Shiyue Zhang, Mohit Bansal. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

While large language models (LLMs) have proven to be effective on a variety of tasks, they are also known hallucinate information. To measure whether an LLM prefers factually consistent continuations its input, we propose new benchmark called FIB (Factual Inconsistency Benchmark) that focuses the task summarization. Specifically, our involves comparing scores assigns versus inconsistent summary for input news article. For summaries, use human-written reference summaries manually verify as...

Jiyuan Zhang, Yang Feng, Dong Wang, Andrew Abel, Shiyue Andi Zhang. Proceedings of the 55th Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2017.

Data collection for natural language (NL) understanding tasks has increasingly included human explanations alongside data points, allowing past works to introduce models that both perform a task and generate NL their outputs. Yet date, model-generated have been evaluated on the basis of surface-level similarities explanations, through automatic metrics like BLEU evaluations. We argue these evaluations are insufficient, since they fail indicate whether support actual model behavior...

Recurrent neural networks (RNNs) have shown clear superiority in sequence modeling, particularly the ones with gated units, such as long short-term memory (LSTM) and recurrent unit (GRU). However, dynamic properties behind remarkable performance remain unclear many applications, e.g., automatic speech recognition (ASR). This paper employs visualization techniques to study behavior of LSTM GRU when performing tasks. Our experiments show some interesting patterns memory, them inspired simple...

Neural machine translation (NMT) has achieved notable success in recent times, however it is also widely recognized that this approach limitations with handling infrequent words and word pairs. This paper presents a novel memory-augmented NMT (M-NMT) architecture, which stores knowledge about how (usually infrequently encountered ones) should be translated memory then utilizes them to assist the neural model. We use mechanism combine learned from conventional statistical system rules by an...

Abstract Currently, much research effort has been devoted to improving the exciton utilization efficiency and narrowing emission spectra of ultraviolet (UV) fluorophores for organic light‐emitting diode (OLED) applications, while almost no attention paid optimizing their light out‐coupling efficiency. Here, we developed a linear donor‐acceptor‐donor (D–A–D) triad, namely CDFDB, which possesses high‐lying reverse intersystem crossing (hRISC) property. Thanks its integrated narrowband UV...

In view of the current and urgent environmental protection needs, use industrial solid waste in China’s Ningdong is becoming more important. this paper, NaP zeolite with good physical properties synthesized by using coal gasification coarse slag (CGCS) as raw material, without addition a silicon aluminum source, template agent, high-temperature calcination. Add small amount NaOH deionized water to CGCS adjust molar ratio SiO2:Al2O3:Na2O:H2O = 5.2:1.0:5.0:100. The effects aging time,...

Rock mass classification is important in preliminary design of geotechnical engineering projects. Using the columnar jointed basalt at foundation Baihetan Hydropower Station as an example, this paper presents a scheme rock. Unlike many common rock masses, obvious characteristic that it discontinuous geometry while continuous mechanics. Due to inapplicability existing systems, scheme, combined with integrity, weak plane tightness, and permeability, proposed. The new system has five grades...

With the development of very high resolution satellite image acquisition technology, remote sensing scene classification has become an important and challenging task. In this article, aiming at tackling task, we propose a hybrid architecture, i.e., aggregated deep Fisher feature (ADFF), which can make full use convolutional features' rich semantic information unsupervised encoding's robustness. Unlike previous methods, first explore optimal encoding layer in pretraining CNN model, naturally...

Shiyue Zhang, Asli Celikyilmaz, Jianfeng Gao, Mohit Bansal. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

The problems of unfaithful summaries have been widely discussed under the context abstractive summarization. Though extractive summarization is less prone to common unfaithfulness issues summaries, does that mean equal faithful? Turns out answer no. In this work, we define a typology with five types broad (including and beyond not-entailment) can appear in including incorrect coreference, incomplete discourse, as well other misleading information. We ask humans label these 1600 English...

More than 43% of the languages spoken in world are endangered, and language loss currently occurs at an accelerated rate because globalization neocolonialism. Saving revitalizing endangered has become very important for maintaining cultural diversity on our planet. In this work, we focus discussing how NLP can help revitalize languages. We first suggest three principles that may practitioners to foster mutual understanding collaboration with communities, discuss ways which potentially assist...

Near-infrared (NIR) organic light-emitting diodes (OLEDs) suffer from the low external electroluminescence (EL) quantum efficiency (EQE), which is a critical obstacle for potential applications. Herein, 1-oxo-1-phenalene-2,3-dicarbonitrile (OPDC) employed as an electron-withdrawing aromatic ring, and by incorporating with triphenylamine (TPA) biphenylphenylamine (BBPA) donors, two novel NIR emitters thermally activated delayed fluorescence (TADF) characteristics, namely OPDC-DTPA OPDC-DBBPA,...

Virtual try-on (VTON) technology has gained attention due to its potential transform online retail by enabling realistic clothing visualization of images and videos. However, most existing methods struggle achieve high-quality results across image video tasks, especially in long scenarios. In this work, we introduce CatV2TON, a simple effective vision-based virtual (V2TON) method that supports both tasks with single diffusion transformer model. By temporally concatenating garment person...

Building on the success of diffusion models, significant advancements have been made in multimodal image generation tasks. Among these, human has emerged as a promising technique, offering potential to revolutionize fashion design process. However, existing methods often focus solely text-to-image or reference-based generation, which fails satisfy increasingly sophisticated demands. To address limitations flexibility and precision we introduce ComposeAnyone, controllable layout-to-human...

While conditional diffusion models have achieved remarkable success in various applications, they require abundant data to train from scratch, which is often infeasible practice. To address this issue, transfer learning has emerged as an essential paradigm small regimes. Despite its empirical success, the theoretical underpinnings of remain unexplored. In paper, we take first step towards understanding sample efficiency through lens representation learning. Inspired by practical training...

Sea ice and snow are crucial components of the cryosphere climate system. Both sea spring in Northern Hemisphere (NH) have been decreasing at an alarming rate a changing climate. Changes NH linked with variety weather extremes including cold spells, heatwaves, droughts wildfires. Understanding these linkages will benefit predictions extremes. However, existing work on this has largely fragmented subject to large uncertainties physical pathways methodologies. This prevented further...

Adding another constituent into a binary system, known as ternary strategy, represents simple and effective approach to boosting the power conversion efficiency (PCE) of organic solar cells (OSCs). Herein, we have prepared new nonfused ring small-molecule acceptor with medium bandgap, named DFTQA-2FIC, which possesses high-lying lowest unoccupied molecular orbital energy level strong intramolecular charge-transfer effect. We elaborately utilized it third component in typical PM6:Y6 blend...

Recently, with the success of deep convolutional neural networks (CNNs), many end-to-end learning algorithms have yielded excellent results. However, in field land use and cover (LULC), very CNNs cannot be driven even tens thousands images. In contrast to transferring methods that only employ a model pretrained an irrelevant data set (e.g., ImageNet) directly inherit parameters without refining, we explore approach for effectively driving CNN small capacity. We propose novel concept called...