate potential risks associated with language models, recent AI detection research proposes incorporating watermarks into machine-generated text through random vocabulary restrictions and utilizing this information for detection. While these watermarks only induce a slight deterioration in perplexity, our empirical investigation reveals a significant detriment to the performance of conditional text generation. To address this issue, we introduce a simple yet effective semantic-aware watermarking algorithm that considers the characteristics of conditional text generation and the input context. Experimental results demonstrate that our proposed method yields substantial improvements across various text generation models, including BART and Flan-T5, in tasks such as summarization and data-to-text generation while maintaining detection ability.

cs.CL↗

arXiv2024

Evaluating and Enhancing Large Language Models Performance in Domain-specific Medicine: Osteoarthritis Management with DocOA

The efficacy of large language models (LLMs) in domain-specific medicine, particularly for managing complex diseases such as osteoarthritis (OA), remains largely unexplored. This study focused on evaluating and enhancing the clinical capabilities of LLMs in specific domains, using osteoarthritis (OA) management as a case study. A domain specific benchmark framework was developed, which evaluate LLMs across a spectrum from domain-specific knowledge to clinical applications in real-world clinical scenarios. DocOA, a specialized LLM tailored for OA management that integrates retrieval-augmented generation (RAG) and instruction prompts, was developed. The study compared the performance of GPT-3.5, GPT-4, and a specialized assistant, DocOA, using objective and human evaluations. Results showed that general LLMs like GPT-3.5 and GPT-4 were less effective in the specialized domain of OA management, particularly in providing personalized treatment recommendations. However, DocOA showed significant improvements. This study introduces a novel benchmark framework which assesses the domain-specific abilities of LLMs in multiple aspects, highlights the limitations of generalized LLMs in clinical contexts, and demonstrates the potential of tailored approaches for developing domain-specific medical LLMs.

cs.CL↗

arXiv2023

Inverse Reinforcement Learning for Text Summarization

We introduce inverse reinforcement learning (IRL) as an effective paradigm for training abstractive summarization models, imitating human summarization behaviors. Our IRL model estimates the reward function using a suite of important sub-rewards for summarization and concurrently optimizes the policy network. Experimental results across datasets in different domains (CNN/DailyMail and WikiHow) and various model sizes (BART-base and BART-large) demonstrate the superiority of our proposed IRL model for summarization over MLE and RL baselines. The resulting summaries exhibit greater similarity to human-crafted gold references, outperforming MLE and RL baselines on metrics such as ROUGE, coverage, novelty, compression ratio, factuality, and human evaluations.

cs.CL↗

arXiv2023

Correspondence between Color Glass Condensate and High-Twist Formalism

The Color Glass Condensate (CGC) effective theory and the collinear factorization at high-twist (HT) are two well-known frameworks describing perturbative QCD multiple scatterings in nuclear media. It has long been recognized that these two formalisms have their own domain of validity in different kinematics regions. Taking direct photon production in proton-nucleus collisions as an example, we clarify for the first time the relation between CGC and HT at the level of a physical observable. We show that the CGC formalism beyond shock-wave approximation, and with the Landau-Pomeranchuk-Migdal interference effect is consistent with the HT formalism in the transition region where they overlap. Such a unified picture paves the way for mapping out the phase diagram of parton density in nuclear medium from dilute to dense region.

hep-ph↗

arXiv2023

Survey of Vulnerabilities in Large Language Models Revealed by Adversarial Attacks

Large Language Models (LLMs) are swiftly advancing in architecture and capability, and as they integrate more deeply into complex systems, the urgency to scrutinize their security properties grows. This paper surveys research in the emerging interdisciplinary field of adversarial attacks on LLMs, a subfield of trustworthy ML, combining the perspectives of Natural Language Processing and Security. Prior work has shown that even safety-aligned LLMs (via instruction tuning and reinforcement learning through human feedback) can be susceptible to adversarial attacks, which exploit weaknesses and mislead AI systems, as evidenced by the prevalence of `jailbreak' attacks on models like ChatGPT and Bard. In this survey, we first provide an overview of large language models, describe their safety alignment, and categorize existing research based on various learning structures: textual-only attacks, multi-modal attacks, and additional attack methods specifically targeting complex systems, such as federated learning or multi-agent systems. We also offer comprehensive remarks on works that focus on the fundamental sources of vulnerabilities and potential defenses. To make this field more accessible to newcomers, we present a systematic review of existing works, a structured typology of adversarial attack concepts, and additional resources, including slides for presentations on related topics at the 62nd Annual Meeting of the Association for Computational Linguistics (ACL'24).

cs.CL↗

arXiv2023

ConSlide: Asynchronous Hierarchical Interaction Transformer with Breakup-Reorganize Rehearsal for Continual Whole Slide Image Analysis

Whole slide image (WSI) analysis has become increasingly important in the medical imaging community, enabling automated and objective diagnosis, prognosis, and therapeutic-response prediction. However, in clinical practice, the ever-evolving environment hamper the utility of WSI analysis models. In this paper, we propose the FIRST continual learning framework for WSI analysis, named ConSlide, to tackle the challenges of enormous image size, utilization of hierarchical structure, and catastrophic forgetting by progressive model updating on multiple sequential datasets. Our framework contains three key components. The Hierarchical Interaction Transformer (HIT) is proposed to model and utilize the hierarchical structural knowledge of WSI. The Breakup-Reorganize (BuRo) rehearsal method is developed for WSI data replay with efficient region storing buffer and WSI reorganizing operation. The asynchronous updating mechanism is devised to encourage the network to learn generic and specific knowledge respectively during the replay stage, based on a nested cross-scale similarity learning (CSSL) module. We evaluated the proposed ConSlide on four public WSI datasets from TCGA projects. It performs best over other state-of-the-art methods with a fair WSI-based continual learning setting and achieves a better trade-off of the overall performance and forgetting on previous task

cs.CV↗

arXiv2023

Reducing the gap between streaming and non-streaming Transducer-based ASR by adaptive two-stage knowledge distillation

Transducer is one of the mainstream frameworks for streaming speech recognition. There is a performance gap between the streaming and non-streaming transducer models due to limited context. To reduce this gap, an effective way is to ensure that their hidden and output distributions are consistent, which can be achieved by hierarchical knowledge distillation. However, it is difficult to ensure the distribution consistency simultaneously because the learning of the output distribution depends on the hidden one. In this paper, we propose an adaptive two-stage knowledge distillation method consisting of hidden layer learning and output layer learning. In the former stage, we learn hidden representation with full context by applying mean square error loss function. In the latter stage, we design a power transformation based adaptive smoothness method to learn stable output distribution. It achieved 19\% relative reduction in word error rate, and a faster response for the first token compared with the original streaming model in LibriSpeech corpus.

cs.CL↗

arXiv2023

SFCNeXt: a simple fully convolutional network for effective brain age estimation with small sample size

Deep neural networks (DNN) have been designed to predict the chronological age of a healthy brain from T1-weighted magnetic resonance images (T1 MRIs), and the predicted brain age could serve as a valuable biomarker for the early detection of development-related or aging-related disorders. Recent DNN models for brain age estimations usually rely too much on large sample sizes and complex network structures for multi-stage feature refinement. However, in clinical application scenarios, researchers usually cannot obtain thousands or tens of thousands of MRIs in each data center for thorough training of these complex models. This paper proposes a simple fully convolutional network (SFCNeXt) for brain age estimation in small-sized cohorts with biased age distributions. The SFCNeXt consists of Single Pathway Encoded ConvNeXt (SPEC) and Hybrid Ranking Loss (HRL), aiming to estimate brain ages in a lightweight way with a sufficient exploration of MRI, age, and ranking features of each batch of subjects. Experimental results demonstrate the superiority and efficiency of our approach.

eess.IV↗

arXiv2023

On triharmonic hypersurfaces in space forms

In this paper we study triharmonic hypersurfaces immersed in a space form $N^{n+1}(c)$. We prove that any proper CMC triharmonic hypersurface in the sphere $\mathbb S^{n+1}$ has constant scalar curvature; any CMC triharmonic hypersurface in the hyperbolic space $\mathbb H^{n+1}$ is minimal. Moreover, we show that any CMC triharmonic hypersurface in the Euclidean space $\mathbb R^{n+1}$ is minimal provided that the multiplicity of the principal curvature zero is at most one. In particular, we are able to prove that every CMC triharmonic hypersurface in the Euclidean space $\mathbb R^{6}$ is minimal.These results extend some recent works due to Montaldo-Oniciuc-Ratto and Chen-Guan, and give affirmative answer to the generalized Chen's conjecture.

math.DG↗

arXiv2023

Bilinear magnetoresistance in HgTe topological insulator: opposite signs at opposite interfaces demonstrated by gate control

Spin-orbit effects appearing in topological insulators (TI) and at Rashba interfaces are currently revolutionizing how we can manipulate spins and have led to several newly discovered effects, from spin-charge interconversion and spin-orbit torques to novel magnetoresistance phenomena. In particular, a puzzling magnetoresistance has been evidenced, bilinear in electric and magnetic fields. Here, we report the observation of bilinear magnetoresistance (BMR) in strained HgTe, a prototypical TI. We show that both the amplitude and sign of this BMR can be tuned by controlling, with an electric gate, the relative proportions of the opposite contributions of opposite surfaces. At magnetic fields of 1 T, the magnetoresistance is of the order of 1 \% and has a larger figure of merit than previously measured TIs. We propose a theoretical model giving a quantitative account of our experimental data. This phenomenon, unique to TI, offers novel opportunities to tune their electrical response for spintronics.

cond-mat.mes-hall↗

arXiv2023

On the biharmonic hypersurfaces with three distinct principal curvatures in space forms

In [16] there was proved that any biharmonic hypersurface with at most three distinct principal curvatures in space forms has constant mean curvature. At the very last step of the proof, the argument relied on the fact that the resultant of two polynomials is a non-zero polynomial. In this paper we point out that, in fact, there is a case, and only one, when this resultant is the zero polynomial and therefore the original proof is not fully complete. Further, we prove that in this special case we still obtain that the hypersurface has constant mean curvature.

math.DG↗

arXiv2022

Artificial Intelligence Security Competition (AISC)

The security of artificial intelligence (AI) is an important research area towards safe, reliable, and trustworthy AI systems. To accelerate the research on AI security, the Artificial Intelligence Security Competition (AISC) was organized by the Zhongguancun Laboratory, China Industrial Control Systems Cyber Emergency Response Team, Institute for Artificial Intelligence, Tsinghua University, and RealAI as part of the Zhongguancun International Frontier Technology Innovation Competition (https://www.zgc-aisc.com/en). The competition consists of three tracks, including Deepfake Security Competition, Autonomous Driving Security Competition, and Face Recognition Security Competition. This report will introduce the competition rules of these three tracks and the solutions of top-ranking teams in each track.

cs.CR↗