arXiv subjects

Wei Lu

Publications and source records attributed to Wei Lu.

At least 199 records · Page 11Linked to original sources

arXiv2024

From Model-centered to Human-Centered: Revision Distance as a Metric for Text Evaluation in LLMs-based Applications

Evaluating large language models (LLMs) is fundamental, particularly in the context of practical applications. Conventional evaluation methods, typically designed primarily for LLM development, yield numerical scores that ignore the user experience. Therefore, our study shifts the focus from model-centered to human-centered evaluation in the context of AI-powered writing assistance applications. Our proposed metric, termed ``Revision Distance,'' utilizes LLMs to suggest revision edits that mimic the human writing process. It is determined by counting the revision edits generated by LLMs. Benefiting from the generated revision edit details, our metric can provide a self-explained text evaluation result in a human-understandable manner beyond the context-independent score. Our results show that for the easy-writing task, ``Revision Distance'' is consistent with established metrics (ROUGE, Bert-score, and GPT-score), but offers more insightful, detailed feedback and better distinguishes between texts. Moreover, in the context of challenging academic writing tasks, our metric still delivers reliable evaluations where other metrics tend to struggle. Furthermore, our metric also holds significant potential for scenarios lacking reference texts.

cs.CL↗

arXiv2024

Sailor: Open Language Models for South-East Asia

We present Sailor, a family of open language models ranging from 0.5B to 7B parameters, tailored for South-East Asian (SEA) languages. These models are continually pre-trained from Qwen1.5, a great language model for multilingual use cases. From Qwen1.5, Sailor models accept 200B to 400B tokens, primarily covering the languages of English, Chinese, Vietnamese, Thai, Indonesian, Malay, and Lao. The training leverages several techniques, including BPE dropout for improving the model robustness, aggressive data cleaning and deduplication, and small proxy models to optimize data mixture. Experimental results on four typical tasks indicate that Sailor models demonstrate strong performance across different benchmarks, including commonsense reasoning, question answering, reading comprehension and examination. Embracing the open-source spirit, we share our insights through this report to spark a wider interest in developing large language models for multilingual use cases.

cs.CL↗

arXiv2024

AeonG: An Efficient Built-in Temporal Support in Graph Databases

Real world graphs are often dynamic and evolve over time. It is crucial for storing and querying graph evolution in graph databases. However, existing works either suffer from high storage overhead or lack efficient temporal query support, or both. In this paper, we propose AeonG, a new graph database with built-in temporal support. AeonG is based on a novel temporal graph model. To fit this model, we design a storage engine and a query engine. Our storage engine is hybrid, with one current storage to manage the most recent versions of graph objects, and another historical storage to manage the previous versions of graph objects. This separation makes the performance degradation of querying the most recent graph object versions as slight as possible. To reduce the historical storage overhead, we propose a novel anchor+delta strategy, in which we periodically create a complete version (namely anchor) of a graph object, and maintain every change (namely delta) between two adjacent anchors of the same object. To boost temporal query processing, we propose an anchor-based version retrieval technique in the query engine to skip unnecessary historical version traversals. Extensive experiments are conducted on both real and synthetic datasets. The results show that AeonG achieves up to 5.73X lower storage consumption and 2.57X lower temporal query latency against state-of-the-art approaches, while introducing only 9.74% performance degradation for supporting temporal features

cs.DB↗

arXiv2024

How biomedical papers accumulated their clinical citations: A large-scale retrospective analysis based on PubMed

This paper explored the temporal characteristics of clinical citations of biomedical papers, including how long it takes to receive its first clinical citation (the initial stage) and how long it takes to receive two or more clinical citations after its first clinical citation (the build-up stage). Over 23 million biomedical papers in PubMed between 1940 and 2013 and their clinical citations are used as the research data. We divide these biomedical papers into three groups and four categories from clinical citation level and translational science perspectives. We compare the temporal characteristics of biomedical papers of different groups or categories. From the perspective of clinical citation level, the results show that highly clinically cited papers had obvious advantages of receiving clinical citations over medium and lowly clinically cited papers in both the initial and build-up stages. Meanwhile, as the number of clinical citations increased in the build-up stage, the difference in the length of time to receive the corresponding number of clinical citations among the three groups of biomedical papers significantly increased. From the perspective of translational science, the results reveal that biomedical papers closer to clinical science more easily receive clinical citations than papers closer to basic science in both the initial and build-up stages. Moreover, we found that highly clinically cited papers had the desperate advantage of receiving clinical citations over even the clinical guidelines or clinical trials. The robustness analysis of the two aspects demonstrates the reliability of our results.

cs.DL↗

arXiv2024

Constrained Layout Generation with Factor Graphs

This paper addresses the challenge of object-centric layout generation under spatial constraints, seen in multiple domains including floorplan design process. The design process typically involves specifying a set of spatial constraints that include object attributes like size and inter-object relations such as relative positioning. Existing works, which typically represent objects as single nodes, lack the granularity to accurately model complex interactions between objects. For instance, often only certain parts of an object, like a room's right wall, interact with adjacent objects. To address this gap, we introduce a factor graph based approach with four latent variable nodes for each room, and a factor node for each constraint. The factor nodes represent dependencies among the variables to which they are connected, effectively capturing constraints that are potentially of a higher order. We then develop message-passing on the bipartite graph, forming a factor graph neural network that is trained to produce a floorplan that aligns with the desired requirements. Our approach is simple and generates layouts faithful to the user requirements, demonstrated by a large improvement in IOU scores over existing methods. Additionally, our approach, being inferential and accurate, is well-suited to the practical human-in-the-loop design process where specifications evolve iteratively, offering a practical and powerful tool for AI-guided design.

cs.CV↗

arXiv2024

A Two-Phase Recall-and-Select Framework for Fast Model Selection

As the ubiquity of deep learning in various machine learning applications has amplified, a proliferation of neural network models has been trained and shared on public model repositories. In the context of a targeted machine learning assignment, utilizing an apt source model as a starting point typically outperforms the strategy of training from scratch, particularly with limited training data. Despite the investigation and development of numerous model selection strategies in prior work, the process remains time-consuming, especially given the ever-increasing scale of model repositories. In this paper, we propose a two-phase (coarse-recall and fine-selection) model selection framework, aiming to enhance the efficiency of selecting a robust model by leveraging the models' training performances on benchmark datasets. Specifically, the coarse-recall phase clusters models showcasing similar training performances on benchmark datasets in an offline manner. A light-weight proxy score is subsequently computed between this model cluster and the target dataset, which serves to recall a significantly smaller subset of potential candidate models in a swift manner. In the following fine-selection phase, the final model is chosen by fine-tuning the recalled models on the target dataset with successive halving. To accelerate the process, the final fine-tuning performance of each potential model is predicted by mining the model's convergence trend on the benchmark datasets, which aids in filtering lower performance models more earlier during fine-tuning. Through extensive experimentation on tasks covering natural language processing and computer vision, it has been demonstrated that the proposed methodology facilitates the selection of a high-performing model at a rate about 3x times faster than conventional baseline methods. Our code is available at https://github.com/plasware/two-phase-selection.

cs.LG↗

arXiv2024

Hilbert's Irreducibility Theorem for Linear Differential Operators

We prove a differential analogue of Hilbert's irreducibility theorem. Let $\mathcal{L}$ be a linear differential operator with coefficients in $C(\mathbb{X})(x)$ that is irreducible over $\overline{C(\mathbb{X})}(x)$, where $\mathbb{X}$ is an irreducible affine algebraic variety over an algebraically closed field $C$ of characteristic zero. We show that the set of $c\in \mathbb{X}(C)$ such that the specialized operator $\mathcal{L}^c$ of $\mathcal{L}$ remains irreducible over $C(x)$ is Zariski dense in $\mathbb{X}(C)$.

math.RA↗

arXiv2024

Lion: Minimizing Distributed Transactions through Adaptive Replica Provision (Extended Version)

Distributed transaction processing often involves multiple rounds of cross-node communications, and therefore tends to be slow. To improve performance, existing approaches convert distributed transactions into single-node transactions by either migrating co-accessed partitions onto the same nodes or establishing a super node housing replicas of the entire database. However, migration-based methods might cause transactions to be blocked due to waiting for data migration, while the super node can become a bottleneck. In this paper, we present Lion, a novel transaction processing protocol that utilizes partition-based replication to reduce the occurrence of distributed transactions. Lion aims to assign a node with one replica from each partition involved in a given transaction's read or write operations. To ensure such a node is available, we propose an adaptive replica provision mechanism, enhanced with an LSTM-based workload prediction algorithm, to determine the appropriate node for locating replicas of co-accessed partitions. The adaptation of replica placement is conducted preemptively and asynchronously, thereby minimizing its impact on performance. By employing this adaptive replica placement strategy, we ensure that the majority of transactions can be efficiently processed on a single node without additional overhead. Only a small fraction of transactions will need to be treated as regular distributed transactions when such a node is unavailable. Consequently, Lion effectively minimizes distributed transactions while avoiding any disruption caused by data migration or the creation of a super node. We conduct extensive experiments to compare Lion against various transaction processing protocols. The results show that Lion achieves up to 2.7x higher throughput and 76.4% better scalability against these state-of-the-art approaches.

cs.DC↗

arXiv2024

Generative Models are Self-Watermarked: Declaring Model Authentication through Re-Generation

As machine- and AI-generated content proliferates, protecting the intellectual property of generative models has become imperative, yet verifying data ownership poses formidable challenges, particularly in cases of unauthorized reuse of generated data. The challenge of verifying data ownership is further amplified by using Machine Learning as a Service (MLaaS), which often functions as a black-box system. Our work is dedicated to detecting data reuse from even an individual sample. Traditionally, watermarking has been leveraged to detect AI-generated content. However, unlike watermarking techniques that embed additional information as triggers into models or generated content, potentially compromising output quality, our approach identifies latent fingerprints inherently present within the outputs through re-generation. We propose an explainable verification procedure that attributes data ownership through re-generation, and further amplifies these fingerprints in the generative models through iterative data re-generation. This methodology is theoretically grounded and demonstrates viability and robustness using recent advanced text and image generative models. Our methodology is significant as it goes beyond protecting the intellectual property of APIs and addresses important issues such as the spread of misinformation and academic misconduct. It provides a useful tool to ensure the integrity of sources and authorship, expanding its application in different scenarios where authenticity and ownership verification are essential.

cs.LG↗

arXiv2024

Synthetic Data (Almost) from Scratch: Generalized Instruction Tuning for Language Models

We introduce Generalized Instruction Tuning (called GLAN), a general and scalable method for instruction tuning of Large Language Models (LLMs). Unlike prior work that relies on seed examples or existing datasets to construct instruction tuning data, GLAN exclusively utilizes a pre-curated taxonomy of human knowledge and capabilities as input and generates large-scale synthetic instruction data across all disciplines. Specifically, inspired by the systematic structure in human education system, we build the taxonomy by decomposing human knowledge and capabilities to various fields, sub-fields and ultimately, distinct disciplines semi-automatically, facilitated by LLMs. Subsequently, we generate a comprehensive list of subjects for every discipline and proceed to design a syllabus tailored to each subject, again utilizing LLMs. With the fine-grained key concepts detailed in every class session of the syllabus, we are able to generate diverse instructions with a broad coverage across the entire spectrum of human knowledge and skills. Extensive experiments on large language models (e.g., Mistral) demonstrate that GLAN excels in multiple dimensions from mathematical reasoning, coding, academic exams, logical reasoning to general instruction following without using task-specific training data of these tasks. In addition, GLAN allows for easy customization and new fields or skills can be added by simply incorporating a new node into our taxonomy.

cs.CL↗

arXiv2024

Anomaly Detection of Particle Orbit in Accelerator using LSTM Deep Learning Technology

A stable, reliable, and controllable orbit lock system is crucial to an electron (or ion) accelerator because the beam orbit and beam energy instability strongly affect the quality of the beam delivered to experimental halls. Currently, when the orbit lock system fails operators must manually intervene. This paper develops a Machine Learning based fault detection methodology to identify orbit lock anomalies and notify accelerator operations staff of the off-normal behavior. Our method is unsupervised, so it does not require labeled data. It uses Long-Short Memory Networks (LSTM) Auto Encoder to capture normal patterns and predict future values of monitoring sensors in the orbit lock system. Anomalies are detected when the prediction error exceeds a threshold. We conducted experiments using monitoring data from Jefferson Lab's Continuous Electron Beam Accelerator Facility (CEBAF). The results are promising: the percentage of real anomalies identified by our solution is 68.6%-89.3% using monitoring data of a single component in the orbit lock control system. The accuracy can be as high as 82%.

cs.LG↗

arXiv2023

Infinite families of MDS and almost MDS codes from BCH codes

In this paper, the sufficient and necessary condition for the minimum distance of the BCH codes over $\mathbb{F}_q$ with length $q+1$ and designed distance 3 to be 3 and 4 are provided. Let $d$ be the minimum distance of the BCH code $\mathcal{C}_{(q,q+1,3,h)}$. We prove that (1) for any $q$, $d=3$ if and only if $\gcd(2h+1,q+1)>1$; (2) for $q$ odd, $d=4$ if and only if $\gcd(2h+1,q+1)=1$. By combining these conditions with the dimensions of these codes, the parameters of this BCH code are determined completely when $q$ is odd. Moreover, several infinite families of MDS and almost MDS (AMDS) codes are shown. Furthermore, the sufficient conditions for these AMDS codes to be distance-optimal and dimension-optimal locally repairable codes are presented. Based on these conditions, several examples are also given.

cs.IT↗

arXiv2023

Experimental demonstration of mice tumor control with a laser-accelerated high-energy electron radiotherapy prototype

Radiotherapy using very-high-energy electron (VHEE) beams (50-300 MeV) has attracted considerable attention due to its advantageous dose deposition characteristics, enabling deep penetration and the potential for ultra-high dose rate treatment. One promising approach to compactly delivering these high energy electron beams in a cost-effective manner is laser wakefield acceleration (LWFA), which offers ultra-strong accelerating gradients. However, the transition from this concept to a functional machine intended for tumor treatment is still being investigated. Here we present the first self-developed prototype for LWFA-based VHEE radiotherapy, exhibiting high compactness (occupying less than 5 square meters) and high operational stability (validated over a period of one month). Subsequently, we employed this device to irradiate a tumor implanted in a mouse model. Following a dose delivery of $5.8\pm0.2$ Gy with precise tumor conformity, all irradiated mice exhibited pronounced control of tumor growth. For comparison, this tumor-control efficacy was similar to that achieved using commercial X-ray radiotherapy equipment operating at equivalent doses. These results demonstrate the potential of a compact laser-driven VHEE system for preclinical studies involving small animal models and its promising prospects for future clinical translation in cancer therapy.

physics.med-ph↗

arXiv2023

Tell2Design: A Dataset for Language-Guided Floor Plan Generation

We consider the task of generating designs directly from natural language descriptions, and consider floor plan generation as the initial research area. Language conditional generative models have recently been very successful in generating high-quality artistic images. However, designs must satisfy different constraints that are not present in generating artistic images, particularly spatial and relational constraints. We make multiple contributions to initiate research on this task. First, we introduce a novel dataset, \textit{Tell2Design} (T2D), which contains more than $80k$ floor plan designs associated with natural language instructions. Second, we propose a Sequence-to-Sequence model that can serve as a strong baseline for future research. Third, we benchmark this task with several text-conditional image generation models. We conclude by conducting human evaluations on the generated samples and providing an analysis of human performance. We hope our contributions will propel the research on language-guided design generation forward.

cs.CL↗

arXiv2023

A Scalable, High-Efficiency, Low-Energy-Spread, Laser Wakefield Accelerator using a Tri-plateau Plasma Channel

The emergence of multi-petawatt laser facilities is expected to push forward the maximum energy gain that can be achieved in a single stage of a LWFA to tens of GeV, which begs the question - is it likely to impact particle physics by providing a truly compact particle collider? Colliders have very stringent requirements on beam energy, acceleration efficiency and beam quality. In this article, we propose a LWFA scheme that can for the first time simultaneously achieve hitherto unrealized acceleration efficiency from the laser to the electron beam of >20% and a sub-one percent energy spread using a stepwise plasma structure and a nonlinearly chirped laser pulse. Three-dimensional high-fidelity simulations show that the nonlinear chirp can effectively mitigate the laser waveform distortion and lengthen the acceleration distance. This combined with an inter-stage rephasing process in the stepwise plasma can triple the beam energy gain compared to that in a uniform plasma for a fixed laser energy thereby dramatically increasing the efficiency. A dynamic beam loading effect can almost perfectly cancel the energy chirp that arises during the acceleration, leading to the sub-percent energy spread. This scheme is highly scalable and can be applied to peta-watt LWFA scenarios. Scaling laws are obtained that suggest electron beams with energy gain of >100 GeV, charge of 2 nC, and with an energy spread <1% can be realized with a high laser pulse to particle beam energy transfer efficiency in a LWFA driven by a peta-watt laser, which could be the basis for a proof of concept of one arm of a future electron-positron collider.

physics.acc-ph↗

arXiv2023

Stability Problems on D-finite Functions

This paper continues the studies of symbolic integration by focusing on the stability problems on D-finite functions. We introduce the notion of stability index in order to investigate the order growth of the differential operators satisfied by iterated integrals of D-finite functions and determine bounds and exact formula for stability indices of several special classes of differential operators. With the basic properties of stability index, we completely solve the stability problem on general hyperexponential functions.

cs.SC↗

arXiv2023

Unraveling Feature Extraction Mechanisms in Neural Networks

The underlying mechanism of neural networks in capturing precise knowledge has been the subject of consistent research efforts. In this work, we propose a theoretical approach based on Neural Tangent Kernels (NTKs) to investigate such mechanisms. Specifically, considering the infinite network width, we hypothesize the learning dynamics of target models may intuitively unravel the features they acquire from training data, deepening our insights into their internal mechanisms. We apply our approach to several fundamental models and reveal how these models leverage statistical features during gradient descent and how they are integrated into final decisions. We also discovered that the choice of activation function can affect feature extraction. For instance, the use of the \textit{ReLU} activation function could potentially introduce a bias in features, providing a plausible explanation for its replacement with alternative functions in recent pre-trained language models. Additionally, we find that while self-attention and CNN models may exhibit limitations in learning n-grams, multiplication-based models seem to excel in this area. We verify these theoretical findings through experiments and find that they can be applied to analyze language modeling tasks, which can be regarded as a special variant of classification. Our contributions offer insights into the roles and capacities of fundamental components within large language models, thereby aiding the broader understanding of these complex systems.

cs.CL↗

arXiv2023

Tuna: Instruction Tuning using Feedback from Large Language Models

Instruction tuning of open-source large language models (LLMs) like LLaMA, using direct outputs from more powerful LLMs such as Instruct-GPT and GPT-4, has proven to be a cost-effective way to align model behaviors with human preferences. However, the instruction-tuned model has only seen one response per instruction, lacking the knowledge of potentially better responses. In this paper, we propose finetuning an instruction-tuned LLM using our novel \textit{probabilistic ranking} and \textit{contextual ranking} approaches to increase the likelihood of generating better responses. Probabilistic ranking enables the instruction-tuned model to inherit the relative rankings of high-quality and low-quality responses from the teacher LLM. On the other hand, learning with contextual ranking allows the model to refine its own response distribution using the contextual understanding ability of stronger LLMs. Furthermore, we apply probabilistic ranking and contextual ranking sequentially to the instruction-tuned LLM. The resulting model, which we call \textbf{Tuna}, consistently improves the performance on Super Natural Instructions (119 test tasks), LMentry (25 test tasks), Vicuna QA, and can even obtain better results than several strong reinforcement learning baselines. Our code and data are available at \url{ https://github.com/microsoft/LMOps}.

cs.CL↗