Weijian Li

Junda Wang, Weijian Li, Han Wang et al.

Causal inference and model interpretability research are gaining increasing attention, especially in the domains of healthcare and bioinformatics. Despite recent successes in this field, decorrelating features under nonlinear environments with human interpretable representations has not been adequately investigated. To address this issue, we introduce a novel method with a variable decorrelation regularizer to handle both linear and nonlinear confounding. Moreover, we employ association rules as new representations using association rule mining based on the original features to further proximate human decision patterns to increase model interpretability. Extensive experiments are conducted on four healthcare datasets (one synthetically generated and three real-world collections on different diseases). Quantitative results in comparison to baseline approaches on parameter estimation and causality computation indicate the model's superior performance. Furthermore, expert evaluation given by healthcare professionals validates the effectiveness and interpretability of the proposed model.

Tim Tsz-Kit Lau, Weijian Li, Chenwei Xu et al.

An appropriate choice of batch sizes in large-scale model training is crucial, yet it involves an intrinsic yet inevitable dilemma: large-batch training improves training efficiency in terms of memory utilization, while generalization performance often deteriorates due to small amounts of gradient noise. Despite this dilemma, the common practice of choosing batch sizes in language model training often prioritizes training efficiency -- employing either constant large sizes with data parallelism or implementing batch size warmup schedules. However, such batch size schedule designs remain heuristic and often fail to adapt to training dynamics, presenting the challenge of designing adaptive batch size schedules. Given the abundance of available datasets and the data-hungry nature of language models, data parallelism has become an indispensable distributed training paradigm, enabling the use of larger batch sizes for gradient computation. However, vanilla data parallelism requires replicas of model parameters, gradients, and optimizer states at each worker, which prohibits training larger models with billions of parameters. To optimize memory usage, more advanced parallelism strategies must be employed. In this work, we propose general-purpose and theoretically principled adaptive batch size schedules compatible with data parallelism and model parallelism. We develop a practical implementation with PyTorch Fully Sharded Data Parallel, facilitating the pretraining of language models of different sizes. We empirically demonstrate that our proposed approaches outperform constant batch sizes and heuristic batch size warmup schedules in the pretraining of models in the Llama 2 family, with particular focus on smaller models with up to 3 billion parameters. We also establish theoretical convergence guarantees for such adaptive batch size schedules with Adam for general smooth nonconvex objectives.

Junda Wang, Weijian Li, Han Wang et al.

Causal inference and model interpretability are gaining increasing attention, particularly in the biomedical domain. Despite recent advance, decorrelating features in nonlinear environments with human-interpretable representations remains underexplored. In this study, we introduce a novel method called causal rule generation with target trial emulation framework (CRTRE), which applies randomize trial design principles to estimate the causal effect of association rules. We then incorporate such association rules for the downstream applications such as prediction of disease onsets. Extensive experiments on six healthcare datasets, including synthetic data, real-world disease collections, and MIMIC-III/IV, demonstrate the model's superior performance. Specifically, our method achieved a $β$ error of 0.907, outperforming DWR (1.024) and SVM (1.141). On real-world datasets, our model achieved accuracies of 0.789, 0.920, and 0.300 for Esophageal Cancer, Heart Disease, and Cauda Equina Syndrome prediction task, respectively, consistently surpassing baseline models. On the ICD code prediction tasks, it achieved AUC Macro scores of 92.8 on MIMIC-III and 96.7 on MIMIC-IV, outperforming the state-of-the-art models KEPT and MSMN. Expert evaluations further validate the model's effectiveness, causality, and interpretability.

Tim Tsz-Kit Lau, Weijian Li, Chenwei Xu et al.

Modern deep neural networks often require distributed training with many workers due to their large size. As the number of workers increases, communication overheads become the main bottleneck in data-parallel minibatch stochastic gradient methods with per-iteration gradient synchronization. Local gradient methods like Local SGD reduce communication by only synchronizing model parameters and/or gradients after several local steps. Despite an understanding of their convergence and the importance of batch sizes for training efficiency and generalization, optimal batch sizes for local gradient methods are difficult to determine. We introduce adaptive batch size strategies for local gradient methods that increase batch sizes adaptively to reduce minibatch gradient variance. We provide convergence guarantees under homogeneous data conditions and support our claims with image classification and language modeling experiments, demonstrating the effectiveness of our strategies for both training efficiency and generalization.

Weijian Li, Masato Kikuchi, Tadachika Ozono

Large-scale e-commerce sites can collect and analyze a large number of user preferences and behaviors, and thus can recommend highly trusted products to users. However, it is very difficult for individuals or non-corporate groups to obtain large-scale user data. Therefore, we consider whether knowledge of the decision-making domain can be used to obtain user preferences and combine it with content-based filtering to design an information retrieval system. This study describes the process of building a product information browsing support system with high satisfaction based on product similarity and multiple other perspectives about products on the Internet. We present the architecture of the proposed system and explain the working principle of its constituent modules. Finally, we demonstrate the effectiveness of the proposed system through an evaluation experiment and a questionnaire.

Wei Zhu, Haitian Zheng, Haofu Liao et al.

Deep learning algorithms mine knowledge from the training data and thus would likely inherit the dataset's bias information. As a result, the obtained model would generalize poorly and even mislead the decision process in real-life applications. We propose to remove the bias information misused by the target task with a cross-sample adversarial debiasing (CSAD) method. CSAD explicitly extracts target and bias features disentangled from the latent representation generated by a feature extractor and then learns to discover and remove the correlation between the target and bias features. The correlation measurement plays a critical role in adversarial debiasing and is conducted by a cross-sample neural mutual information estimator. Moreover, we propose joint content and local structural representation learning to boost mutual information estimation for better performance. We conduct thorough experiments on publicly available datasets to validate the advantages of the proposed method over state-of-the-art approaches.

Yikai Wang, Weijian Li

In recent years, natural language processing (NLP) has become one of the most important areas with various applications in human's life. As the most fundamental task, the field of word embedding still requires more attention and research. Currently, existing works about word embedding are focusing on proposing novel embedding algorithms and dimension reduction techniques on well-trained word embeddings. In this paper, we propose to use a novel joint signal separation method - JIVE to jointly decompose various trained word embeddings into joint and individual components. Through this decomposition framework, we can easily investigate the similarity and difference among different word embeddings. We conducted extensive empirical study on word2vec, FastText and GLoVE trained on different corpus and with different dimensions. We compared the performance of different decomposed components based on sentiment analysis on Twitter and Stanford sentiment treebank. We found that by mapping different word embeddings into the joint component, sentiment performance can be greatly improved for the original word embeddings with lower performance. Moreover, we found that by concatenating different components together, the same model can achieve better performance. These findings provide great insights into the word embeddings and our work offer a new of generating word embeddings by fusing.

Weijian Li, Wei Zhu, E. Ray Dorsey et al.

Parkinsons Disease is a neurological disorder and prevalent in elderly people. Traditional ways to diagnose the disease rely on in-person subjective clinical evaluations on the quality of a set of activity tests. The high-resolution longitudinal activity data collected by smartphone applications nowadays make it possible to conduct remote and convenient health assessment. However, out-of-lab tests often suffer from poor quality controls as well as irregularly collected observations, leading to noisy test results. To address these issues, we propose a novel time-series based approach to predicting Parkinson's Disease with raw activity test data collected by smartphones in the wild. The proposed method first synchronizes discrete activity tests into multimodal features at unified time points. Next, it distills and enriches local and global representations from noisy data across modalities and temporal observations by two attention modules. With the proposed mechanisms, our model is capable of handling noisy observations and at the same time extracting refined temporal features for improved prediction performance. Quantitative and qualitative results on a large public dataset demonstrate the effectiveness of the proposed approach.

Yuhang Lu, Weijian Li, Kang Zheng et al.

Accurate segmentation of critical anatomical structures is at the core of medical image analysis. The main bottleneck lies in gathering the requisite expert-labeled image annotations in a scalable manner. Methods that permit to produce accurate anatomical structure segmentation without using a large amount of fully annotated training images are highly desirable. In this work, we propose a novel contribution of Contour Transformer Network (CTN), a one-shot anatomy segmentor including a naturally built-in human-in-the-loop mechanism. Segmentation is formulated by learning a contour evolution behavior process based on graph convolutional networks (GCNs). Training of our CTN model requires only one labeled image exemplar and leverages additional unlabeled data through newly introduced loss functions that measure the global shape and appearance consistency of contours. We demonstrate that our one-shot learning method significantly outperforms non-learning-based methods and performs competitively to the state-of-the-art fully supervised deep learning approaches. With minimal human-in-the-loop editing feedback, the segmentation performance can be further improved and tailored towards the observer desired outcomes. This can facilitate the clinician designed imaging-based biomarker assessments (to support personalized quantitative clinical diagnosis) and outperforms fully supervised baselines.

Weijian Li, Viet-Duy Nguyen, Haofu Liao et al.

Automated whole slide image (WSI) tagging has become a growing demand due to the increasing volume and diversity of WSIs collected nowadays in histopathology. Various methods have been studied to classify WSIs with single tags but none of them focuses on labeling WSIs with multiple tags. To this end, we propose a novel end-to-end trainable deep neural network named Patch Transformer which can effectively predict multiple slide-level tags from WSI patches based on both the correlations and the uniqueness between the tags. Specifically, the proposed method learns patch characteristics considering 1) patch-wise relations through a patch transformation module and 2) tag-wise uniqueness for each tagging task through a multi-tag attention module. Extensive experiments on a large and diverse dataset consisting of 4,920 WSIs prove the effectiveness of the proposed model.

Mengshi Qi, Weijian Li, Zhengyuan Yang et al.

Scene graph generation refers to the task of automatically mapping an image into a semantic structural graph, which requires correctly labeling each extracted object and their interaction relationships. Despite the recent success in object detection using deep learning techniques, inferring complex contextual relationships and structured graph representations from visual data remains a challenging topic. In this study, we propose a novel Attentive Relational Network that consists of two key modules with an object detection backbone to approach this problem. The first module is a semantic transformation module utilized to capture semantic embedded relation features, by translating visual features and linguistic features into a common semantic space. The other module is a graph self-attention module introduced to embed a joint graph representation through assigning various importance weights to neighboring nodes. Finally, accurate scene graphs are produced by the relation inference module to recognize all entities and the corresponding relations. We evaluate our proposed method on the widely-adopted Visual Genome Dataset, and the results demonstrate the effectiveness and superiority of our model.