Ying Wei

Yichen Wu, Hongming Piao, Long-Kai Huang et al.

Continual Learning (CL) with foundation models has recently emerged as a promising paradigm to exploit abundant knowledge acquired during pre-training for tackling sequential tasks. However, existing prompt-based and Low-Rank Adaptation-based (LoRA-based) methods often require expanding a prompt/LoRA pool or retaining samples of previous tasks, which poses significant scalability challenges as the number of tasks grows. To address these limitations, we propose Scalable Decoupled LoRA (SD-LoRA) for class incremental learning, which continually separates the learning of the magnitude and direction of LoRA components without rehearsal. Our empirical and theoretical analysis reveals that SD-LoRA tends to follow a low-loss trajectory and converges to an overlapping low-loss region for all learned tasks, resulting in an excellent stability-plasticity trade-off. Building upon these insights, we introduce two variants of SD-LoRA with further improved parameter efficiency. All parameters of SD-LoRAs can be end-to-end optimized for CL objectives. Meanwhile, they support efficient inference by allowing direct evaluation with the finally trained model, obviating the need for component selection. Extensive experiments across multiple CL benchmarks and foundation models consistently validate the effectiveness of SD-LoRA. The code is available at https://github.com/WuYichen-97/SD-Lora-CL.

Yunqiao Yang, Long-Kai Huang, Shengzhuang Chen et al.

Model editing aims to data-efficiently correct predictive errors of large pre-trained models while ensuring generalization to neighboring failures and locality to minimize unintended effects on unrelated examples. While significant progress has been made in editing Transformer-based large language models, effective strategies for editing vision Transformers (ViTs) in computer vision remain largely untapped. In this paper, we take initial steps towards correcting predictive errors of ViTs, particularly those arising from subpopulation shifts. Taking a locate-then-edit approach, we first address the where-to-edit challenge by meta-learning a hypernetwork on CutMix-augmented data generated for editing reliability. This trained hypernetwork produces generalizable binary masks that identify a sparse subset of structured model parameters, responsive to real-world failure samples. Afterward, we solve the how-to-edit problem by simply fine-tuning the identified parameters using a variant of gradient descent to achieve successful edits. To validate our method, we construct an editing benchmark that introduces subpopulation shifts towards natural underrepresented images and AI-generated images, thereby revealing the limitations of pre-trained ViTs for object recognition. Our approach not only achieves superior performance on the proposed benchmark but also allows for adjustable trade-offs between generalization and locality. Our code is available at https://github.com/hustyyq/Where-to-Edit.

Yunqiao Yang, Long-Kai Huang, Ying Wei

A multitude of prevalent pre-trained models mark a major milestone in the development of artificial intelligence, while fine-tuning has been a common practice that enables pretrained models to figure prominently in a wide array of target datasets. Our empirical results reveal that off-the-shelf finetuning techniques are far from adequate to mitigate negative transfer caused by two types of underperforming features in a pre-trained model, including rare features and spuriously correlated features. Rooted in structural causal models of predictions after fine-tuning, we propose a Concept-wise fine-tuning (Concept-Tuning) approach which refines feature representations in the level of patches with each patch encoding a concept. Concept-Tuning minimizes the negative impacts of rare features and spuriously correlated features by (1) maximizing the mutual information between examples in the same category with regard to a slice of rare features (a patch) and (2) applying front-door adjustment via attention neural networks in channels and feature slices (patches). The proposed Concept-Tuning consistently and significantly (by up to 4.76%) improves prior state-of-the-art fine-tuning methods on eleven datasets, diverse pre-training strategies (supervised and self-supervised ones), various network architectures, and sample sizes in a target dataset.

Haokun Lin, Haoli Bai, Zhili Liu et al.

Vision-language pre-trained models have achieved impressive performance on various downstream tasks. However, their large model sizes hinder their utilization on platforms with limited computational resources. We find that directly using smaller pre-trained models and applying magnitude-based pruning on CLIP models leads to inflexibility and inferior performance. Recent efforts for VLP compression either adopt uni-modal compression metrics resulting in limited performance or involve costly mask-search processes with learnable masks. In this paper, we first propose the Module-wise Pruning Error (MoPE) metric, accurately assessing CLIP module importance by performance decline on cross-modal tasks. Using the MoPE metric, we introduce a unified pruning framework applicable to both pre-training and task-specific fine-tuning compression stages. For pre-training, MoPE-CLIP effectively leverages knowledge from the teacher model, significantly reducing pre-training costs while maintaining strong zero-shot capabilities. For fine-tuning, consecutive pruning from width to depth yields highly competitive task-specific models. Extensive experiments in two stages demonstrate the effectiveness of the MoPE metric, and MoPE-CLIP outperforms previous state-of-the-art VLP compression methods.

Yemin Yu, Luotian Yuan, Ying Wei et al.

Machine learning-assisted retrosynthesis prediction models have been gaining widespread adoption, though their performances oftentimes degrade significantly when deployed in real-world applications embracing out-of-distribution (OOD) molecules or reactions. Despite steady progress on standard benchmarks, our understanding of existing retrosynthesis prediction models under the premise of distribution shifts remains stagnant. To this end, we first formally sort out two types of distribution shifts in retrosynthesis prediction and construct two groups of benchmark datasets. Next, through comprehensive experiments, we systematically compare state-of-the-art retrosynthesis prediction models on the two groups of benchmarks, revealing the limitations of previous in-distribution evaluation and re-examining the advantages of each model. More remarkably, we are motivated by the above empirical insights to propose two model-agnostic techniques that can improve the OOD generalization of arbitrary off-the-shelf retrosynthesis prediction algorithms. Our preliminary experiments show their high potential with an average performance improvement of 4.6%, and the established benchmarks serve as a foothold for further retrosynthesis prediction research towards OOD generalization.

Shuailei Ma, Chen-Wei Xie, Ying Wei et al.

Prompt learning has emerged as an efficient alternative for fine-tuning foundational models, such as CLIP, for various downstream tasks. However, there is no work that provides a comprehensive explanation for the working mechanism of the multi-modal prompts. In this paper, we conduct a direct analysis of the multi-modal prompts by asking the following questions: $(i)$ How do the learned multi-modal prompts improve the recognition performance? $(ii)$ What do the multi-modal prompts learn? To answer these questions, we begin by isolating the component of the formula where the prompt influences the calculation of self-attention at each layer in two distinct ways, \ie, $(1)$ introducing prompt embeddings makes the $[cls]$ token focus on foreground objects. $(2)$ the prompts learn a bias term during the update of token embeddings, allowing the model to adapt to the target domain. Subsequently, we conduct extensive visualization and statistical experiments on the eleven diverse downstream recognition datasets. From the experiments, we reveal that the learned prompts improve the performance mainly through the second way, which acts as the dataset bias to improve the recognition performance of the pre-trained model on the corresponding dataset. Meanwhile, we propose the bias tuning way to validate our finding. With a deeper understanding of the multi-modal prompt, we hope our work can inspire new and solid research in this direction.

Long-Kai Huang, Ying Wei, Yu Rong et al.

Transferability estimation has been an essential tool in selecting a pre-trained model and the layers in it for transfer learning, to transfer, so as to maximize the performance on a target task and prevent negative transfer. Existing estimation algorithms either require intensive training on target tasks or have difficulties in evaluating the transferability between layers. To this end, we propose a simple, efficient, and effective transferability measure named TransRate. Through a single pass over examples of a target task, TransRate measures the transferability as the mutual information between features of target examples extracted by a pre-trained model and their labels. We overcome the challenge of efficient mutual information estimation by resorting to coding rate that serves as an effective alternative to entropy. From the perspective of feature representation, the resulting TransRate evaluates both completeness (whether features contain sufficient information of a target task) and compactness (whether features of each class are compact enough for good generalization) of pre-trained features. Theoretically, we have analyzed the close connection of TransRate to the performance after transfer learning. Despite its extraordinary simplicity in 10 lines of codes, TransRate performs remarkably well in extensive evaluations on 32 pre-trained models and 16 downstream tasks.

Huaxiu Yao, Longkai Huang, Linjun Zhang et al.

Meta-learning has proven to be a powerful paradigm for transferring the knowledge from previous tasks to facilitate the learning of a novel task. Current dominant algorithms train a well-generalized model initialization which is adapted to each task via the support set. The crux lies in optimizing the generalization capability of the initialization, which is measured by the performance of the adapted model on the query set of each task. Unfortunately, this generalization measure, evidenced by empirical results, pushes the initialization to overfit the meta-training tasks, which significantly impairs the generalization and adaptation to novel tasks. To address this issue, we actively augment a meta-training task with "more data" when evaluating the generalization. Concretely, we propose two task augmentation methods, including MetaMix and Channel Shuffle. MetaMix linearly combines features and labels of samples from both the support and query sets. For each class of samples, Channel Shuffle randomly replaces a subset of their channels with the corresponding ones from a different class. Theoretical studies show how task augmentation improves the generalization of meta-learning. Moreover, both MetaMix and Channel Shuffle outperform state-of-the-art results by a large margin across many datasets and are compatible with existing meta-learning algorithms.

Yifan Zhang, Ying Wei, Qingyao Wu et al.

Deep learning based medical image diagnosis has shown great potential in clinical medicine. However, it often suffers two major difficulties in real-world applications: 1) only limited labels are available for model training, due to expensive annotation costs over medical images; 2) labeled images may contain considerable label noise (e.g., mislabeling labels) due to diagnostic difficulties of diseases. To address these, we seek to exploit rich labeled data from relevant domains to help the learning in the target task via {Unsupervised Domain Adaptation} (UDA). Unlike most UDA methods that rely on clean labeled data or assume samples are equally transferable, we innovatively propose a Collaborative Unsupervised Domain Adaptation algorithm, which conducts transferability-aware adaptation and conquers label noise in a collaborative way. We theoretically analyze the generalization performance of the proposed method, and also empirically evaluate it on both medical and general images. Promising experimental results demonstrate the superiority and generalization of the proposed method.

Yu Rong, Yatao Bian, Tingyang Xu et al.

How to obtain informative representations of molecules is a crucial prerequisite in AI-driven drug design and discovery. Recent researches abstract molecules as graphs and employ Graph Neural Networks (GNNs) for molecular representation learning. Nevertheless, two issues impede the usage of GNNs in real scenarios: (1) insufficient labeled molecules for supervised training; (2) poor generalization capability to new-synthesized molecules. To address them both, we propose a novel framework, GROVER, which stands for Graph Representation frOm self-superVised mEssage passing tRansformer. With carefully designed self-supervised tasks in node-, edge- and graph-level, GROVER can learn rich structural and semantic information of molecules from enormous unlabelled molecular data. Rather, to encode such complex information, GROVER integrates Message Passing Networks into the Transformer-style architecture to deliver a class of more expressive encoders of molecules. The flexibility of GROVER allows it to be trained efficiently on large-scale molecular dataset without requiring any supervision, thus being immunized to the two issues mentioned above. We pre-train GROVER with 100 million parameters on 10 million unlabelled molecules -- the biggest GNN and the largest training dataset in molecular representation learning. We then leverage the pre-trained GROVER for molecular property prediction followed by task-specific fine-tuning, where we observe a huge improvement (more than 6% on average) from current state-of-the-art methods on 11 challenging benchmarks. The insights we gained are that well-designed self-supervision losses and largely-expressive pre-trained models enjoy the significant potential on performance boosting.

Yifan Zhang, Shuaicheng Niu, Zhen Qiu et al.

The outbreak of novel coronavirus disease 2019 (COVID-19) has already infected millions of people and is still rapidly spreading all over the globe. Most COVID-19 patients suffer from lung infection, so one important diagnostic method is to screen chest radiography images, e.g., X-Ray or CT images. However, such examinations are time-consuming and labor-intensive, leading to limited diagnostic efficiency. To solve this issue, AI-based technologies, such as deep learning, have been used recently as effective computer-aided means to improve diagnostic efficiency. However, one practical and critical difficulty is the limited availability of annotated COVID-19 data, due to the prohibitive annotation costs and urgent work of doctors to fight against the pandemic. This makes the learning of deep diagnosis models very challenging. To address this, motivated by that typical pneumonia has similar characteristics with COVID-19 and many pneumonia datasets are publicly available, we propose to conduct domain knowledge adaptation from typical pneumonia to COVID-19. There are two main challenges: 1) the discrepancy of data distributions between domains; 2) the task difference between the diagnosis of typical pneumonia and COVID-19. To address them, we propose a new deep domain adaptation method for COVID-19 diagnosis, namely COVID-DA. Specifically, we alleviate the domain discrepancy via feature adversarial adaptation and handle the task difference issue via a novel classifier separation scheme. In this way, COVID-DA is able to diagnose COVID-19 effectively with only a small number of COVID-19 annotations. Extensive experiments verify the effectiveness of COVID-DA and its great potential for real-world applications.

Yinghua Zhang, Yu Zhang, Ying Wei et al.

Deep domain adaptation models learn a neural network in an unlabeled target domain by leveraging the knowledge from a labeled source domain. This can be achieved by learning a domain-invariant feature space. Though the learned representations are separable in the source domain, they usually have a large variance and samples with different class labels tend to overlap in the target domain, which yields suboptimal adaptation performance. To fill the gap, a Fisher loss is proposed to learn discriminative representations which are within-class compact and between-class separable. Experimental results on two benchmark datasets show that the Fisher loss is a general and effective loss for deep domain adaptation. Noticeable improvements are brought when it is used together with widely adopted transfer criteria, including MMD, CORAL and domain adversarial loss. For example, an absolute improvement of 6.67% in terms of the mean accuracy is attained when the Fisher loss is used together with the domain adversarial loss on the Office-Home dataset.

Huaxiu Yao, Chuxu Zhang, Ying Wei et al.

Towards the challenging problem of semi-supervised node classification, there have been extensive studies. As a frontier, Graph Neural Networks (GNNs) have aroused great interest recently, which update the representation of each node by aggregating information of its neighbors. However, most GNNs have shallow layers with a limited receptive field and may not achieve satisfactory performance especially when the number of labeled nodes is quite small. To address this challenge, we innovatively propose a graph few-shot learning (GFL) algorithm that incorporates prior knowledge learned from auxiliary graphs to improve classification accuracy on the target graph. Specifically, a transferable metric space characterized by a node embedding and a graph-specific prototype embedding function is shared between auxiliary graphs and the target, facilitating the transfer of structural knowledge. Extensive experiments and ablation studies on four real-world graph datasets demonstrate the effectiveness of our proposed model.

Ying Wei, Peilin Zhao, Huaxiu Yao et al.

Automated machine learning aims to automate the whole process of machine learning, including model configuration. In this paper, we focus on automated hyperparameter optimization (HPO) based on sequential model-based optimization (SMBO). Though conventional SMBO algorithms work well when abundant HPO trials are available, they are far from satisfactory in practical applications where a trial on a huge dataset may be so costly that an optimal hyperparameter configuration is expected to return in as few trials as possible. Observing that human experts draw on their expertise in a machine learning model by trying configurations that once performed well on other datasets, we are inspired to speed up HPO by transferring knowledge from historical HPO trials on other datasets. We propose an end-to-end and efficient HPO algorithm named as Transfer Neural Processes (TNP), which achieves transfer learning by incorporating trials on other datasets, initializing the model with well-generalized parameters, and learning an initial set of hyperparameters to evaluate. Experiments on extensive OpenML datasets and three computer vision datasets show that the proposed model can achieve state-of-the-art performance in at least one order of magnitude less trials.

Huaxiu Yao, Ying Wei, Junzhou Huang et al.

In order to learn quickly with few samples, meta-learning utilizes prior knowledge learned from previous tasks. However, a critical challenge in meta-learning is task uncertainty and heterogeneity, which can not be handled via globally sharing knowledge among tasks. In this paper, based on gradient-based meta-learning, we propose a hierarchically structured meta-learning (HSML) algorithm that explicitly tailors the transferable knowledge to different clusters of tasks. Inspired by the way human beings organize knowledge, we resort to a hierarchical task clustering structure to cluster tasks. As a result, the proposed approach not only addresses the challenge via the knowledge customization to different clusters of tasks, but also preserves knowledge generalization among a cluster of similar tasks. To tackle the changing of task relationship, in addition, we extend the hierarchical structure to a continual learning environment. The experimental results show that our approach can achieve state-of-the-art performance in both toy-regression and few-shot image classification problems.

Huaxiu Yao, Yiding Liu, Ying Wei et al.

Spatial-temporal prediction is a fundamental problem for constructing smart city, which is useful for tasks such as traffic control, taxi dispatching, and environmental policy making. Due to data collection mechanism, it is common to see data collection with unbalanced spatial distributions. For example, some cities may release taxi data for multiple years while others only release a few days of data; some regions may have constant water quality data monitored by sensors whereas some regions only have a small collection of water samples. In this paper, we tackle the problem of spatial-temporal prediction for the cities with only a short period of data collection. We aim to utilize the long-period data from other cities via transfer learning. Different from previous studies that transfer knowledge from one single source city to a target city, we are the first to leverage information from multiple cities to increase the stability of transfer. Specifically, our proposed model is designed as a spatial-temporal network with a meta-learning paradigm. The meta-learning paradigm learns a well-generalized initialization of the spatial-temporal network, which can be effectively adapted to target cities. In addition, a pattern-based spatial-temporal memory is designed to distill long-term temporal information (i.e., periodicity). We conduct extensive experiments on two tasks: traffic (taxi and bike) prediction and water quality prediction. The experiments demonstrate the effectiveness of our proposed model over several competitive baseline models.

Zheng Li, Ying Wei, Yu Zhang et al.

Aspect-level sentiment classification (ASC) aims at identifying sentiment polarities towards aspects in a sentence, where the aspect can behave as a general Aspect Category (AC) or a specific Aspect Term (AT). However, due to the especially expensive and labor-intensive labeling, existing public corpora in AT-level are all relatively small. Meanwhile, most of the previous methods rely on complicated structures with given scarce data, which largely limits the efficacy of the neural models. In this paper, we exploit a new direction named coarse-to-fine task transfer, which aims to leverage knowledge learned from a rich-resource source domain of the coarse-grained AC task, which is more easily accessible, to improve the learning in a low-resource target domain of the fine-grained AT task. To resolve both the aspect granularity inconsistency and feature mismatch between domains, we propose a Multi-Granularity Alignment Network (MGAN). In MGAN, a novel Coarse2Fine attention guided by an auxiliary task can help the AC task modeling at the same fine-grained level with the AT task. To alleviate the feature false alignment, a contrastive feature alignment method is adopted to align aspect-specific feature representations semantically. In addition, a large-scale multi-domain dataset for the AC task is provided. Empirically, extensive experiments demonstrate the effectiveness of the MGAN.

Yu Zhang, Ying Wei, Qiang Yang

Multitask learning has shown promising performance in many applications and many multitask models have been proposed. In order to identify an effective multitask model for a given multitask problem, we propose a learning framework called learning to multitask (L2MT). To achieve the goal, L2MT exploits historical multitask experience which is organized as a training set consists of several tuples, each of which contains a multitask problem with multiple tasks, a multitask model, and the relative test error. Based on such training set, L2MT first uses a proposed layerwise graph neural network to learn task embeddings for all the tasks in a multitask problem and then learns an estimation function to estimate the relative test error based on task embeddings and the representation of the multitask model based on a unified formulation. Given a new multitask problem, the estimation function is used to identify a suitable multitask model. Experiments on benchmark datasets show the effectiveness of the proposed L2MT framework.

Ying Wei, Yu Zhang, Qiang Yang

Transfer learning borrows knowledge from a source domain to facilitate learning in a target domain. Two primary issues to be addressed in transfer learning are what and how to transfer. For a pair of domains, adopting different transfer learning algorithms results in different knowledge transferred between them. To discover the optimal transfer learning algorithm that maximally improves the learning performance in the target domain, researchers have to exhaustively explore all existing transfer learning algorithms, which is computationally intractable. As a trade-off, a sub-optimal algorithm is selected, which requires considerable expertise in an ad-hoc way. Meanwhile, it is widely accepted in educational psychology that human beings improve transfer learning skills of deciding what to transfer through meta-cognitive reflection on inductive transfer learning practices. Motivated by this, we propose a novel transfer learning framework known as Learning to Transfer (L2T) to automatically determine what and how to transfer are the best by leveraging previous transfer learning experiences. We establish the L2T framework in two stages: 1) we first learn a reflection function encrypting transfer learning skills from experiences; and 2) we infer what and how to transfer for a newly arrived pair of domains by optimizing the reflection function. Extensive experiments demonstrate the L2T's superiority over several state-of-the-art transfer learning algorithms and its effectiveness on discovering more transferable knowledge.