Wu-Jun Li

Hao Lin, Ke Wu, Jie Li et al.

Distributed learning is commonly used for training deep learning models, especially large models. In distributed learning, manual parallelism (MP) methods demand considerable human effort and have limited flexibility. Hence, automatic parallelism (AP) methods have recently been proposed for automating the parallel strategy optimization process. Existing AP methods suffer from sub-optimal solutions because they do not jointly optimize the two categories of parallel strategies (i.e., inter-layer parallelism and intra-layer parallelism). In this paper, we propose a novel AP method called UniAP, which unifies inter- and intra-layer automatic parallelism by mixed integer quadratic programming. To the best of our knowledge, UniAP is the first parallel method that can jointly optimize the two categories of parallel strategies to find an optimal solution. Experimental results show that UniAP outperforms state-of-the-art methods by up to 3.80$\times$ in throughput and reduces strategy optimization time by up to 107$\times$ across five Transformer-based models.

Yan-Shuo Liang, Wu-Jun Li

Online continual learning, especially when task identities and task boundaries are unavailable, is a challenging continual learning setting. One representative kind of methods for online continual learning is replay-based methods, in which a replay buffer called memory is maintained to keep a small part of past samples for overcoming catastrophic forgetting. When tackling with online continual learning, most existing replay-based methods focus on single-label problems in which each sample in the data stream has only one label. But multi-label problems may also happen in the online continual learning setting in which each sample may have more than one label. In the online setting with multi-label samples, the class distribution in data stream is typically highly imbalanced, and it is challenging to control class distribution in memory since changing the number of samples belonging to one class may affect the number of samples belonging to other classes. But class distribution in memory is critical for replay-based memory to get good performance, especially when the class distribution in data stream is highly imbalanced. In this paper, we propose a simple but effective method, called optimizing class distribution in memory (OCDM), for multi-label online continual learning. OCDM formulates the memory update mechanism as an optimization problem and updates the memory by solving this problem. Experiments on two widely used multi-label datasets show that OCDM can control the class distribution in memory well and can outperform other state-of-the-art methods.

Yi-Rui Yang, Kun Wang, Wu-Jun Li

Federated learning (FL) has recently become a hot research topic, in which Byzantine robustness, communication efficiency and privacy preservation are three important aspects. However, the tension among these three aspects makes it hard to simultaneously take all of them into account. In view of this challenge, we theoretically analyze the conditions that a communication compression method should satisfy to be compatible with existing Byzantine-robust methods and privacy-preserving methods. Motivated by the analysis results, we propose a novel communication compression method called consensus sparsification (ConSpar). To the best of our knowledge, ConSpar is the first communication compression method that is designed to be compatible with both Byzantine-robust methods and privacy-preserving methods. Based on ConSpar, we further propose a novel FL framework called FedREP, which is Byzantine-robust, communication-efficient and privacy-preserving. We theoretically prove the Byzantine robustness and the convergence of FedREP. Empirical results show that FedREP can significantly outperform communication-efficient privacy-preserving baselines. Furthermore, compared with Byzantine-robust communication-efficient baselines, FedREP can achieve comparable accuracy with the extra advantage of privacy preservation.

Yize Jiang, Xinze Li, Yuanyuan Zhang et al.

Existing protein-ligand docking studies typically focus on the self-docking scenario, which is less practical in real applications. Moreover, some studies involve heavy frameworks requiring extensive training, posing challenges for convenient and efficient assessment of docking methods. To fill these gaps, we design PoseX, an open-source benchmark to evaluate both self-docking and cross-docking, enabling a practical and comprehensive assessment of algorithmic advances. Specifically, we curated a novel dataset comprising 718 entries for self-docking and 1,312 entries for cross-docking; second, we incorporated 23 docking methods in three methodological categories, including physics-based methods (e.g., Schrödinger Glide), AI docking methods (e.g., DiffDock) and AI co-folding methods (e.g., AlphaFold3); third, we developed a relaxation method for post-processing to minimize conformational energy and refine binding poses; fourth, we built a leaderboard to rank submitted models in real-time. We derived some key insights and conclusions from extensive experiments: (1) AI approaches have consistently outperformed physics-based methods in overall docking success rate. (2) Most intra- and intermolecular clashes of AI approaches can be greatly alleviated with relaxation, which means combining AI modeling with physics-based post-processing could achieve excellent performance. (3) AI co-folding methods exhibit ligand chirality issues, except for Boltz-1x, which introduced physics-inspired potentials to fix hallucinations, suggesting modeling on stereochemistry improves the structural plausibility markedly. (4) Specifying binding pockets significantly promotes docking performance, indicating that pocket information can be leveraged adequately, particularly for AI co-folding methods, in future modeling efforts. The code, dataset, and leaderboard are released at https://github.com/CataAI/PoseX.

Yang-Fan Zhou, Kai-Lang Yao, Wu-Jun Li

Cytopathology report generation is a necessary step for the standardized examination of pathology images. However, manually writing detailed reports brings heavy workloads for pathologists. To improve efficiency, some existing works have studied automatic generation of cytopathology reports, mainly by applying image caption generation frameworks with visual encoders originally proposed for natural images. A common weakness of these works is that they do not explicitly model the structural information among cells, which is a key feature of pathology images and provides significant information for making diagnoses. In this paper, we propose a novel graph-based framework called GNNFormer, which seamlessly integrates graph neural network (GNN) and Transformer into the same framework, for cytopathology report generation. To the best of our knowledge, GNNFormer is the first report generation method that explicitly models the structural information among cells in pathology images. It also effectively fuses structural information among cells, fine-grained morphology features of cells and background features to generate high-quality reports. Experimental results on the NMI-WSI dataset show that GNNFormer can outperform other state-of-the-art baselines.

Chang-Wei Shi, Yi-Rui Yang, Wu-Jun Li

Distributed learning is essential for training large-scale deep models. Asynchronous SGD (ASGD) and its variants are commonly used distributed learning methods, particularly in scenarios where the computing capabilities of workers in the cluster are heterogeneous. Momentum has been acknowledged for its benefits in both optimization and generalization in deep model training. However, existing works have found that naively incorporating momentum into ASGD can impede the convergence. In this paper, we propose a novel method called ordered momentum (OrMo) for ASGD. In OrMo, momentum is incorporated into ASGD by organizing the gradients in order based on their iteration indexes. We theoretically prove the convergence of OrMo with both constant and delay-adaptive learning rates for non-convex problems. To the best of our knowledge, this is the first work to establish the convergence analysis of ASGD with momentum without dependence on the maximum delay. Empirical results demonstrate that OrMo can achieve better convergence performance compared with ASGD and other asynchronous methods with momentum.

Kai-Lang Yao, Wu-Jun Li

Link prediction is a fundamental problem in many graph based applications, such as protein-protein interaction prediction. Graph neural network (GNN) has recently been widely used for link prediction. However, existing GNN based link prediction (GNN-LP) methods suffer from scalability problem during training for large-scale graphs, which has received little attention by researchers. In this paper, we first give computation complexity analysis of existing GNN-LP methods, which reveals that the scalability problem stems from their symmetric learning strategy adopting the same class of GNN models to learn representation for both head and tail nodes. Then we propose a novel method, called asymmetric learning (AML), for GNN-LP. The main idea of AML is to adopt a GNN model for learning head node representation while using a multi-layer perceptron (MLP) model for learning tail node representation. Furthermore, AML proposes a row-wise sampling strategy to generate mini-batch for training, which is a necessary component to make the asymmetric learning strategy work for training speedup. To the best of our knowledge, AML is the first GNN-LP method adopting an asymmetric learning strategy for node representation learning. Experiments on three real large-scale datasets show that AML is 1.7X~7.3X faster in training than baselines with a symmetric learning strategy, while having almost no accuracy loss.

Hanqun Cao, Hongrui Zhang, Junde Xu et al.

Protein language models (PLMs) have advanced computational protein science through large-scale pretraining and scalable architectures. In parallel, reinforcement learning (RL) has broadened exploration and enabled precise multi-objective optimization in protein design. Yet whether RL can push PLMs beyond their pretraining priors to uncover latent sequence-structure-function rules remains unclear. We address this by pairing RL with PLMs across four domains: antimicrobial peptide design, kinase variant optimization, antibody engineering, and inverse folding. Using diverse RL algorithms and model classes, we ask if RL improves sampling efficiency and, more importantly, if it reveals capabilities not captured by supervised learning. Across benchmarks, RL consistently boosts success rates and sample efficiency. Performance follows a three-factor interaction: task headroom, reward fidelity, and policy capacity jointly determine gains. When rewards are accurate and informative, policies have sufficient capacity, and tasks leave room beyond supervised baselines, improvements scale; when rewards are noisy or capacity is constrained, gains saturate despite exploration. This view yields practical guidance for RL in protein design: prioritize reward modeling and calibration before scaling policy size, match algorithm and regularization strength to task difficulty, and allocate capacity where marginal gains are largest. Implementation is available at https://github.com/chq1155/RL-PLM.

Huan-Yi Su, Ke Wu, Yu-Hao Huang et al.

Recently, many works have proposed various financial large language models (FinLLMs) by pre-training from scratch or fine-tuning open-sourced LLMs on financial corpora. However, existing FinLLMs exhibit unsatisfactory performance in understanding financial text when numeric variables are involved in questions. In this paper, we propose a novel LLM, called numeric-sensitive large language model (NumLLM), for Chinese finance. We first construct a financial corpus from financial textbooks which is essential for improving numeric capability of LLMs during fine-tuning. After that, we train two individual low-rank adaptation (LoRA) modules by fine-tuning on our constructed financial corpus. One module is for adapting general-purpose LLMs to financial domain, and the other module is for enhancing the ability of NumLLM to understand financial text with numeric variables. Lastly, we merge the two LoRA modules into the foundation model to obtain NumLLM for inference. Experiments on financial question-answering benchmark show that NumLLM can boost the performance of the foundation model and can achieve the best overall performance compared to all baselines, on both numeric and non-numeric questions.

Yan-Shuo Liang, Wu-Jun Li

Continual learning requires the model to learn multiple tasks sequentially. In continual learning, the model should possess the ability to maintain its performance on old tasks (stability) and the ability to adapt to new tasks continuously (plasticity). Recently, parameter-efficient fine-tuning (PEFT), which involves freezing a pre-trained model and injecting a small number of learnable parameters to adapt to downstream tasks, has gained increasing popularity in continual learning. Although existing continual learning methods based on PEFT have demonstrated superior performance compared to those not based on PEFT, most of them do not consider how to eliminate the interference of the new task on the old tasks, which inhibits the model from making a good trade-off between stability and plasticity. In this work, we propose a new PEFT method, called interference-free low-rank adaptation (InfLoRA), for continual learning. InfLoRA injects a small number of parameters to reparameterize the pre-trained weights and shows that fine-tuning these injected parameters is equivalent to fine-tuning the pre-trained weights within a subspace. Furthermore, InfLoRA designs this subspace to eliminate the interference of the new task on the old tasks, making a good trade-off between stability and plasticity. Experimental results show that InfLoRA outperforms existing state-of-the-art continual learning methods on multiple datasets.

Xiao-Fan Niu, Wu-Jun Li

Knowledge graph embedding aims at translating the knowledge graph into numerical representations by transforming the entities and relations into continuous low-dimensional vectors. Recently, many methods [1, 5, 3, 2, 6] have been proposed to deal with this problem, but existing single-thread implementations of them are time-consuming for large-scale knowledge graphs. Here, we design a unified parallel framework to parallelize these methods, which achieves a significant time reduction without influencing the accuracy. We name our framework as ParaGraphE, which provides a library for parallel knowledge graph embedding. The source code can be downloaded from https://github.com/LIBBLE/LIBBLE-MultiThread/tree/master/ParaGraphE .

Yu-Hao Huang, Chang Xu, Yueying Wu et al.

Time series generation models are crucial for applications like data augmentation and privacy preservation. Most existing time series generation models are typically designed to generate data from one specified domain. While leveraging data from other domain for better generalization is proved to work in other application areas, this approach remains challenging for time series modeling due to the large divergence in patterns among different real world time series categories. In this paper, we propose a multi-domain time series diffusion model with domain prompts, named TimeDP. In TimeDP, we utilize a time series semantic prototype module which defines time series prototypes to represent time series basis, each prototype vector serving as "word" representing some elementary time series feature. A prototype assignment module is applied to extract the extract domain specific prototype weights, for learning domain prompts as generation condition. During sampling, we extract "domain prompt" with few-shot samples from the target domain and use the domain prompts as condition to generate time series samples. Experiments demonstrate that our method outperforms baselines to provide the state-of-the-art in-domain generation quality and strong unseen domain generation capability.

Jin Han, Yun Hong, Wu-Jun Li

Virtual screening (VS) is a critical step in computer-aided drug discovery, aiming to identify molecules that bind to a specific target receptor like protein. Traditional VS methods, such as docking, are often too time-consuming for screening large-scale molecular databases. Recent advances in deep learning have demonstrated that learning vector representations for both proteins and molecules using contrastive learning can outperform traditional docking methods. However, given that target databases often contain billions of molecules, real-valued vector representations adopted by existing methods can still incur significant memory and time costs in VS. To address this problem, in this paper we propose a hashing-based contrastive learning method, called DrugHash, for VS. DrugHash treats VS as a retrieval task that uses efficient binary hash codes for retrieval. In particular, DrugHash designs a simple yet effective hashing strategy to enable end-to-end learning of binary hash codes for both protein and molecule modalities, which can dramatically reduce the memory and time costs with higher accuracy compared with existing methods. Experimental results show that DrugHash can outperform existing methods to achieve state-of-the-art accuracy, with a memory saving of 32$\times$ and a speed improvement of 3.5$\times$.

Jin Han, Tianfan Fu, Wu-Jun Li

Protein inverse folding, the design of an amino acid sequence based on a target 3D structure, is a fundamental problem of computational protein engineering. Existing methods either generate sequences without leveraging external knowledge or relying on protein language models (PLMs). The former omits the evolutionary information stored in protein databases, while the latter is parameter-inefficient and inflexible to adapt to ever-growing protein data. To overcome the above drawbacks, in this paper we propose a novel method, called retrieval-augmented denoising diffusion (RadDiff), for protein inverse folding. Given the target protein backbone, RadDiff uses a hierarchical search strategy to efficiently retrieve structurally similar proteins from large protein databases. The retrieved structures are then aligned residue-by-residue to the target to construct a position-specific amino acid profile, which serves as an evolutionary-informed prior that conditions the denoising process. A lightweight integration module is further designed to incorporate this prior effectively. Experimental results on the CATH, PDB, and TS50 datasets show that RadDiff consistently outperforms existing methods, improving sequence recovery rate by up to 19%. Experimental results also demonstrate that RadDiff generates highly foldable sequences and scales effectively with database size.

Yan-Shuo Liang, Jia-Rui Chen, Wu-Jun Li

Continual learning (CL), which requires the model to learn multiple tasks sequentially, is crucial for large language models (LLMs). Recently, low-rank adaptation~(LoRA), one of the most representative parameter-efficient fine-tuning (PEFT) methods, has gained increasing attention in CL of LLMs. However, most existing CL methods based on LoRA typically expand a new LoRA branch to learn each new task and force the new and old LoRA branches to influence old tasks equally, potentially leading to forgetting. In this work, we propose a new method, called gated integration of low-rank adaptation (GainLoRA), for CL of LLMs. GainLoRA expands a new LoRA branch for each new task and introduces gating modules to integrate the new and old LoRA branches. Furthermore, GainLoRA leverages the new gating module to minimize the influence from the new LoRA branch to old tasks, effectively mitigating forgetting and improving the model's overall performance. Experimental results on CL benchmarks demonstrate that GainLoRA outperforms existing state-of-the-art methods.

Yangyijian Liu, Jun Li, Wu-Jun Li

The high memory and computation demand of large language models (LLMs) makes them challenging to be deployed on consumer devices due to limited GPU memory. Offloading can mitigate the memory constraint but often suffers from low GPU utilization, leading to low inference efficiency. In this work, we propose a novel framework, called pipelined offloading (PIPO), for efficient inference on consumer devices. PIPO designs a fine-grained offloading pipeline, complemented with optimized data transfer and computation, to achieve high concurrency and efficient scheduling for inference. Experimental results show that compared with state-of-the-art baseline, PIPO increases GPU utilization from below 40% to over 90% and achieves up to 3.1$\times$ higher throughput, running on a laptop equipped with a RTX3060 GPU of 6GB memory.

Jin Han, Wu-Jun Li

Protein structure similarity search (PSSS), which tries to search proteins with similar structures, plays a crucial role across diverse domains from drug design to protein function prediction and molecular evolution. Traditional alignment-based PSSS methods, which directly calculate alignment on the protein structures, are highly time-consuming with high memory cost. Recently, alignment-free methods, which represent protein structures as fixed-length real-valued vectors, are proposed for PSSS. Although these methods have lower time and memory cost than alignment-based methods, their time and memory cost is still too high for large-scale PSSS, and their accuracy is unsatisfactory. In this paper, we propose a novel method, called $\underline{\text{p}}$r$\underline{\text{o}}$tein $\underline{\text{s}}$tructure $\underline{\text{h}}$ashing (POSH), for PSSS. POSH learns a binary vector representation for each protein structure, which can dramatically reduce the time and memory cost for PSSS compared with real-valued vector representation based methods. Furthermore, in POSH we also propose expressive hand-crafted features and a structure encoder to well model both node and edge interactions in proteins. Experimental results on real datasets show that POSH can outperform other methods to achieve state-of-the-art accuracy. Furthermore, POSH achieves a memory saving of more than six times and speed improvement of more than four times, compared with other methods.

Kun Wang, Yi-Rui Yang, Wu-Jun Li

Asynchronous federated learning (AFL) is an effective method to address the challenge of device heterogeneity in cross-device federated learning. However, AFL is usually incompatible with existing secure aggregation protocols used to protect user privacy in federated learning because most existing secure aggregation protocols are based on synchronous aggregation. To address this problem, we propose a novel secure aggregation protocol named buffered asynchronous secure aggregation (BASA) in this paper. Compared with existing protocols, BASA is fully compatible with AFL and provides secure aggregation under the condition that each user only needs one round of communication with the server without relying on any synchronous interaction among users. Based on BASA, we propose the first AFL method which achieves secure aggregation without extra requirements on hardware. We empirically demonstrate that BASA outperforms existing secure aggregation protocols for cross-device federated learning in terms of training efficiency and scalability.

Yi-Rui Yang, Chang-Wei Shi, Wu-Jun Li

Byzantine-robust distributed learning (BRDL), in which computing devices are likely to behave abnormally due to accidental failures or malicious attacks, has recently become a hot research topic. However, even in the independent and identically distributed (i.i.d.) case, existing BRDL methods will suffer from a significant drop on model accuracy due to the large variance of stochastic gradients. Increasing batch sizes is a simple yet effective way to reduce the variance. However, when the total number of gradient computation is fixed, a too-large batch size will lead to a too-small iteration number (update number), which may also degrade the model accuracy. In view of this challenge, we mainly study the optimal batch size when the total number of gradient computation is fixed in this work. In particular, we theoretically and empirically show that when the total number of gradient computation is fixed, the optimal batch size in BRDL increases with the fraction of Byzantine workers. Therefore, compared to the case without attacks, the batch size should be set larger when under Byzantine attacks. However, for existing BRDL methods, large batch sizes will lead to a drop on model accuracy, even if there is no Byzantine attack. To deal with this problem, we propose a novel BRDL method, called Byzantine-robust stochastic gradient descent with normalized momentum (ByzSGDnm), which can alleviate the drop on model accuracy in large-batch cases. Moreover, we theoretically prove the convergence of ByzSGDnm for general non-convex cases under Byzantine attacks. Empirical results show that ByzSGDnm has a comparable performance to existing BRDL methods under bit-flipping failure, but can outperform existing BRDL methods under deliberately crafted attacks.

Xiao Ma, Wu-Jun Li

Multi-agent reinforcement learning (MARL) algorithms have made promising progress in recent years by leveraging the centralized training and decentralized execution (CTDE) paradigm. However, existing MARL algorithms still suffer from the sample inefficiency problem. In this paper, we propose a simple yet effective approach, called state-based episodic memory (SEM), to improve sample efficiency in MARL. SEM adopts episodic memory (EM) to supervise the centralized training procedure of CTDE in MARL. To the best of our knowledge, SEM is the first work to introduce EM into MARL. We can theoretically prove that, when using for MARL, SEM has lower space complexity and time complexity than state and action based EM (SAEM), which is originally proposed for single-agent reinforcement learning. Experimental results on StarCraft multi-agent challenge (SMAC) show that introducing episodic memory into MARL can improve sample efficiency and SEM can reduce storage cost and time cost compared with SAEM.

Zhao-Heng Yin, Lingfeng Sun, Hengbo Ma et al.

Animals are able to imitate each others' behavior, despite their difference in biomechanics. In contrast, imitating the other similar robots is a much more challenging task in robotics. This problem is called cross domain imitation learning~(CDIL). In this paper, we consider CDIL on a class of similar robots. We tackle this problem by introducing an imitation learning algorithm based on invariant representation. We propose to learn invariant state and action representations, which aligns the behavior of multiple robots so that CDIL becomes possible. Compared with previous invariant representation learning methods for similar purpose, our method does not require human-labeled pairwise data for training. Instead, we use cycle-consistency and domain confusion to align the representation and increase its robustness. We test the algorithm on multiple robots in simulator and show that unseen new robot instances can be trained with existing expert demonstrations successfully. Qualitative results also demonstrate that the proposed method is able to learn similar representations for different robots with similar behaviors, which is essential for successful CDIL.

Ming-Wei Li, Qing-Yuan Jiang, Wu-Jun Li

Due to its low storage cost and fast query speed, hashing has been widely used in large-scale image retrieval tasks. Hash bucket search returns data points within a given Hamming radius to each query, which can enable search at a constant or sub-linear time cost. However, existing hashing methods cannot achieve satisfactory retrieval performance for hash bucket search in complex scenarios, since they learn only one hash code for each image. More specifically, by using one hash code to represent one image, existing methods might fail to put similar image pairs to the buckets with a small Hamming distance to the query when the semantic information of images is complex. As a result, a large number of hash buckets need to be visited for retrieving similar images, based on the learned codes. This will deteriorate the efficiency of hash bucket search. In this paper, we propose a novel hashing framework, called multiple code hashing (MCH), to improve the performance of hash bucket search. The main idea of MCH is to learn multiple hash codes for each image, with each code representing a different region of the image. Furthermore, we propose a deep reinforcement learning algorithm to learn the parameters in MCH. To the best of our knowledge, this is the first work that proposes to learn multiple hash codes for each image in image retrieval. Experiments demonstrate that MCH can achieve a significant improvement in hash bucket search, compared with existing methods that learn only one hash code for each image.

Quan Cui, Qing-Yuan Jiang, Xiu-Shen Wei et al.

Retrieving content relevant images from a large-scale fine-grained dataset could suffer from intolerably slow query speed and highly redundant storage cost, due to high-dimensional real-valued embeddings which aim to distinguish subtle visual differences of fine-grained objects. In this paper, we study the novel fine-grained hashing topic to generate compact binary codes for fine-grained images, leveraging the search and storage efficiency of hash learning to alleviate the aforementioned problems. Specifically, we propose a unified end-to-end trainable network, termed as ExchNet. Based on attention mechanisms and proposed attention constraints, it can firstly obtain both local and global features to represent object parts and whole fine-grained objects, respectively. Furthermore, to ensure the discriminative ability and semantic meaning's consistency of these part-level features across images, we design a local feature alignment approach by performing a feature exchanging operation. Later, an alternative learning algorithm is employed to optimize the whole ExchNet and then generate the final binary hash codes. Validated by extensive experiments, our proposal consistently outperforms state-of-the-art generic hashing methods on five fine-grained datasets, which shows our effectiveness. Moreover, compared with other approximate nearest neighbor methods, ExchNet achieves the best speed-up and storage reduction, revealing its efficiency and practicality.

Shen-Yi Zhao, Chang-Wei Shi, Yin-Peng Xie et al.

Stochastic gradient descent~(SGD) and its variants have been the dominating optimization methods in machine learning. Compared to SGD with small-batch training, SGD with large-batch training can better utilize the computational power of current multi-core systems such as graphics processing units~(GPUs) and can reduce the number of communication rounds in distributed training settings. Thus, SGD with large-batch training has attracted considerable attention. However, existing empirical results showed that large-batch training typically leads to a drop in generalization accuracy. Hence, how to guarantee the generalization ability in large-batch training becomes a challenging task. In this paper, we propose a simple yet effective method, called stochastic normalized gradient descent with momentum~(SNGM), for large-batch training. We prove that with the same number of gradient computations, SNGM can adopt a larger batch size than momentum SGD~(MSGD), which is one of the most widely used variants of SGD, to converge to an $ε$-stationary point. Empirical results on deep learning verify that when adopting the same large batch size, SNGM can achieve better test accuracy than MSGD and other state-of-the-art large-batch training methods.

Zhao-Heng Yin, Wu-Jun Li

Animals are able to discover the topological map (graph) of surrounding environment, which will be used for navigation. Inspired by this biological phenomenon, researchers have recently proposed to generate graph representation for Markov decision process (MDP) and use such graphs for planning in reinforcement learning (RL). However, existing graph generation methods suffer from many drawbacks. One drawback is that existing methods do not learn an abstraction for graphs, which results in high memory and computation cost. This drawback also makes generated graph non-robust, which degrades the planning performance. Another drawback is that existing methods cannot be used for facilitating exploration which is important in RL. In this paper, we propose a new method, called topological map abstraction (TOMA), for graph generation. TOMA can generate an abstract graph representation for MDP, which costs much less memory and computation cost than existing methods. Furthermore, TOMA can be used for facilitating exploration. In particular, we propose planning to explore, in which TOMA is used to accelerate exploration by guiding the agent towards unexplored states. A novel experience replay module called vertex memory is also proposed to improve exploration performance. Experimental results show that TOMA can outperform existing methods to achieve the state-of-the-art performance.

Yi-Rui Yang, Wu-Jun Li

Distributed learning has become a hot research topic due to its wide application in clusterbased large-scale learning, federated learning, edge computing and so on. Most traditional distributed learning methods typically assume no failure or attack. However, many unexpected cases, such as communication failure and even malicious attack, may happen in real applications. Hence, Byzantine learning (BL), which refers to distributed learning with failure or attack, has recently attracted much attention. Most existing BL methods are synchronous, which are impractical in some applications due to heterogeneous or offline workers. In these cases, asynchronous BL (ABL) is usually preferred. In this paper, we propose a novel method, called buffered asynchronous stochastic gradient descent (BASGD), for ABL. To the best of our knowledge, BASGD is the first ABL method that can resist non-omniscient attacks without storing any instances on server. Furthermore, we also propose an improved variant of BASGD, called BASGD with momentum (BASGDm), by introducing momentum into BASGD. BASGDm can resist both non-omniscient and omniscient attacks. Compared with those methods which need to store instances on server, BASGD and BASGDm have a wider scope of application. Both BASGD and BASGDm are compatible with various aggregation rules. Moreover, both BASGD and BASGDm are proved to be convergent and be able to resist failure or attack. Empirical results show that our methods significantly outperform existing ABL baselines when there exists failure or attack on workers.

Shen-Yi Zhao, Yin-Peng Xie, Wu-Jun Li

Existing research shows that the batch size can seriously affect the performance of stochastic gradient descent~(SGD) based learning, including training speed and generalization ability. A larger batch size typically results in less parameter updates. In distributed training, a larger batch size also results in less frequent communication. However, a larger batch size can make a generalization gap more easily. Hence, how to set a proper batch size for SGD has recently attracted much attention. Although some methods about setting batch size have been proposed, the batch size problem has still not been well solved. In this paper, we first provide theory to show that a proper batch size is related to the gap between initialization and optimum of the model parameter. Then based on this theory, we propose a novel method, called \underline{s}tagewise \underline{e}nlargement of \underline{b}atch \underline{s}ize~(\mbox{SEBS}), to set proper batch size for SGD. More specifically, \mbox{SEBS} adopts a multi-stage scheme, and enlarges the batch size geometrically by stage. We theoretically prove that, compared to classical stagewise SGD which decreases learning rate by stage, \mbox{SEBS} can reduce the number of parameter updates without increasing generalization error. SEBS is suitable for \mbox{SGD}, momentum \mbox{SGD} and AdaGrad. Empirical results on real data successfully verify the theories of \mbox{SEBS}. Furthermore, empirical results also show that SEBS can outperform other baselines.

Wen-Pu Cai, Wu-Jun Li

With the development of deep neural networks, the size of network models becomes larger and larger. Model compression has become an urgent need for deploying these network models to mobile or embedded devices. Model quantization is a representative model compression technique. Although a lot of quantization methods have been proposed, many of them suffer from a high quantization error caused by a long-tail distribution of network weights. In this paper, we propose a novel quantization method, called weight normalization based quantization (WNQ), for model compression. WNQ adopts weight normalization to avoid the long-tail distribution of network weights and subsequently reduces the quantization error. Experiments on CIFAR-100 and ImageNet show that WNQ can outperform other baselines to achieve state-of-the-art performance.

Shen-Yi Zhao, Hao Gao, Wu-Jun Li

Stochastic gradient decent~(SGD) and its variants, including some accelerated variants, have become popular for training in machine learning. However, in all existing SGD and its variants, the sample size in each iteration~(epoch) of training is the same as the size of the full training set. In this paper, we propose a new method, called \underline{ada}ptive \underline{s}ample \underline{s}election~(ADASS), for training acceleration. During different epoches of training, ADASS only need to visit different training subsets which are adaptively selected from the full training set according to the Lipschitz constants of the loss functions on samples. It means that in ADASS the sample size in each epoch of training can be smaller than the size of the full training set, by discarding some samples. ADASS can be seamlessly integrated with existing optimization methods, such as SGD and momentum SGD, for training acceleration. Theoretical results show that the learning accuracy of ADASS is comparable to that of counterparts with full training set. Furthermore, empirical results on both shallow models and deep models also show that ADASS can accelerate the training process of existing methods without sacrificing accuracy.

Xiao Ma, Shen-Yi Zhao, Wu-Jun Li

Exploration strategy design is one of the challenging problems in reinforcement learning~(RL), especially when the environment contains a large state space or sparse rewards. During exploration, the agent tries to discover novel areas or high reward~(quality) areas. In most existing methods, the novelty and quality in the neighboring area of the current state are not well utilized to guide the exploration of the agent. To tackle this problem, we propose a novel RL framework, called \underline{c}lustered \underline{r}einforcement \underline{l}earning~(CRL), for efficient exploration in RL. CRL adopts clustering to divide the collected states into several clusters, based on which a bonus reward reflecting both novelty and quality in the neighboring area~(cluster) of the current state is given to the agent. Experiments on a continuous control task and several \emph{Atari 2600} games show that CRL can outperform other state-of-the-art methods to achieve the best performance in most cases.

Shen-Yi Zhao, Hao Gao, Wu-Jun Li

Distributed stochastic gradient descent~(DSGD) has been widely used for optimizing large-scale machine learning models, including both convex and non-convex models. With the rapid growth of model size, huge communication cost has been the bottleneck of traditional DSGD. Recently, many communication compression methods have been proposed. Memory-based distributed stochastic gradient descent~(M-DSGD) is one of the efficient methods since each worker communicates a sparse vector in each iteration so that the communication cost is small. Recent works propose the convergence rate of M-DSGD when it adopts vanilla SGD. However, there is still a lack of convergence theory for M-DSGD when it adopts momentum SGD. In this paper, we propose a universal convergence analysis for M-DSGD by introducing \emph{transformation equation}. The transformation equation describes the relation between traditional DSGD and M-DSGD so that we can transform M-DSGD to its corresponding DSGD. Hence we get the convergence rate of M-DSGD with momentum for both convex and non-convex problems. Furthermore, we combine M-DSGD and stagewise learning that the learning rate of M-DSGD in each stage is a constant and is decreased by stage, instead of iteration. Using the transformation equation, we propose the convergence rate of stagewise M-DSGD which bridges the gap between theory and practice.

Chang-Wei Shi, Shen-Yi Zhao, Yin-Peng Xie et al.

With the rapid growth of data, distributed momentum stochastic gradient descent~(DMSGD) has been widely used in distributed learning, especially for training large-scale deep models. Due to the latency and limited bandwidth of the network, communication has become the bottleneck of distributed learning. Communication compression with sparsified gradient, abbreviated as \emph{sparse communication}, has been widely employed to reduce communication cost. All existing works about sparse communication in DMSGD employ local momentum, in which the momentum only accumulates stochastic gradients computed by each worker locally. In this paper, we propose a novel method, called \emph{\underline{g}}lobal \emph{\underline{m}}omentum \emph{\underline{c}}ompression~(GMC), for sparse communication. Different from existing works that utilize local momentum, GMC utilizes global momentum. Furthermore, to enhance the convergence performance when using more aggressive sparsification compressors (e.g., RBGS), we extend GMC to GMC+. We theoretically prove the convergence of GMC and GMC+. To the best of our knowledge, this is the first work that introduces global momentum for sparse communication in distributed learning. Empirical results demonstrate that, compared with the local momentum counterparts, our GMC and GMC+ can achieve higher test accuracy and exhibit faster convergence, especially under non-IID data distribution.

Ming-Wei Li, Qing-Yuan Jiang, Wu-Jun Li

Traditional person re-identification (ReID) methods typically represent person images as real-valued features, which makes ReID inefficient when the gallery set is extremely large. Recently, some hashing methods have been proposed to make ReID more efficient. However, these hashing methods will deteriorate the accuracy in general, and the efficiency of them is still not high enough. In this paper, we propose a novel hashing method, called deep multi-index hashing (DMIH), to improve both efficiency and accuracy for ReID. DMIH seamlessly integrates multi-index hashing and multi-branch based networks into the same framework. Furthermore, a novel block-wise multi-index hashing table construction approach and a search-aware multi-index (SAMI) loss are proposed in DMIH to improve the search efficiency. Experiments on three widely used datasets show that DMIH can outperform other state-of-the-art baselines, including both hashing methods and real-valued methods, in terms of both efficiency and accuracy.

Qing-Yuan Jiang, Ming-Wei Li, Wu-Jun Li

Due to its low storage cost and fast query speed, hashing has been widely used for large-scale approximate nearest neighbor (ANN) search. Bucket search, also called hash lookup, can achieve fast query speed with a sub-linear time cost based on the inverted index table constructed from hash codes. Many metrics have been adopted to evaluate hashing algorithms. However, all existing metrics are improper to evaluate the hash codes for bucket search. On one hand, all existing metrics ignore the retrieval time cost which is an important factor reflecting the performance of search. On the other hand, some of them, such as mean average precision (MAP), suffer from the uncertainty problem as the ranked list is based on integer-valued Hamming distance, and are insensitive to Hamming radius as these metrics only depend on relative Hamming distance. Other metrics, such as precision at Hamming radius R, fail to evaluate global performance as these metrics only depend on one specific Hamming radius. In this paper, we first point out the problems of existing metrics which have been ignored by the hashing community, and then propose a novel evaluation metric called radius aware mean average precision (RAMAP) to evaluate hash codes for bucket search. Furthermore, two coding strategies are also proposed to qualitatively show the problems of existing metrics. Experiments demonstrate that our proposed RAMAP can provide more proper evaluation than existing metrics.

Kai-Lang Yao, Wu-Jun Li

Recommender systems (RS), which have been an essential part in a wide range of applications, can be formulated as a matrix completion (MC) problem. To boost the performance of MC, matrix completion with side information, called inductive matrix completion (IMC), was further proposed. In real applications, the factorized version of IMC is more favored due to its efficiency of optimization and implementation. Regarding the factorized version, traditional IMC method can be interpreted as learning an individual representation for each feature, which is independent from each other. Moreover, representations for the same features are shared across all users/items. However, the independent characteristic for features and shared characteristic for the same features across all users/items may limit the expressiveness of the model. The limitation also exists in variants of IMC, such as deep learning based IMC models. To break the limitation, we generalize recent advances of self-attention mechanism to IMC and propose a context-aware model called collaborative self-attention (CSA), which can jointly learn context-aware representations for features and perform inductive matrix completion process. Extensive experiments on three large-scale datasets from real RS applications demonstrate effectiveness of CSA.

Dong Xu, Jianhui Ji, Haikuan Huang et al.

Answer selection (answer ranking) is one of the key steps in many kinds of question answering (QA) applications, where deep models have achieved state-of-the-art performance. Among these deep models, recurrent neural network (RNN) based models are most popular, typically with better performance than convolutional neural network (CNN) based models. Nevertheless, it is difficult for RNN based models to capture the information about long-range dependency among words in the sentences of questions and answers. In this paper, we propose a new deep model, called gated group self-attention (GGSA), for answer selection. GGSA is inspired by global self-attention which is originally proposed for machine translation and has not been explored in answer selection. GGSA tackles the problem of global self-attention that local and global information cannot be well distinguished. Furthermore, an interaction mechanism between questions and answers is also proposed to enhance GGSA by a residual structure. Experimental results on two popular QA datasets show that GGSA can outperform existing answer selection models to achieve state-of-the-art performance. Furthermore, GGSA can also achieve higher accuracy than global self-attention for the answer selection task, with a lower computation cost.

Dong Xu, Wu-Jun Li

Answer selection is an important subtask of question answering (QA), where deep models usually achieve better performance. Most deep models adopt question-answer interaction mechanisms, such as attention, to get vector representations for answers. When these interaction based deep models are deployed for online prediction, the representations of all answers need to be recalculated for each question. This procedure is time-consuming for deep models with complex encoders like BERT which usually have better accuracy than simple encoders. One possible solution is to store the matrix representation (encoder output) of each answer in memory to avoid recalculation. But this will bring large memory cost. In this paper, we propose a novel method, called hashing based answer selection (HAS), to tackle this problem. HAS adopts a hashing strategy to learn a binary matrix representation for each answer, which can dramatically reduce the memory cost for storing the matrix representations of answers. Hence, HAS can adopt complex encoders like BERT in the model, but the online prediction of HAS is still fast with a low memory cost. Experimental results on three popular answer selection datasets show that HAS can outperform existing models to achieve state-of-the-art performance.

Shen-Yi Zhao, Hao Gao, Wu-Jun Li

Due to its efficiency and ease to implement, stochastic gradient descent (SGD) has been widely used in machine learning. In particular, SGD is one of the most popular optimization methods for distributed learning. Recently, quantized SGD (QSGD), which adopts quantization to reduce the communication cost in SGD-based distributed learning, has attracted much attention. Although several QSGD methods have been proposed, some of them are heuristic without theoretical guarantee, and others have high quantization variance which makes the convergence become slow. In this paper, we propose a new method, called Quantized Epoch-SGD (QESGD), for communication-efficient distributed learning. QESGD compresses (quantizes) the parameter with variance reduction, so that it can get almost the same performance as that of SGD with less communication cost. QESGD is implemented on the Parameter Server framework, and empirical results on distributed deep learning show that QESGD can outperform other state-of-the-art quantization methods to achieve the best performance.

Shen-Yi Zhao, Gong-Duo Zhang, Ming-Wei Li et al.

Distributed sparse learning with a cluster of multiple machines has attracted much attention in machine learning, especially for large-scale applications with high-dimensional data. One popular way to implement sparse learning is to use $L_1$ regularization. In this paper, we propose a novel method, called proximal \mbox{SCOPE}~(\mbox{pSCOPE}), for distributed sparse learning with $L_1$ regularization. pSCOPE is based on a \underline{c}ooperative \underline{a}utonomous \underline{l}ocal \underline{l}earning~(\mbox{CALL}) framework. In the \mbox{CALL} framework of \mbox{pSCOPE}, we find that the data partition affects the convergence of the learning procedure, and subsequently we define a metric to measure the goodness of a data partition. Based on the defined metric, we theoretically prove that pSCOPE is convergent with a linear convergence rate if the data partition is good enough. We also prove that better data partition implies faster convergence rate. Furthermore, pSCOPE is also communication efficient. Experimental results on real data sets show that pSCOPE can outperform other state-of-the-art distributed methods for sparse learning.

Kai-Lang Yao, Wu-Jun Li, Jianbo Yang et al.

Geometric matrix completion (GMC) has been proposed for recommendation by integrating the relationship (link) graphs among users/items into matrix completion (MC). Traditional GMC methods typically adopt graph regularization to impose smoothness priors for MC. Recently, geometric deep learning on graphs (GDLG) is proposed to solve the GMC problem, showing better performance than existing GMC methods including traditional graph regularization based methods. To the best of our knowledge, there exists only one GDLG method for GMC, which is called RMGCNN. RMGCNN combines graph convolutional network (GCN) and recurrent neural network (RNN) together for GMC. In the original work of RMGCNN, RMGCNN demonstrates better performance than pure GCN-based method. In this paper, we propose a new GMC method, called convolutional geometric matrix completion (CGMC), for recommendation with graphs among users/items. CGMC is a pure GCN-based method with a newly designed graph convolutional network. Experimental results on real datasets show that CGMC can outperform other state-of-the-art methods including RMGCNN in terms of both accuracy and speed.

Gong-Duo Zhang, Shen-Yi Zhao, Hao Gao et al.

Linear classification has been widely used in many high-dimensional applications like text classification. To perform linear classification for large-scale tasks, we often need to design distributed learning methods on a cluster of multiple machines. In this paper, we propose a new distributed learning method, called feature-distributed stochastic variance reduced gradient (FD-SVRG) for high-dimensional linear classification. Unlike most existing distributed learning methods which are instance-distributed, FD-SVRG is feature-distributed. FD-SVRG has lower communication cost than other instance-distributed methods when the data dimensionality is larger than the number of data instances. Experimental results on real data demonstrate that FD-SVRG can outperform other state-of-the-art distributed methods for high-dimensional linear classification in terms of both communication cost and wall-clock time, when the dimensionality is larger than the number of instances in training data.

Qing-Yuan Jiang, Xue Cui, Wu-Jun Li

Hashing has been widely used for large-scale search due to its low storage cost and fast query speed. By using supervised information, supervised hashing can significantly outperform unsupervised hashing. Recently, discrete supervised hashing and deep hashing are two representative progresses in supervised hashing. On one hand, hashing is essentially a discrete optimization problem. Hence, utilizing supervised information to directly guide discrete (binary) coding procedure can avoid sub-optimal solution and improve the accuracy. On the other hand, deep hashing, which integrates deep feature learning and hash-code learning into an end-to-end architecture, can enhance the feedback between feature learning and hash-code learning. The key in discrete supervised hashing is to adopt supervised information to directly guide the discrete coding procedure in hashing. The key in deep hashing is to adopt the supervised information to directly guide the deep feature learning procedure. However, there have not existed works which can use the supervised information to directly guide both discrete coding procedure and deep feature learning procedure in the same framework. In this paper, we propose a novel deep hashing method, called deep discrete supervised hashing (DDSH), to address this problem. DDSH is the first deep hashing method which can utilize supervised information to directly guide both discrete coding procedure and deep feature learning procedure, and thus enhance the feedback between these two important procedures. Experiments on three real datasets show that DDSH can outperform other state-of-the-art baselines, including both discrete hashing and deep hashing baselines, for image retrieval.

Qing-Yuan Jiang, Wu-Jun Li

Hashing has been widely used for large-scale approximate nearest neighbor search because of its storage and search efficiency. Recent work has found that deep supervised hashing can significantly outperform non-deep supervised hashing in many applications. However, most existing deep supervised hashing methods adopt a symmetric strategy to learn one deep hash function for both query points and database (retrieval) points. The training of these symmetric deep supervised hashing methods is typically time-consuming, which makes them hard to effectively utilize the supervised information for cases with large-scale database. In this paper, we propose a novel deep supervised hashing method, called asymmetric deep supervised hashing (ADSH), for large-scale nearest neighbor search. ADSH treats the query points and database points in an asymmetric way. More specifically, ADSH learns a deep hash function only for query points, while the hash codes for database points are directly learned. The training of ADSH is much more efficient than that of traditional symmetric deep supervised hashing methods. Experiments show that ADSH can achieve state-of-the-art performance in real applications.

Qing-Yuan Jiang, Wu-Jun Li

Due to its storage and retrieval efficiency, cross-modal hashing~(CMH) has been widely used for cross-modal similarity search in multimedia applications. According to the training strategy, existing CMH methods can be mainly divided into two categories: relaxation-based continuous methods and discrete methods. In general, the training of relaxation-based continuous methods is faster than discrete methods, but the accuracy of relaxation-based continuous methods is not satisfactory. On the contrary, the accuracy of discrete methods is typically better than relaxation-based continuous methods, but the training of discrete methods is time-consuming. In this paper, we propose a novel CMH method, called discrete latent factor model based cross-modal hashing~(DLFH), for cross modal similarity search. DLFH is a discrete method which can directly learn the binary hash codes for CMH. At the same time, the training of DLFH is efficient. Experiments on real datasets show that DLFH can achieve significantly better accuracy than existing methods, and the training time of DLFH is comparable to that of relaxation-based continuous methods which are much faster than existing discrete methods.

Luo Luo, Cheng Chen, Zhihua Zhang et al.

The frequent directions (FD) technique is a deterministic approach for online sketching that has many applications in machine learning. The conventional FD is a heuristic procedure that often outputs rank deficient matrices. To overcome the rank deficiency problem, we propose a new sketching strategy called robust frequent directions (RFD) by introducing a regularization term. RFD can be derived from an optimization problem. It updates the sketch matrix and the regularization term adaptively and jointly. RFD reduces the approximation error of FD without increasing the computational cost. We also apply RFD to online learning and propose an effective hyperparameter-free online Newton algorithm. We derive a regret bound for our online Newton algorithm based on RFD, which guarantees the robustness of the algorithm. The experimental studies demonstrate that the proposed method outperforms state-of-the-art second order online learning algorithms.

Shen-Yi Zhao, Gong-Duo Zhang, Wu-Jun Li

Stochastic gradient descent~(SGD) and its variants have attracted much attention in machine learning due to their efficiency and effectiveness for optimization. To handle large-scale problems, researchers have recently proposed several lock-free strategy based parallel SGD~(LF-PSGD) methods for multi-core systems. However, existing works have only proved the convergence of these LF-PSGD methods for convex problems. To the best of our knowledge, no work has proved the convergence of the LF-PSGD methods for non-convex problems. In this paper, we provide the theoretical proof about the convergence of two representative LF-PSGD methods, Hogwild! and AsySVRG, for non-convex problems. Empirical results also show that both Hogwild! and AsySVRG are convergent on non-convex problems, which successfully verifies our theoretical results.

Dong Xu, Wu-Jun Li

Recently, bidirectional recurrent neural network (BRNN) has been widely used for question answering (QA) tasks with promising performance. However, most existing BRNN models extract the information of questions and answers by directly using a pooling operation to generate the representation for loss or similarity calculation. Hence, these existing models don't put supervision (loss or similarity calculation) at every time step, which will lose some useful information. In this paper, we propose a novel BRNN model called full-time supervision based BRNN (FTS-BRNN), which can put supervision at every time step. Experiments on the factoid QA task show that our FTS-BRNN can outperform other baselines to achieve the state-of-the-art accuracy.

Qing-Yuan Jiang, Wu-Jun Li

Due to its low storage cost and fast query speed, cross-modal hashing (CMH) has been widely used for similarity search in multimedia retrieval applications. However, almost all existing CMH methods are based on hand-crafted features which might not be optimally compatible with the hash-code learning procedure. As a result, existing CMH methods with handcrafted features may not achieve satisfactory performance. In this paper, we propose a novel cross-modal hashing method, called deep crossmodal hashing (DCMH), by integrating feature learning and hash-code learning into the same framework. DCMH is an end-to-end learning framework with deep neural networks, one for each modality, to perform feature learning from scratch. Experiments on two real datasets with text-image modalities show that DCMH can outperform other baselines to achieve the state-of-the-art performance in cross-modal retrieval applications.

Luo Luo, Zihao Chen, Zhihua Zhang et al.

In this paper, we discuss the problem of minimizing the sum of two convex functions: a smooth function plus a non-smooth function. Further, the smooth part can be expressed by the average of a large number of smooth component functions, and the non-smooth part is equipped with a simple proximal mapping. We propose a proximal stochastic second-order method, which is efficient and scalable. It incorporates the Hessian in the smooth part of the function and exploits multistage scheme to reduce the variance of the stochastic gradient. We prove that our method can achieve linear rate of convergence.

Shen-Yi Zhao, Ru Xiang, Ying-Hao Shi et al.

Many machine learning models, such as logistic regression~(LR) and support vector machine~(SVM), can be formulated as composite optimization problems. Recently, many distributed stochastic optimization~(DSO) methods have been proposed to solve the large-scale composite optimization problems, which have shown better performance than traditional batch methods. However, most of these DSO methods are not scalable enough. In this paper, we propose a novel DSO method, called \underline{s}calable \underline{c}omposite \underline{op}timization for l\underline{e}arning~({SCOPE}), and implement it on the fault-tolerant distributed platform \mbox{Spark}. SCOPE is both computation-efficient and communication-efficient. Theoretical analysis shows that SCOPE is convergent with linear convergence rate when the objective function is convex. Furthermore, empirical results on real datasets show that SCOPE can outperform other state-of-the-art distributed learning methods on Spark, including both batch learning methods and DSO methods.