Search

Scientific document summarization has been a challenging task due to the long structure of the input text. The long input hinders the simultaneous effective modeling of both global high-order relations between sentences and local intra-sentence relations which is the most critical step in extractive summarization. However, existing methods mostly focus on one type of relation, neglecting the simultaneous effective modeling of both relations, which can lead to insufficient learning of semantic representations. In this paper, we propose HAESum, a novel approach utilizing graph neural networks to locally and globally model documents based on their hierarchical discourse structure. First, intra-sentence relations are learned using a local heterogeneous graph. Subsequently, a novel hypergraph self-attention layer is introduced to further enhance the characterization of high-order inter-sentence relations. We validate our approach on two benchmark datasets, and the experimental results demonstrate the effectiveness of HAESum and the importance of considering hierarchical structures in modeling long scientific documents. Our code will be available at \url{https://github.com/MoLICHENXI/HAESum}, Comment: Accepted to NAACL 2024 Findings

Prompt-based learning paradigm has demonstrated remarkable efficacy in enhancing the adaptability of pretrained language models (PLMs), particularly in few-shot scenarios. However, this learning paradigm has been shown to be vulnerable to backdoor attacks. The current clean-label attack, employing a specific prompt as a trigger, can achieve success without the need for external triggers and ensure correct labeling of poisoned samples, which is more stealthy compared to the poisoned-label attack, but on the other hand, it faces significant issues with false activations and poses greater challenges, necessitating a higher rate of poisoning. Using conventional negative data augmentation methods, we discovered that it is challenging to trade off between effectiveness and stealthiness in a clean-label setting. In addressing this issue, we are inspired by the notion that a backdoor acts as a shortcut and posit that this shortcut stems from the contrast between the trigger and the data utilized for poisoning. In this study, we propose a method named Contrastive Shortcut Injection (CSI), by leveraging activation values, integrates trigger design and data selection strategies to craft stronger shortcut features. With extensive experiments on full-shot and few-shot text classification tasks, we empirically validate CSI's high effectiveness and high stealthiness at low poisoning rates. Notably, we found that the two approaches play leading roles in full-shot and few-shot settings, respectively., Comment: 10 pages, 6 figures, conference

The acousto-optic interaction known as stimulated Brillouin scattering (SBS) has emerged as fundamental principles for realizing crucial components and functionalities in integrated photonics. However, the main challenge of integrated Brillouin devices is how to effectively confine both optical and acoustic waves. Apart from that, the manufacturing processes for these devices need to be compatible with standard fabrication platforms, and streamlined to facilitate their large-scale integration. Here, we demonstrate a novel suspended nanowire structure that can tightly confine photons and phonons. Furthermore, tailored for this structure, we introduce a loading-effect-based three-dimensional microfabrication technique, compatible with complementary metal-oxide-semiconductor (CMOS) technology. This innovative technique allows for the fabrication of the entire structure using a single-step lithography exposure, significantly streamlining the fabrication process. Leveraging this structure and fabrication scheme, we have achieved a Brillouin gain coefficient of 1100 1/W/m on the silicon-on-insulator platform within a compact footprint. It can support a Brillouin net gain over 4.1 dB with modest pump powers. We believe that this structure can significantly advance the development of SBS on chip, unlocking new opportunities for the large-scale integration of Brillouin-based photonic devices., Comment: 5 pages, 4 figures

Empowering independent control of optical and acoustic modes and enhancing the photon-phonon interaction, integrated photonics boosts the advancements of on-chip stimulated Brillouin scattering (SBS). However, achieving acoustic waveguides with low loss, tailorability, and easy fabrication remains a challenge. Here, inspired by the optical anti-resonance in hollow-core fibers, we propose suspended anti-resonant acoustic waveguides (SARAWs) with superior confinement and high selectivity of acoustic modes, supporting both forward and backward SBS on chip. Furthermore, this structure streamlines the design and fabrication processes. Leveraging the advantages of SARAWs, we have showcased a series of record-breaking results for SBS within a compact footprint on the silicon-on-insulator platform. For forward SBS, a centimeter-scale SARAW supports a large net gain exceeding 6.4 dB. For backward SBS, we have observed an unprecedented Brillouin frequency shift of 27.6 GHz and a mechanical quality factor of up to 1,960 in silicon waveguides. This paradigm of acoustic waveguide propels SBS into a new era, unlocking new opportunities in the fields of optomechanics, phononic circuits, and hybrid quantum systems., Comment: 18 pages, 13 figures

Phase noise or frequency noise is a key metrics to evaluate the short term stability of a laser. This property is of a great interest for the applications but delicate to characterize, especially for narrow line-width lasers. In this letter, we demonstrate a digital cross correlation scheme to characterize the absolute phase noise of sub-hertz line-width lasers. Three 1,542 nm ultra-stable lasers are used in this approach. For each measurement two lasers act as references to characterize a third one. Phase noise power spectral density from 0.5 Hz to 0.8 MHz Fourier frequencies can be derived for each laser by a mere change in the configuration of the lasers. This is the first time showing the phase noise of sub-hertz line-width lasers with no reference limitation. We also present an analysis of the laser phase noise performance., Comment: 4 pages, 5 figures

Photonic synthesis of radiofrequency revived the quest for unrivalled microwave purity by its seducing ability to convey the benefits of the optics to the microwave world. In this work, we perform a high-fidelity transfer of frequency stability between an optical reference and a microwave signal via a low-noise fiber-based frequency comb and cutting-edge photo-detection techniques. We demonstrate the generation of the purest microwave signal with a fractional frequency stability below 6.5 x 10^-16 at 1 s and a timing noise floor below 41 zs.Hz^-1/2 (phase noise below -173 dBc.Hz^-1 for a 12 GHz carrier). This outclasses existing sources and promises a new era for state-of-the-art microwave generation. The characterization is achieved by building two auxiliary low noise optoelectronic microwave reference and using a heterodyne cross-correlation scheme with lowermost detection noise. This unprecedented level of purity can impact domains such as radar systems, telecommunications and time-frequency metrology. Furthermore, the measurement methods developed here can benefit the characterization of a broad range of signals, beyond comb-based systems., Comment: 21 pages, 3 figures