Searcharxiv⌕ Search

arXiv subjects

Jian Wu

Publications and source records attributed to Jian Wu.

At least 235 records · Page 13Linked to original sources
arXiv2023

DeepPatent2: A Large-Scale Benchmarking Corpus for Technical Drawing Understanding

Recent advances in computer vision (CV) and natural language processing have been driven by exploiting big data on practical applications. However, these research fields are still limited by the sheer volume, versatility, and diversity of the available datasets. CV tasks, such as image captioning, which has primarily been carried out on natural images, still struggle to produce accurate and meaningful captions on sketched images often included in scientific and technical documents. The advancement of other tasks such as 3D reconstruction from 2D images requires larger datasets with multiple viewpoints. We introduce DeepPatent2, a large-scale dataset, providing more than 2.7 million technical drawings with 132,890 object names and 22,394 viewpoints extracted from 14 years of US design patent documents. We demonstrate the usefulness of DeepPatent2 with conceptual captioning. We further provide the potential usefulness of our dataset to facilitate other research areas such as 3D image reconstruction and image retrieval.

cs.CV↗
arXiv2023

Fast Model Debias with Machine Unlearning

Recent discoveries have revealed that deep neural networks might behave in a biased manner in many real-world scenarios. For instance, deep networks trained on a large-scale face recognition dataset CelebA tend to predict blonde hair for females and black hair for males. Such biases not only jeopardize the robustness of models but also perpetuate and amplify social biases, which is especially concerning for automated decision-making processes in healthcare, recruitment, etc., as they could exacerbate unfair economic and social inequalities among different groups. Existing debiasing methods suffer from high costs in bias labeling or model re-training, while also exhibiting a deficiency in terms of elucidating the origins of biases within the model. To this respect, we propose a fast model debiasing framework (FMD) which offers an efficient approach to identify, evaluate and remove biases inherent in trained models. The FMD identifies biased attributes through an explicit counterfactual concept and quantifies the influence of data samples with influence functions. Moreover, we design a machine unlearning-based strategy to efficiently and effectively remove the bias in a trained model with a small counterfactual dataset. Experiments on the Colored MNIST, CelebA, and Adult Income datasets along with experiments with large language models demonstrate that our method achieves superior or competing accuracies compared with state-of-the-art methods while attaining significantly fewer biases and requiring much less debiasing cost. Notably, our method requires only a small external dataset and updating a minimal amount of model parameters, without the requirement of access to training data that may be too large or unavailable in practice.

cs.LG↗
arXiv2023

Sample-efficient Multi-objective Molecular Optimization with GFlowNets

Many crucial scientific problems involve designing novel molecules with desired properties, which can be formulated as a black-box optimization problem over the discrete chemical space. In practice, multiple conflicting objectives and costly evaluations (e.g., wet-lab experiments) make the diversity of candidates paramount. Computational methods have achieved initial success but still struggle with considering diversity in both objective and search space. To fill this gap, we propose a multi-objective Bayesian optimization (MOBO) algorithm leveraging the hypernetwork-based GFlowNets (HN-GFN) as an acquisition function optimizer, with the purpose of sampling a diverse batch of candidate molecular graphs from an approximate Pareto front. Using a single preference-conditioned hypernetwork, HN-GFN learns to explore various trade-offs between objectives. We further propose a hindsight-like off-policy strategy to share high-performing molecules among different preferences in order to speed up learning for HN-GFN. We empirically illustrate that HN-GFN has adequate capacity to generalize over preferences. Moreover, experiments in various real-world MOBO settings demonstrate that our framework predominantly outperforms existing methods in terms of candidate quality and sample efficiency. The code is available at https://github.com/violet-sto/HN-GFN.

cs.LG↗
arXiv2023

Maximizing Equitable Reach and Accessibility of ETDs

This poster addresses accessibility issues of electronic theses and dissertations (ETDs) in digital libraries (DLs). ETDs are available primarily as PDF files, which present barriers to equitable access, especially for users with visual impairments, cognitive or learning disabilities, or for anyone needing more efficient and effective ways of finding relevant information within these long documents. We propose using AI techniques, including natural language processing (NLP), computer vision, and text analysis, to convert PDFs into machine-readable HTML documents with semantic tags and structure, extracting figures and tables, and generating summaries and keywords. Our goal is to increase the accessibility of ETDs and to make this important scholarship available to a wider audience.

cs.DL↗
arXiv2023

Design of the LLRF control system for MA cavity at CSNS RCS

The China Spallation Neutron Source (CSNS) beam power was successfully reached 125 kW with a low beam loss in February 2022. In order to increase beam power, during the summer in 2022, we employ magnetic-alloy (MA) cavity in the rapid cycling synchrotron (RCS). It is a wideband cavity (Q=2), allows the second harmonic rf (h= 4) operation, with the existing ferrite cavity to realize the dual-harmonic acceleration. The second harmonic (h=4) is used for the bunch shape control and alleviating the space charge effects. We design of the low-level RF(LLRF) control system for MA cavity, in this paper, we describe the system design and implementation, and the preliminary test results.

physics.acc-ph↗
arXiv2023

Modeling Lead-vehicle Kinematics For Rear-end Crash Scenario Generation

The use of virtual safety assessment as the primary method for evaluating vehicle safety technologies has emphasized the importance of crash scenario generation. One of the most common crash types is the rear-end crash, which involves a lead vehicle and a following vehicle. Most studies have focused on the following vehicle, assuming that the lead vehicle maintains a constant acceleration/deceleration before the crash. However, there is no evidence for this premise in the literature. This study aims to address this knowledge gap by thoroughly analyzing and modeling the lead vehicle's behavior as a first step in generating rear-end crash scenarios. Accordingly, the study employed a piecewise linear model to parameterize the speed profiles of lead vehicles, utilizing two rear-end pre-crash/near-crash datasets. These datasets were merged and categorized into multiple sub-datasets; for each one, a multivariate distribution was constructed to represent the corresponding parameters. Subsequently, a synthetic dataset was generated using these distribution models and validated by comparison with the original combined dataset. The results highlight diverse lead-vehicle speed patterns, indicating that a more accurate model, such as the proposed piecewise linear model, is required instead of the conventional constant acceleration/deceleration model. Crashes generated with the proposed models accurately match crash data across the full severity range, surpassing existing lead-vehicle kinematics models in both severity range and accuracy. By providing more realistic speed profiles for the lead vehicle, the model developed in the study contributes to creating realistic rear-end crash scenarios and reconstructing real-life crashes.

cs.RO↗
arXiv2023

VisionFM: a Multi-Modal Multi-Task Vision Foundation Model for Generalist Ophthalmic Artificial Intelligence

We present VisionFM, a foundation model pre-trained with 3.4 million ophthalmic images from 560,457 individuals, covering a broad range of ophthalmic diseases, modalities, imaging devices, and demography. After pre-training, VisionFM provides a foundation to foster multiple ophthalmic artificial intelligence (AI) applications, such as disease screening and diagnosis, disease prognosis, subclassification of disease phenotype, and systemic biomarker and disease prediction, with each application enhanced with expert-level intelligence and accuracy. The generalist intelligence of VisionFM outperformed ophthalmologists with basic and intermediate levels in jointly diagnosing 12 common ophthalmic diseases. Evaluated on a new large-scale ophthalmic disease diagnosis benchmark database, as well as a new large-scale segmentation and detection benchmark database, VisionFM outperformed strong baseline deep neural networks. The ophthalmic image representations learned by VisionFM exhibited noteworthy explainability, and demonstrated strong generalizability to new ophthalmic modalities, disease spectrum, and imaging devices. As a foundation model, VisionFM has a large capacity to learn from diverse ophthalmic imaging data and disparate datasets. To be commensurate with this capacity, in addition to the real data used for pre-training, we also generated and leveraged synthetic ophthalmic imaging data. Experimental results revealed that synthetic data that passed visual Turing tests, can also enhance the representation learning capability of VisionFM, leading to substantial performance gains on downstream ophthalmic AI tasks. Beyond the ophthalmic AI applications developed, validated, and demonstrated in this work, substantial further applications can be achieved in an efficient and cost-effective manner using VisionFM as the foundation.

eess.IV↗
arXiv2023

The second multi-channel multi-party meeting transcription challenge (M2MeT) 2.0): A benchmark for speaker-attributed ASR

With the success of the first Multi-channel Multi-party Meeting Transcription challenge (M2MeT), the second M2MeT challenge (M2MeT 2.0) held in ASRU2023 particularly aims to tackle the complex task of \emph{speaker-attributed ASR (SA-ASR)}, which directly addresses the practical and challenging problem of ``who spoke what at when" at typical meeting scenario. We particularly established two sub-tracks. The fixed training condition sub-track, where the training data is constrained to predetermined datasets, but participants can use any open-source pre-trained model. The open training condition sub-track, which allows for the use of all available data and models without limitation. In addition, we release a new 10-hour test set for challenge ranking. This paper provides an overview of the dataset, track settings, results, and analysis of submitted systems, as a benchmark to show the current state of speaker-attributed ASR.

cs.SD↗
arXiv2023

On decoder-only architecture for speech-to-text and large language model integration

Large language models (LLMs) have achieved remarkable success in the field of natural language processing, enabling better human-computer interaction using natural language. However, the seamless integration of speech signals into LLMs has not been explored well. The "decoder-only" architecture has also not been well studied for speech processing tasks. In this research, we introduce Speech-LLaMA, a novel approach that effectively incorporates acoustic information into text-based large language models. Our method leverages Connectionist Temporal Classification and a simple audio encoder to map the compressed acoustic features to the continuous semantic space of the LLM. In addition, we further probe the decoder-only architecture for speech-to-text tasks by training a smaller scale randomly initialized speech-LLaMA model from speech-text paired data alone. We conduct experiments on multilingual speech-to-text translation tasks and demonstrate a significant improvement over strong baselines, highlighting the potential advantages of decoder-only models for speech-to-text conversion.

eess.AS↗
arXiv2023

OneSeg: Self-learning and One-shot Learning based Single-slice Annotation for 3D Medical Image Segmentation

As deep learning methods continue to improve medical image segmentation performance, data annotation is still a big bottleneck due to the labor-intensive and time-consuming burden on medical experts, especially for 3D images. To significantly reduce annotation efforts while attaining competitive segmentation accuracy, we propose a self-learning and one-shot learning based framework for 3D medical image segmentation by annotating only one slice of each 3D image. Our approach takes two steps: (1) self-learning of a reconstruction network to learn semantic correspondence among 2D slices within 3D images, and (2) representative selection of single slices for one-shot manual annotation and propagating the annotated data with the well-trained reconstruction network. Extensive experiments verify that our new framework achieves comparable performance with less than 1% annotated data compared with fully supervised methods and generalizes well on several out-of-distribution testing sets.

cs.CV↗
arXiv2023

GCL: Gradient-Guided Contrastive Learning for Medical Image Segmentation with Multi-Perspective Meta Labels

Since annotating medical images for segmentation tasks commonly incurs expensive costs, it is highly desirable to design an annotation-efficient method to alleviate the annotation burden. Recently, contrastive learning has exhibited a great potential in learning robust representations to boost downstream tasks with limited labels. In medical imaging scenarios, ready-made meta labels (i.e., specific attribute information of medical images) inherently reveal semantic relationships among images, which have been used to define positive pairs in previous work. However, the multi-perspective semantics revealed by various meta labels are usually incompatible and can incur intractable "semantic contradiction" when combining different meta labels. In this paper, we tackle the issue of "semantic contradiction" in a gradient-guided manner using our proposed Gradient Mitigator method, which systematically unifies multi-perspective meta labels to enable a pre-trained model to attain a better high-level semantic recognition ability. Moreover, we emphasize that the fine-grained discrimination ability is vital for segmentation-oriented pre-training, and develop a novel method called Gradient Filter to dynamically screen pixel pairs with the most discriminating power based on the magnitude of gradients. Comprehensive experiments on four medical image segmentation datasets verify that our new method GCL: (1) learns informative image representations and considerably boosts segmentation performance with limited labels, and (2) shows promising generalizability on out-of-distribution datasets.

cs.CV↗
arXiv2023

t-SOT FNT: Streaming Multi-talker ASR with Text-only Domain Adaptation Capability

Token-level serialized output training (t-SOT) was recently proposed to address the challenge of streaming multi-talker automatic speech recognition (ASR). T-SOT effectively handles overlapped speech by representing multi-talker transcriptions as a single token stream with $\langle \text{cc}\rangle$ symbols interspersed. However, the use of a naive neural transducer architecture significantly constrained its applicability for text-only adaptation. To overcome this limitation, we propose a novel t-SOT model structure that incorporates the idea of factorized neural transducers (FNT). The proposed method separates a language model (LM) from the transducer's predictor and handles the unnatural token order resulting from the use of $\langle \text{cc}\rangle$ symbols in t-SOT. We achieve this by maintaining multiple hidden states and introducing special handling of the $\langle \text{cc}\rangle$ tokens within the LM. The proposed t-SOT FNT model achieves comparable performance to the original t-SOT model while retaining the ability to reduce word error rate (WER) on both single and multi-talker datasets through text-only adaptation.

eess.AS↗
arXiv2023

Bose condensation of upper-branch exciton-polaritons in a transferrable microcavity

Exciton-polaritons are composite bosonic quasiparticles arising from the strong coupling of excitonic transitions and optical modes. Exciton-polaritons have triggered wide exploration in the past decades not only due to their rich quantum phenomena such as superfluidity, superconductivity and quantized vortices but also due to their potential applications for unconventional coherent light sources and all-optical control elements. Here, we report the observation of Bose-Einstein condensation of the upper polariton branch in a transferrable WS$_2$ monolayer microcavity. Near the condensation threshold, we observe a nonlinear increase in upper polariton intensity. This sharp increase in intensity is accompanied by a decrease of the linewidth and an increase of the upper polariton temporal coherence, all of which are hallmarks of Bose-Einstein condensation. By simulating the quantum Boltzmann equation, we show that the upper polariton condensation only occurs for a particular range of particle density. We can attribute the creation of Bose condensation of the upper polariton to the following requirements: 1) the upper polariton is more excitonic than the lower one; 2) there is relatively more pumping in the upper branch; and 3) the conversion time from the upper to the lower polariton branch is long compared to the lifetime of the upper polaritons.

cond-mat.quant-gas↗
arXiv2023

Integrated photonics modular arithmetic processor

Integrated photonics computing has emerged as a promising approach to overcome the limitations of electronic processors in the post-Moore era, capitalizing on the superiority of photonic systems. However, present integrated photonics computing systems face challenges in achieving high-precision calculations, consequently limiting their potential applications, and their heavy reliance on analog-to-digital (AD) and digital-to-analog (DA) conversion interfaces undermines their performance. Here we propose an innovative photonic computing architecture featuring scalable calculation precision and a novel photonic conversion interface. By leveraging Residue Number System (RNS) theory, the high-precision calculation is decomposed into multiple low-precision modular arithmetic operations executed through optical phase manipulation. Those operations directly interact with the digital system via our proposed optical digital-to-phase converter (ODPC) and phase-to-digital converter (OPDC). Through experimental demonstrations, we showcase a calculation precision of 9 bits and verify the feasibility of the ODPC/OPDC photonic interface. This approach paves the path towards liberating photonic computing from the constraints imposed by limited precision and AD/DA converters.

physics.optics↗
arXiv2023

Bilingual Streaming ASR with Grapheme units and Auxiliary Monolingual Loss

We introduce a bilingual solution to support English as secondary locale for most primary locales in hybrid automatic speech recognition (ASR) settings. Our key developments constitute: (a) pronunciation lexicon with grapheme units instead of phone units, (b) a fully bilingual alignment model and subsequently bilingual streaming transformer model, (c) a parallel encoder structure with language identification (LID) loss, (d) parallel encoder with an auxiliary loss for monolingual projections. We conclude that in comparison to LID loss, our proposed auxiliary loss is superior in specializing the parallel encoders to respective monolingual locales, and that contributes to stronger bilingual learning. We evaluate our work on large-scale training and test tasks for bilingual Spanish (ES) and bilingual Italian (IT) applications. Our bilingual models demonstrate strong English code-mixing capability. In particular, the bilingual IT model improves the word error rate (WER) for a code-mix IT task from 46.5% to 13.8%, while also achieving a close parity (9.6%) with the monolingual IT model (9.5%) over IT tests.

eess.AS↗
arXiv2023

PyStructureFactor: A Python code for the molecular structure factor in tunneling ionization rates

Tunneling ionization is at the core of strong-field and attosecond science. In this paper, we present PyStructureFactor - a general Python code towards the calculation of the structure factor in the tunneling ionization rate of common molecules under intense laser fields. The numerical implementation is based on the well-developed weak-field asymptotic theory in the integral representation. The information of the electronic structure of the molecules is obtained via the PySCF quantum chemistry package. PyStructureFactor is a general computational framework that can be utilized to compute the molecular structure factor of various types of molecules, including polar and non-polar diatomic molecules, degenerate molecules, and open-shell molecules. Examples are given that are benchmarked against known results with good agreements. The present PyStructureFactor is implemented in an efficient manner and is easily applicable towards larger molecules.

physics.atom-ph↗
arXiv2023

It's Not Just GitHub: Identifying Data and Software Sources Included in Publications

Paper publications are no longer the only form of research product. Due to recent initiatives by publication venues and funding institutions, open access datasets and software products are increasingly considered research products and URIs to these products are growing more prevalent in scholarly publications. However, as with all URIs, resources found on the live Web are not permanent. Archivists and institutions including Software Heritage, Internet Archive, and Zenodo are working to preserve data and software products as valuable parts of reproducibility, a cornerstone of scientific research. While some hosting platforms are well-known and can be identified with regular expressions, there are a vast number of smaller, more niche hosting platforms utilized by researchers to host their data and software. If it is not feasible to manually identify all hosting platforms used by researchers, how can we identify URIs to open-access data and software (OADS) to aid in their preservation? We used a hybrid classifier to classify URIs as OADS URIs and non-OADS URIs. We found that URIs to Git hosting platforms (GHPs) including GitHub, GitLab, SourceForge, and Bitbucket accounted for 33\% of OADS URIs. Non-GHP OADS URIs are distributed across almost 50,000 unique hostnames. We determined that using a hybrid classifier allows for the identification of OADS URIs in less common hosting platforms which can benefit discoverability for preserving datasets and software products as research products for reproducibility.

cs.DL↗
arXiv2023

Ord2Seq: Regarding Ordinal Regression as Label Sequence Prediction

Ordinal regression refers to classifying object instances into ordinal categories. It has been widely studied in many scenarios, such as medical disease grading, movie rating, etc. Known methods focused only on learning inter-class ordinal relationships, but still incur limitations in distinguishing adjacent categories thus far. In this paper, we propose a simple sequence prediction framework for ordinal regression called Ord2Seq, which, for the first time, transforms each ordinal category label into a special label sequence and thus regards an ordinal regression task as a sequence prediction process. In this way, we decompose an ordinal regression task into a series of recursive binary classification steps, so as to subtly distinguish adjacent categories. Comprehensive experiments show the effectiveness of distinguishing adjacent categories for performance improvement and our new approach exceeds state-of-the-art performances in four different scenarios. Codes are available at https://github.com/wjh892521292/Ord2Seq.

cs.AI↗