arXiv subjects

Yu Zhao

Publications and source records attributed to Yu Zhao.

At least 181 records · Page 10Linked to original sources

arXiv2024

Retrieval-style In-Context Learning for Few-shot Hierarchical Text Classification

Hierarchical text classification (HTC) is an important task with broad applications, while few-shot HTC has gained increasing interest recently. While in-context learning (ICL) with large language models (LLMs) has achieved significant success in few-shot learning, it is not as effective for HTC because of the expansive hierarchical label sets and extremely-ambiguous labels. In this work, we introduce the first ICL-based framework with LLM for few-shot HTC. We exploit a retrieval database to identify relevant demonstrations, and an iterative policy to manage multi-layer hierarchical labels. Particularly, we equip the retrieval database with HTC label-aware representations for the input texts, which is achieved by continual training on a pretrained language model with masked language modeling (MLM), layer-wise classification (CLS, specifically for HTC), and a novel divergent contrastive learning (DCL, mainly for adjacent semantically-similar labels) objective. Experimental results on three benchmark datasets demonstrate superior performance of our method, and we can achieve state-of-the-art results in few-shot HTC.

cs.CL↗

arXiv2024

Next-Generation Simulation Illuminates Scientific Problems of Organised Complexity

As artificial intelligence becomes increasingly prevalent in scientific research, data-driven methodologies appear to overshadow traditional approaches in resolving scientific problems. In this Perspective, we revisit a classic classification of scientific problems and acknowledge that a series of unresolved problems remain. Throughout the history of researching scientific problems, scientists have continuously formed new paradigms facilitated by advances in data, algorithms, and computational power. To better tackle unresolved problems, especially those of organised complexity, a novel paradigm is necessitated. While recognising that the strengths of new paradigms have expanded the scope of resolvable scientific problems, we aware that the continued advancement of data, algorithms, and computational power alone is hardly to bring a new paradigm. We posit that the integration of paradigms, which capitalises on the strengths of each, represents a promising approach. Specifically, we focus on next-generation simulation (NGS), which can serve as a platform to integrate methods from different paradigms. We propose a methodology, sophisticated behavioural simulation (SBS), to realise it. SBS represents a higher level of paradigms integration based on foundational models to simulate complex systems, such as social systems involving sophisticated human strategies and behaviours. NGS extends beyond the capabilities of traditional mathematical modelling simulations and agent-based modelling simulations, and therefore, positions itself as a potential solution to problems of organised complexity in complex systems.

cs.AI↗

arXiv2024

RAG-based Crowdsourcing Task Decomposition via Masked Contrastive Learning with Prompts

Crowdsourcing is a critical technology in social manufacturing, which leverages an extensive and boundless reservoir of human resources to handle a wide array of complex tasks. The successful execution of these complex tasks relies on task decomposition (TD) and allocation, with the former being a prerequisite for the latter. Recently, pre-trained language models (PLMs)-based methods have garnered significant attention. However, they are constrained to handling straightforward common-sense tasks due to their inherent restrictions involving limited and difficult-to-update knowledge as well as the presence of hallucinations. To address these issues, we propose a retrieval-augmented generation-based crowdsourcing framework that reimagines TD as event detection from the perspective of natural language understanding. However, the existing detection methods fail to distinguish differences between event types and always depend on heuristic rules and external semantic analyzing tools. Therefore, we present a Prompt-Based Contrastive learning framework for TD (PBCT), which incorporates a prompt-based trigger detector to overcome dependence. Additionally, trigger-attentive sentinel and masked contrastive learning are introduced to provide varying attention to trigger and contextual features according to different event types. Experiment results demonstrate the competitiveness of our method in both supervised and zero-shot detection. A case study on printed circuit board manufacturing is showcased to validate its adaptability to unknown professional domains.

cs.CL↗

arXiv2024

Efficient bound preserving and asymptotic preserving semi-implicit schemes for the fast reaction-diffusion system

We consider a special type of fast reaction-diffusion systems in which the coefficients of the reaction terms of the two substances are much larger than those of the diffusion terms while the diffusive motion to the substrate is negligible. Specifically speaking, the rate constants of the reaction terms are $O(1/ε)$ while the diffusion coefficients are $O(1)$ where the parameter $ε$ is small. When the rate constants of the reaction terms become highly large, i.e. $ε$ tends to 0, the singular limit behavior of such a fast reaction-diffusion system is inscribed by the Stefan problem with latent heat, which brings great challenges in numerical simulations. In this paper, we adopt a semi-implicit scheme, which is first-order accurate in time and can accurately approximate the interface propagation even when the reaction becomes extremely fast, that is to say, the parameter $ε$ is sufficiently small. The scheme satisfies the positivity, bound preserving properties and has $L^2$ stability and the linearized stability results of the system. For better performance on numerical simulations, we then construct a semi-implicit Runge-Kutta scheme which is second-order accurate in time. Numerous numerical tests are carried out to demonstrate the properties, such as the order of accuracy, positivity and bound preserving, the capturing of the sharp interface with various $ε$ and to simulate the dynamics of the substances and the substrate, and to explore the heat transfer process, such as solid melting or liquid solidification in two dimensions.

math.NA↗

arXiv2024

Tele-FLM Technical Report

Large language models (LLMs) have showcased profound capabilities in language understanding and generation, facilitating a wide array of applications. However, there is a notable paucity of detailed, open-sourced methodologies on efficiently scaling LLMs beyond 50 billion parameters with minimum trial-and-error cost and computational resources. In this report, we introduce Tele-FLM (aka FLM-2), a 52B open-sourced multilingual large language model that features a stable, efficient pre-training paradigm and enhanced factual judgment capabilities. Tele-FLM demonstrates superior multilingual language modeling abilities, measured by BPB on textual corpus. Besides, in both English and Chinese foundation model evaluation, it is comparable to strong open-sourced models that involve larger pre-training FLOPs, such as Llama2-70B and DeepSeek-67B. In addition to the model weights, we share the core designs, engineering practices, and training details, which we expect to benefit both the academic and industrial communities.

cs.CL↗

arXiv2024

PuzzleTuning: Explicitly Bridge Pathological and Natural Image with Puzzles

Pathological image analysis is a crucial field in computer vision. Due to the annotation scarcity in the pathological field, pre-training with self-supervised learning (SSL) is widely applied to learn on unlabeled images. However, the current SSL-based pathological pre-training: (1) does not explicitly explore the essential focuses of the pathological field, and (2) does not effectively bridge with and thus take advantage of the knowledge from natural images. To explicitly address them, we propose our large-scale PuzzleTuning framework, containing the following innovations. Firstly, we define three task focuses that can effectively bridge knowledge of pathological and natural domain: appearance consistency, spatial consistency, and restoration understanding. Secondly, we devise a novel multiple puzzle restoring task, which explicitly pre-trains the model regarding these focuses. Thirdly, we introduce an explicit prompt-tuning process to incrementally integrate the domain-specific knowledge. It builds a bridge to align the large domain gap between natural and pathological images. Additionally, a curriculum-learning training strategy is designed to regulate task difficulty, making the model adaptive to the puzzle restoring complexity. Experimental results show that our PuzzleTuning framework outperforms the previous state-of-the-art methods in various downstream tasks on multiple datasets. The code, demo, and pre-trained weights are available at https://github.com/sagizty/PuzzleTuning.

eess.IV↗

arXiv2024

The Hallucinations Leaderboard -- An Open Effort to Measure Hallucinations in Large Language Models

Large Language Models (LLMs) have transformed the Natural Language Processing (NLP) landscape with their remarkable ability to understand and generate human-like text. However, these models are prone to ``hallucinations'' -- outputs that do not align with factual reality or the input context. This paper introduces the Hallucinations Leaderboard, an open initiative to quantitatively measure and compare the tendency of each model to produce hallucinations. The leaderboard uses a comprehensive set of benchmarks focusing on different aspects of hallucinations, such as factuality and faithfulness, across various tasks, including question-answering, summarisation, and reading comprehension. Our analysis provides insights into the performance of different models, guiding researchers and practitioners in choosing the most reliable models for their applications.

cs.CL↗

arXiv2024

Multi-Scale Subgraph Contrastive Learning

Graph-level contrastive learning, aiming to learn the representations for each graph by contrasting two augmented graphs, has attracted considerable attention. Previous studies usually simply assume that a graph and its augmented graph as a positive pair, otherwise as a negative pair. However, it is well known that graph structure is always complex and multi-scale, which gives rise to a fundamental question: after graph augmentation, will the previous assumption still hold in reality? By an experimental analysis, we discover the semantic information of an augmented graph structure may be not consistent as original graph structure, and whether two augmented graphs are positive or negative pairs is highly related with the multi-scale structures. Based on this finding, we propose a multi-scale subgraph contrastive learning architecture which is able to characterize the fine-grained semantic information. Specifically, we generate global and local views at different scales based on subgraph sampling, and construct multiple contrastive relationships according to their semantic associations to provide richer self-supervised signals. Extensive experiments and parametric analyzes on eight graph classification real-world datasets well demonstrate the effectiveness of the proposed method.

cs.AI↗

arXiv2024

TeleChat Technical Report

In this technical report, we present TeleChat, a collection of large language models (LLMs) with parameters of 3 billion, 7 billion and 12 billion. It includes pretrained language models as well as fine-tuned chat models that is aligned with human preferences. TeleChat is initially pretrained on an extensive corpus containing a diverse collection of texts from both English and Chinese languages, including trillions of tokens. Subsequently, the model undergoes fine-tuning to align with human preferences, following a detailed methodology that we describe. We evaluate the performance of TeleChat on various tasks, including language understanding, mathematics, reasoning, code generation, and knowledge-based question answering. Our findings indicate that TeleChat achieves comparable performance to other open-source models of similar size across a wide range of public benchmarks. To support future research and applications utilizing LLMs, we release the fine-tuned model checkpoints of TeleChat's 7B and 12B variant, along with code and a portion of our pretraining data, to the public community.

cs.CL↗

arXiv2024

The thesan project: public data release of radiation-hydrodynamic simulations matching reionization-era JWST observations

Cosmological simulations serve as invaluable tools for understanding the Universe. However, the technical complexity and substantial computational resources required to generate such simulations often limit their accessibility within the broader research community. Notable exceptions exist, but most are not suited for simultaneously studying the physics of galaxy formation and cosmic reionization during the first billion years of cosmic history. This is especially relevant now that a fleet of advanced observatories (e.g. James Webb Space Telescope, Nancy Grace Roman Space Telescope, SPHEREx, ELT, SKA) will soon provide an holistic picture of this defining epoch. To bridge this gap, we publicly release all simulation outputs and post-processing products generated within the THESAN simulation project at https://thesan-project.com. This project focuses on the $z \geq 5.5$ Universe, combining a radiation-hydrodynamics solver (AREPO-RT), a well-tested galaxy formation model (IllustrisTNG) and cosmic dust physics to provide a comprehensive view of the Epoch of Reionization. The THESAN suite includes 16 distinct simulations, each varying in volume, resolution, and underlying physical models. This paper outlines the unique features of these new simulations, the production and detailed format of the wide range of derived data products, and the process for data retrieval. Finally, as a case study, we compare our simulation data with a number of recent observations from the James Webb Space Telescope, affirming the accuracy and applicability of THESAN. The examples also serve as prototypes for how to utilise the released dataset to perform comparisons between predictions and observations.

astro-ph.CO↗

arXiv2024

Generating Progressive Images from Pathological Transitions via Diffusion Model

Deep learning is widely applied in computer-aided pathological diagnosis, which alleviates the pathologist workload and provide timely clinical analysis. However, most models generally require large-scale annotated data for training, which faces challenges due to the sampling and annotation scarcity in pathological images. The rapid developing generative models shows potential to generate more training samples from recent studies. However, they also struggle in generalization diversity with limited training data, incapable of generating effective samples. Inspired by the pathological transitions between different stages, we propose an adaptive depth-controlled diffusion (ADD) network to generate pathological progressive images for effective data augmentation. This novel approach roots in domain migration, where a hybrid attention strategy guides the bidirectional diffusion, blending local and global attention priorities. With feature measuring, the adaptive depth-controlled strategy ensures the migration and maintains locational similarity in simulating the pathological feature transition. Based on tiny training set (samples less than 500), the ADD yields cross-domain progressive images with corresponding soft-labels. Experiments on two datasets suggest significant improvements in generation diversity, and the effectiveness with generated progressive samples are highlighted in downstream classifications. The code is available at https://github.com/Rowerliu/ADD.

cs.CV↗

arXiv2024

Towards Optimal Customized Architecture for Heterogeneous Federated Learning with Contrastive Cloud-Edge Model Decoupling

Federated learning, as a promising distributed learning paradigm, enables collaborative training of a global model across multiple network edge clients without the need for central data collecting. However, the heterogeneity of edge data distribution drags the model towards the local minima, which can be distant from the global optimum. Such heterogeneity often leads to slow convergence and substantial communication overhead. To address these issues, we propose a novel federated learning framework called FedCMD, a model decoupling tailored to the Cloud-edge supported federated learning that separates deep neural networks into a body for capturing shared representations in Cloud and a personalized head for migrating data heterogeneity. Our motivation is that, by the deep investigation of the performance of selecting different neural network layers as the personalized head, we found rigidly assigning the last layer as the personalized head in current studies is not always optimal. Instead, it is necessary to dynamically select the personalized layer that maximizes the training performance by taking the representation difference between neighbor layers into account. To find the optimal personalized layer, we utilize the low-dimensional representation of each layer to contrast feature distribution transfer and introduce a Wasserstein-based layer selection method, aimed at identifying the best-match layer for personalization. Additionally, a weighted global aggregation algorithm is proposed based on the selected personalized layer for the practical application of FedCMD. Extensive experiments on ten benchmarks demonstrate the efficiency and superior performance of our solution compared with nine state-of-the-art solutions. All code and results are available at https://github.com/elegy112138/FedCMD.

cs.LG↗

arXiv2024

Analysing The Impact of Sequence Composition on Language Model Pre-Training

Most language model pre-training frameworks concatenate multiple documents into fixed-length sequences and use causal masking to compute the likelihood of each token given its context; this strategy is widely adopted due to its simplicity and efficiency. However, to this day, the influence of the pre-training sequence composition strategy on the generalisation properties of the model remains under-explored. In this work, we find that applying causal masking can lead to the inclusion of distracting information from previous documents during pre-training, which negatively impacts the performance of the models on language modelling and downstream tasks. In intra-document causal masking, the likelihood of each token is only conditioned on the previous tokens in the same document, eliminating potential distracting information from previous documents and significantly improving performance. Furthermore, we find that concatenating related documents can reduce some potential distractions during pre-training, and our proposed efficient retrieval-based sequence construction method, BM25Chunk, can improve in-context learning (+11.6\%), knowledge memorisation (+9.8\%), and context utilisation (+7.2\%) abilities of language models without sacrificing efficiency.

cs.CL↗

arXiv2024

StableMask: Refining Causal Masking in Decoder-only Transformer

The decoder-only Transformer architecture with causal masking and relative position encoding (RPE) has become the de facto choice in language modeling. Despite its exceptional performance across various tasks, we have identified two limitations: First, it requires all attention scores to be non-zero and sum up to 1, even if the current embedding has sufficient self-contained information. This compels the model to assign disproportional excessive attention to specific tokens. Second, RPE-based Transformers are not universal approximators due to their limited capacity at encoding absolute positional information, which limits their application in position-critical tasks. In this work, we propose StableMask: a parameter-free method to address both limitations by refining the causal mask. It introduces pseudo-attention values to balance attention distributions and encodes absolute positional information via a progressively decreasing mask ratio. StableMask's effectiveness is validated both theoretically and empirically, showing significant enhancements in language models with parameter sizes ranging from 71M to 1.4B across diverse datasets and encoding methods. We further show that it naturally supports (1) efficient extrapolation without special tricks such as StreamingLLM and (2) easy integration with existing attention optimization techniques.

cs.CL↗

arXiv2023

Closest targets in Russell graph measure of strongly monotonic efficiency for an extended facet production possibility set

This study modifies the Russell graph measure of efficiency to find the closest target from the assessed decision-making unit to the efficient frontier of an empirical production possibility set using an extended facet approach. The closest target requires a single improvement in either an output term or an input term, and needs some positive levels of all inputs to achieve the output target. The practical advantage of the proposed efficiency measure is demonstrated numerically in comparison to other existing non-radial efficiency measures.

math.OC↗

arXiv2023

Improving Neural Machine Translation by Multi-Knowledge Integration with Prompting

Improving neural machine translation (NMT) systems with prompting has achieved significant progress in recent years. In this work, we focus on how to integrate multi-knowledge, multiple types of knowledge, into NMT models to enhance the performance with prompting. We propose a unified framework, which can integrate effectively multiple types of knowledge including sentences, terminologies/phrases and translation templates into NMT models. We utilize multiple types of knowledge as prefix-prompts of input for the encoder and decoder of NMT models to guide the translation process. The approach requires no changes to the model architecture and effectively adapts to domain-specific translation without retraining. The experiments on English-Chinese and English-German translation demonstrate that our approach significantly outperform strong baselines, achieving high translation quality and terminology match accuracy.

cs.CL↗

arXiv2023

WATUNet: A Deep Neural Network for Segmentation of Volumetric Sweep Imaging Ultrasound

Objective. Limited access to breast cancer diagnosis globally leads to delayed treatment. Ultrasound, an effective yet underutilized method, requires specialized training for sonographers, which hinders its widespread use. Approach. Volume sweep imaging (VSI) is an innovative approach that enables untrained operators to capture high-quality ultrasound images. Combined with deep learning, like convolutional neural networks (CNNs), it can potentially transform breast cancer diagnosis, enhancing accuracy, saving time and costs, and improving patient outcomes. The widely used UNet architecture, known for medical image segmentation, has limitations, such as vanishing gradients and a lack of multi-scale feature extraction and selective region attention. In this study, we present a novel segmentation model known as Wavelet_Attention_UNet (WATUNet). In this model, we incorporate wavelet gates (WGs) and attention gates (AGs) between the encoder and decoder instead of a simple connection to overcome the limitations mentioned, thereby improving model performance. Main results. Two datasets are utilized for the analysis. The public "Breast Ultrasound Images" (BUSI) dataset of 780 images and a VSI dataset of 3818 images. Both datasets contained segmented lesions categorized into three types: no mass, benign mass, and malignant mass. Our segmentation results show superior performance compared to other deep networks. The proposed algorithm attained a Dice coefficient of 0.94 and an F1 score of 0.94 on the VSI dataset and scored 0.93 and 0.94 on the public dataset, respectively.

eess.IV↗

arXiv2023

How to get better embeddings with code pre-trained models? An empirical study

Pre-trained language models have demonstrated powerful capabilities in the field of natural language processing (NLP). Recently, code pre-trained model (PTM), which draw from the experiences of the NLP field, have also achieved state-of-the-art results in many software engineering (SE) downstream tasks. These code PTMs take into account the differences between programming languages and natural languages during pre-training and make adjustments to pre-training tasks and input data. However, researchers in the SE community still inherit habits from the NLP field when using these code PTMs to generate embeddings for SE downstream classification tasks, such as generating semantic embeddings for code snippets through special tokens and inputting code and text information in the same way as pre-training the PTMs. In this paper, we empirically study five different PTMs (i.e. CodeBERT, CodeT5, PLBART, CodeGPT and CodeGen) with three different architectures (i.e. encoder-only, decoder-only and encoder-decoder) on four SE downstream classification tasks (i.e. code vulnerability detection, code clone detection, just-in-time defect prediction and function docstring mismatch detection) with respect to the two aforementioned aspects. Our experimental results indicate that (1) regardless of the architecture of the code PTMs used, embeddings obtained through special tokens do not sufficiently aggregate the semantic information of the entire code snippet; (2) the quality of code embeddings obtained by combing code data and text data in the same way as pre-training the PTMs is poor and cannot guarantee richer semantic information; (3) using the method that aggregates the vector representations of all code tokens, the decoder-only PTMs can obtain code embeddings with semantics as rich as or even better quality than those obtained from the encoder-only and encoder-decoder PTMs.

cs.SE↗