arXiv subjects

Yulin Li

Publications and source records attributed to Yulin Li.

65 records · Page 4Linked to original sources

arXiv2023

Fast-StrucTexT: An Efficient Hourglass Transformer with Modality-guided Dynamic Token Merge for Document Understanding

Transformers achieve promising performance in document understanding because of their high effectiveness and still suffer from quadratic computational complexity dependency on the sequence length. General efficient transformers are challenging to be directly adapted to model document. They are unable to handle the layout representation in documents, e.g. word, line and paragraph, on different granularity levels and seem hard to achieve a good trade-off between efficiency and performance. To tackle the concerns, we propose Fast-StrucTexT, an efficient multi-modal framework based on the StrucTexT algorithm with an hourglass transformer architecture, for visual document understanding. Specifically, we design a modality-guided dynamic token merging block to make the model learn multi-granularity representation and prunes redundant tokens. Additionally, we present a multi-modal interaction module called Symmetry Cross Attention (SCA) to consider multi-modal fusion and efficiently guide the token mergence. The SCA allows one modality input as query to calculate cross attention with another modality in a dual phase. Extensive experiments on FUNSD, SROIE, and CORD datasets demonstrate that our model achieves the state-of-the-art performance and almost 1.9X faster inference time than the state-of-the-art methods.

cs.CV↗

arXiv2023

Quantaloidal Completions of Order-enriched Categories and Their Applications

By introducing the concept of quantaloidal completions for an order-enriched category, relationships between the category of quantaloids and the category of order-enriched categories are studied. It is proved that quantaloidal completions for an order-enriched category can be fully characterized as compatible quotients of the power-set completion. As applications, we show that a special type of injective hull of an order-enriched category is the MacNeille completion; the free quantaloid over an order-enriched category is the Down-set completion.

math.CT↗

arXiv2023

StrucTexTv2: Masked Visual-Textual Prediction for Document Image Pre-training

In this paper, we present StrucTexTv2, an effective document image pre-training framework, by performing masked visual-textual prediction. It consists of two self-supervised pre-training tasks: masked image modeling and masked language modeling, based on text region-level image masking. The proposed method randomly masks some image regions according to the bounding box coordinates of text words. The objectives of our pre-training tasks are reconstructing the pixels of masked image regions and the corresponding masked tokens simultaneously. Hence the pre-trained encoder can capture more textual semantics in comparison to the masked image modeling that usually predicts the masked image patches. Compared to the masked multi-modal modeling methods for document image understanding that rely on both the image and text modalities, StrucTexTv2 models image-only input and potentially deals with more application scenarios free from OCR pre-processing. Extensive experiments on mainstream benchmarks of document image understanding demonstrate the effectiveness of StrucTexTv2. It achieves competitive or even new state-of-the-art performance in various downstream tasks such as image classification, layout analysis, table structure recognition, document OCR, and information extraction under the end-to-end scenario.

cs.CV↗

arXiv2022

Neural-iLQR: A Learning-Aided Shooting Method for Trajectory Optimization

Iterative linear quadratic regulator (iLQR) has gained wide popularity in addressing trajectory optimization problems with nonlinear system models. However, as a model-based shooting method, it relies heavily on an accurate system model to update the optimal control actions and the trajectory determined with forward integration, thus becoming vulnerable to inevitable model inaccuracies. Recently, substantial research efforts in learning-based methods for optimal control problems have been progressing significantly in addressing unknown system models, particularly when the system has complex interactions with the environment. Yet a deep neural network is normally required to fit substantial scale of sampling data. In this work, we present Neural-iLQR, a learning-aided shooting method over the unconstrained control space, in which a neural network with a simple structure is used to represent the local system model. In this framework, the trajectory optimization task is achieved with simultaneous refinement of the optimal policy and the neural network iteratively, without relying on the prior knowledge of the system model. Through comprehensive evaluations on two illustrative control tasks, the proposed method is shown to outperform the conventional iLQR significantly in the presence of inaccuracies in system models.

cs.LG↗

arXiv2021

StrucTexT: Structured Text Understanding with Multi-Modal Transformers

Structured text understanding on Visually Rich Documents (VRDs) is a crucial part of Document Intelligence. Due to the complexity of content and layout in VRDs, structured text understanding has been a challenging task. Most existing studies decoupled this problem into two sub-tasks: entity labeling and entity linking, which require an entire understanding of the context of documents at both token and segment levels. However, little work has been concerned with the solutions that efficiently extract the structured data from different levels. This paper proposes a unified framework named StrucTexT, which is flexible and effective for handling both sub-tasks. Specifically, based on the transformer, we introduce a segment-token aligned encoder to deal with the entity labeling and entity linking tasks at different levels of granularity. Moreover, we design a novel pre-training strategy with three self-supervised tasks to learn a richer representation. StrucTexT uses the existing Masked Visual Language Modeling task and the new Sentence Length Prediction and Paired Boxes Direction tasks to incorporate the multi-modal information across text, image, and layout. We evaluate our method for structured text understanding at segment-level and token-level and show it outperforms the state-of-the-art counterparts with significantly superior performance on the FUNSD, SROIE, and EPHOIE datasets.

cs.CV↗

arXiv2021

Diverse Part Discovery: Occluded Person Re-identification with Part-Aware Transformer

Occluded person re-identification (Re-ID) is a challenging task as persons are frequently occluded by various obstacles or other persons, especially in the crowd scenario. To address these issues, we propose a novel end-to-end Part-Aware Transformer (PAT) for occluded person Re-ID through diverse part discovery via a transformer encoderdecoder architecture, including a pixel context based transformer encoder and a part prototype based transformer decoder. The proposed PAT model enjoys several merits. First, to the best of our knowledge, this is the first work to exploit the transformer encoder-decoder architecture for occluded person Re-ID in a unified deep model. Second, to learn part prototypes well with only identity labels, we design two effective mechanisms including part diversity and part discriminability. Consequently, we can achieve diverse part discovery for occluded person Re-ID in a weakly supervised manner. Extensive experimental results on six challenging benchmarks for three tasks (occluded, partial and holistic Re-ID) demonstrate that our proposed PAT performs favorably against stat-of-the-art methods.

cs.CV↗

arXiv2014

Shielded button electrodes for time-resolved measurements of electron cloud buildup

We report on the design, deployment and signal analysis for shielded button electrodes sensitive to electron cloud buildup at the Cornell Electron Storage Ring. These simple detectors, derived from a beam-position monitor electrode design, have provided detailed information on the physical processes underlying the local production and lifetime of electron densities in the storage ring. Digitizing oscilloscopes are used to record electron fluxes incident on the vacuum chamber wall in 1024 time steps of 100 ps or more. The fine time steps provide a detailed characterization of the cloud, allowing the independent estimation of processes contributing on differing time scales and providing sensitivity to the characteristic kinetic energies of the electrons making up the cloud. By varying the spacing and population of electron and positron beam bunches, we map the time development of the various cloud production and re-absorption processes. The excellent reproducibility of the measurements also permits the measurement of long-term conditioning of vacuum chamber surfaces.

physics.acc-ph↗

arXiv2013

Demonstration of Low Emittance in the Cornell Energy Recovery Linac Injector Prototype

We present a detailed study of the six-dimensional phase space of the electron beam produced by the Cornell Energy Recovery Linac Photoinjector, a high-brightness, high repetition rate (1.3 GHz) DC photoemission source designed to drive a hard x-ray energy recovery linac (ERL). A complete simulation model of the injector has been constructed, verified by measurement, and optimized. Both the horizontal and vertical 2D transverse phase spaces, as well as the time-resolved (sliced) horizontal phase space, were simulated and directly measured at the end of the injector for 19 pC and 77 pC bunches at roughly 8 MeV. These bunch charges were chosen because they correspond to 25 mA and 100 mA average current if operating at the full 1.3 GHz repetition rate. The resulting 90% normalized transverse emittances for 19 (77) pC/bunch were 0.23 +/- 0.02 (0.51 +/- 0.04) microns in the horizontal plane, and 0.14 +/- 0.01 (0.29 +/- 0.02) microns in the vertical plane, respectively. These emittances were measured with a corresponding bunch length of 2.1 +/- 0.1 (3.0 +/- 0.2) ps, respectively. In each case the rms momentum spread was determined to be on the order of 1e-3. Excellent overall agreement between measurement and simulation has been demonstrated. Using the emittances and bunch length measured at 19 pC/bunch, we estimate the electron beam quality in a 1.3 GHz, 5 GeV hard x-ray ERL to be at least a factor of 20 times better than that of existing storage rings when the rms energy spread of each device is considered. These results represent a milestone for the field of high-brightness, high-current photoinjectors.

physics.acc-ph↗

arXiv2011

Photocathode Behavior During High Current Running in the Cornell ERL Photoinjector

The Cornell University Energy Recovery Linac (ERL) photoinjector has recently demonstrated operation at 20 mA for approximately 8 hours, utilizing a multialkali photocathode deposited on a Si substrate. We describe the recipe for photocathode deposition, and will detail the parameters of the run. Post-run analysis of the photocathode indicates the presence of significant damage to the substrate, perhaps due to ion back-bombardment from the residual beamline gas. While the exact cause of the substrate damage remains unknown, we describe multiple surface characterization techniques (X-ray fluorescence spectroscopy, X-ray diffraction, atomic force and scanning electron microscopy) used to study the interesting morphological and crystallographic features of the photocathode surface after its use for high current beam production. Finally, we present a simple model of crystal damage due to ion back-bombardment, which agrees qualitatively with the distribution of damage on the substrate surface.

physics.acc-ph↗

arXiv2011

Thermal emittance measurements of a cesium potassium antimonide photocathode

Thermal emittance measurements of a CsK2Sb photocathode at several laser wavelengths are presented. The emittance is obtained with a solenoid scan technique using a high voltage dc photoemission gun. The thermal emittance is 0.56+/-0.03 mm-mrad/mm(rms) at 532 nm wavelength. The results are compared with a simple photoemission model and found to be in a good agreement.

physics.acc-ph↗

arXiv2008

Benchmarking of 3D space charge codes using direct phase space measurements from photoemission high voltage DC gun

We present a comparison between space charge calculations and direct measurements of the transverse phase space for space charge dominated electron bunches after a high voltage photoemission DC gun followed by an emittance compensation solenoid magnet. The measurements were performed using a double-slit setup for a set of parameters such as charge per bunch and the solenoid current. The data is compared with detailed simulations using 3D space charge codes GPT and Parmela3D with initial particle distributions created from the measured transverse and temporal laser profiles. Beam brightness as a function of beam fraction is calculated for the measured phase space maps and found to approach the theoretical maximum set by the thermal energy and accelerating field at the photocathode.

physics.acc-ph↗