论文检索

Xing Lei, Xuetao Zhang, Donglin Wang

Recently, a state-of-the-art series of algorithms—Goal-Conditioned Weighted Supervised Learning (GCWSL) methods—has been introduced to address the challenges inherent in offline goal-conditioned reinforcement learning (RL). GCWSL optimizes a lower bound on the goal-conditioned RL objective and has demonstrated exceptional performance across a range of goal-reaching tasks, offering a simple, effective, and stable solution. Nonetheless, researches has revealed a critical limitation in GCWSL: the absence of trajectory stitching capabilities. In response, goal data augmentation strategies have been proposed to enhance these methods. However, existing techniques often fail to effectively sample appropriate augmented goals for GCWSL. In this paper, we establish unified principles for goal data augmentation, emphasizing goal diversity, action optimality, and goal reachability. Building on these principles, we propose a Model-based Goal Data Augmentation (MGDA) approach, which leverages a dynamics model to sample more appropriate augmented goals. MGDA uniquely incorporates the local Lipschitz continuity assumption within the learned model to mitigate the effects of compounding errors. Empirical results demonstrate that MGDA significantly improves the performance of GCWSL methods on both state-based and vision-based maze datasets, outperforming previous goal data augmentation techniques in their ability to enhancing stitching capabilities.

Shiqi Lei, Kanghoon Lee, Linjing Li, Jinkyoo Park

The offline datasets for imitation learning (IL) in multi-agent games typically contain player trajectories exhibiting diverse strategies, which necessitate measures to prevent learning algorithms from acquiring undesirable behaviors. Learning representations for these trajectories is an effective approach to depicting the strategies employed by each demonstrator. However, existing learning strategies often require player identification or rely on strong assumptions, which are not appropriate for multi-agent games. Therefore, in this paper, we introduce the Strategy Representation for Imitation Learning (STRIL) framework, which (1) effectively learns strategy representations in multi-agent games, (2) estimates proposed indicators based on these representations, and (3) filters out sub-optimal data using the indicators. STRIL is a plug-in method that can be integrated into existing IL algorithms. We demonstrate the effectiveness of STRIL across competitive multi-agent scenarios, including Two-player Pong, Limit Texas Hold'em, and Connect Four. Our approach successfully acquires strategy representations and indicators, thereby identifying dominant trajectories and significantly enhancing existing IL performance across these environments.

Wonjun Lee, Bumsub Ham, Suhyun Kim

In vision transformers, position embedding (PE) plays a crucial role in capturing the order of tokens. However, in vision transformer structures, there is a limitation in the expressiveness of PE due to the structure where position embedding is simply added to the token embedding. A layer-wise method that delivers PE to each layer and applies independent LNs for token embedding and PE has been adopted to overcome this limitation. In this paper, we identify the conflicting result that occurs in a layer-wise structure when using the global average pooling (GAP) method instead of the class token. To overcome this problem, we propose MPVG, which maximizes the effectiveness of PE in a layer-wise structure with GAP. Specifically, we identify that PE counterbalances token embedding values at each layer in a layer-wise structure. Furthermore, we recognize that the counterbalancing role of PE is insufficient in the layer-wise structure, and we address this by maximizing the effectiveness of PE through MPVG. Through experiments, we demonstrate that PE performs a counterbalancing role and that maintaining this counterbalancing directionality significantly impacts vision transformers. As a result, the experimental results show that MPVG outperforms existing methods across vision transformers on various tasks.

Seungcheol Lee, Moohong Min

Data sharing is necessary for AI to be widely used, but sharing sensitive data with others with privacy is risky. To solve these problems, it is necessary to synthesize realistic tabular data. In many cases, tabular data contains a mixture of continuous, mixed, categorical columns. Moreover, columns of the same type may have multimodal distribution or be highly imbalanced. These issues make it challenging to synthesize tabular data. The synthesized tabular data should reflect the relational meaning between columns of tabular data, so modeling the probability distribution of tabular data is a non-trivial task. Traditional tabular data synthesizing models are based on GAN or diffusion models and are built using fully connected or convolutional layers. However, fully connected layers have the disadvantage of low inductive bias, and convolutional layers are not invariant to the column order of tabular data. Therefore, we assume that converting tabular data into graph-structured data and using a graph neural network would produce better synthetic data than using fully connected layers or convolutional layers. Our study aims to show that GANs constructed with graph neural networks can outperform existing GAN models using fully connected layers or convolutional layers. We propose CG-TGAN, a conditional GAN built using graph neural networks. To learn how to synthesize realistic data, the graph neural networks in the discriminator and generator learn graph-level tasks and node-level tasks together. The discriminator of CG-TGAN learns a graph-level task to distinguish between real and synthetic data and node-level tasks to predict the value of the target node. CG-TGAN’s generator learns a graph-level task to synthesize an overall graph similar to real data and node-level tasks to learn how to synthesize a fake graph with the proper relation between nodes. In this paper, we show that CG-TGAN outperforms GAN-based models and is comparable to diffusion-based models.

Junyong Lee, Jeihee Cho, Shiho Kim

In the Noisy Intermediate-Scale Quantum (NISQ) era, using variational quantum algorithms (VQAs) to solve optimization problems has become a key application. However, these algorithms face significant challenges, such as choosing an effective initial set of parameters and the limited quantum processing time that restricts the number of optimization iterations. In this study, we introduce a new framework for optimizing parameterized quantum circuits (PQCs) that employs a classical optimizer, inspired by Model-Agnostic Meta-Learning (MAML) technique. This approach aim to achieve better parameter initialization that ensures fast convergence. Our framework features a classical neural network, called Learner, which interacts with a PQC using the output of Learner as an initial parameter. During the pre-training phase, Learner is trained with the meta objective function. In the adaptation phase, the framework requires only a few PQC updates to converge to a more accurate value, while the learner remains unchanged. This method is highly adaptable and is effectively extended to various Hamiltonian optimization problems. We validate our approach through experiments, including distribution function mapping and optimization of the Heisenberg XYZ Hamiltonian. The result implies that the Learner successfully estimates initial parameters that generalize across the problem space, enabling fast adaptation.

Junghun Lee, Hyunju Kim, Fanchen Bu, Jihoon Ko, Kijung Shin

In social networks, people influence each other through social links, which can be represented as propagation among nodes in graphs. Influence minimization (IMIN) is the problem of manipulating the structures of an input graph (e.g., removing edges) to reduce the propagation among nodes. IMIN can represent time-critical real-world applications, such as rumor blocking, but IMIN is theoretically difficult and computationally expensive. Moreover, the discrete nature of IMIN hinders the usage of powerful machine learning techniques, which requires differentiable computation. In this work, we propose DiffIM, a novel method for IMIN with two differentiable schemes for acceleration: (1) surrogate modeling for efficient influence estimation, which avoids time-consuming simulations (e.g., Monte Carlo), and (2) the continuous relaxation of decisions, which avoids the evaluation of individual discrete decisions (e.g., removing an edge). We further propose a third accelerating scheme, gradient-driven selection, that chooses edges instantly based on gradients without optimization (spec., gradient descent iterations) on each test instance. Through extensive experiments on real-world graphs, we show that each proposed scheme significantly improves speed with little (or even no) IMIN performance degradation. Our method is Pareto-optimal (i.e., no baseline is faster and more effective than it) and typically several orders of magnitude (spec., up to 15,160X) faster than the most effective baseline, while being more effective.

Jin-Seop Lee, Noo-ri Kim, Jee-Hyong Lee

Self-supervised learning (SSL) methods based on the instance discrimination tasks with InfoNCE have achieved remarkable success. Despite their success, SSL models often struggle to generate effective representations for unseen-domain data. To address this issue, research on unsupervised domain generalization (UDG), which aims to develop SSL models that can generate domain-irrelevant features, has been conducted. Most UDG approaches utilize contrastive learning with InfoNCE to generate representations, and perform feature alignment based on strong assumptions to generalize domain-irrelevant common features from multi-source domains. However, existing methods that rely on instance discrimination tasks are not effective at extracting domain-irrelevant common features. This leads to the suppression of domain-irrelevant common features and the amplification of domain-relevant features, thereby hindering domain generalization. Furthermore, strong assumptions underlying feature alignment can lead to biased feature learning, reducing the diversity of common features. In this paper, we propose a novel approach, DomCLP, Domain-wise Contrastive Learning with Prototype Mixup. We explore how InfoNCE suppresses domain-irrelevant common features and amplifies domain-relevant features. Based on this analysis, we propose Domain-wise Contrastive Learning (DCon) to enhance domain-irrelevant common features. We also propose Prototype Mixup Learning (PMix) to generalize domain-irrelevant common features across multiple domains without relying on strong assumptions. The proposed method consistently outperforms state-of-the-art methods on the PACS and DomainNet datasets across various label fractions, showing significant improvements.

Jeong Woon Lee, Hyoseok Hwang

Visual Reinforcement Learning (RL) facilitates learning directly from raw images; however, the domain gap between training and testing environments frequently leads to a decline in performance within unseen environments. In this paper, we propose Fourier Guided Adaptive Adversarial Augmentation (FGA3), a novel augmentation method that maintains semantic consistency. We focus on style augmentation in the frequency domain by keeping the phase and altering the amplitude to preserve the state of the original data. For adaptive adversarial perturbation, we reformulate the worst-case problem to RL by employing adversarial example training, which leverages value loss and cosine similarity within a semantic space. Moreover, our findings illustrate that cosine similarity is effective in quantifying feature distances within a semantic space. Extensive experiments on DMControl-GB and Procgen have shown that FGA3 is compatible with a wide range of visual RL algorithms, both off-policy and on-policy, and significantly improves the robustness of the agent in unseen environments.

Giwoong Lee, Jiseung Ahn, Jeongyeol Choe

We propose HYBOOD, a hybrid out-of-distribution model based on normalizing flow followed by a simple linear classification model. In real-world settings, it is known that data corruption has a strong influence on model degradation; for example image quality like noise, blur and image geometry like translation, scaling, rotation. MNIST-C, CIFAR10-C are the general synthesized datasets to measure model performance and corruption difficulty in terms of covariate and semantic shifts. HYBOOD shows that the separability between in-distribution, covariate shift, and semantic shift can be represented by distribution distance and log-scale density. We also find out the attributes of covariate shifts are ordered by corruption difficulty ranking (CDR) for the datasets. To the best of our knowledge, this is the first method to measure data corruption difficulty with generative models using Wasserstein Distance, Mutual Information and Minimal Description Length. In this paper, we pose interesting experimental results that the MNIST-C trained generative model is most deteriorated by fog, impulse noise and stripe corruption types. This can be interpreted that those attributes are challenging corruptions to the generative model in uncertainty and complexity. By training in-distribution data only, HYBOOD achieves out-of-distribution detection performance for distinguishable covariate and semantic shifts, and quantifying covariate shift ranking.

Geonho Lee, Janghwan Lee, Sukjin Hong, Minsoo Kim, Euijai Ahn, Du-Seong Chang, Jungwook Choi

Low-rank adaptation (LoRA) has become the dominant method for parameter-efficient LLM fine-tuning, with LoRA-based quantization error compensation (LQEC) emerging as a powerful tool for recovering accuracy in compressed LLMs. However, LQEC has underperformed in sub-4-bit scenarios, with no prior investigation into understanding this limitation. We propose RILQ (Rank-Insensitive LoRA-based Quantization Error Compensation) to boost 2-bit LLM accuracy. Based on rank analysis revealing model-wise activation discrepancy loss's rank-insensitive nature, RILQ employs this loss to adjust adapters cooperatively across layers, enabling robust error compensation with low-rank adapters. Evaluations on LLaMA-2 and LLaMA-3 demonstrate RILQ's consistent improvements in 2-bit quantized inference across various state-of-the-art quantizers and enhanced accuracy in task-specific fine-tuning. RILQ maintains computational efficiency comparable to existing LoRA methods, enabling adapter-merged weight-quantized LLM inference with significantly enhanced accuracy, making it a promising approach for boosting 2-bit LLM performance.

Geon Lee, Wenchao Yu, Kijung Shin, Wei Cheng, Haifeng Chen

Time series data is essential in various applications, including climate modeling, healthcare monitoring, and financial analytics. Understanding the contextual information associated with real-world time series data is often essential for accurate and reliable event predictions. In this paper, we introduce TimeCAP, a time-series processing framework that creatively employs Large Language Models (LLMs) as contextualizers of time series data, extending their typical usage as predictors. TimeCAP incorporates two independent LLM agents: one generates a textual summary capturing the context of the time series, while the other uses this enriched summary to make more informed predictions. In addition, TimeCAP employs a multi-modal encoder that synergizes with the LLM agents, enhancing predictive performance through mutual augmentation of inputs with in-context examples. Experimental results on real-world datasets demonstrate that TimeCAP outperforms state-of-the-art methods for time series event prediction, including those utilizing LLMs as predictors, achieving an average improvement of 28.75% in F1 score.

Ganghun Lee, Minji Kim, Minsu Lee, Byoung-Tak Zhang

In continuous domains, reinforcement learning policies are often based on Gaussian distributions for their generality. However, the unbounded support of Gaussian policy can cause a bias toward sampling boundary actions in many continuous control tasks that impose action limits due to physical constraints. This "boundary action bias'' can negatively impact training in algorithms like Proximal Policy Optimization. Despite this, it has been overlooked in many existing research and applications. In this paper, we revisit this issue by presenting illustrative explanations and analysis from the sampling point of view. Then, we introduce a truncated Gaussian policy with inherent bounds as a minimal alternative to mitigate the bias. However, we find that the plain truncated Gaussian policy may lay the counter-bias, preferring interior actions: to balance the bias, we ultimately propose a scale-adjusted truncated Gaussian policy, where the distribution scale shrinks if the location is near the boundaries. This property makes boundary actions deterministic more than in plain truncated Gaussian, but still less than in original Gaussian. Extensive empirical studies and comparisons on various continuous control tasks demonstrate that the truncated Gaussian policies significantly reduce the rate of boundary action usage, while scale-adjusted ones successfully balance the bias and counter-bias. It generally outperforms the Gaussian policy and shows competitive results compared to other approaches designed to counteract the bias.

Bruce D. Lee, Leonardo F. Toso, Thomas T. Zhang, James Anderson, Nikolai Matni

Representation learning is a powerful tool that enables learning over large multitudes of agents or domains by enforcing that all agents operate on a shared set of learned features. However, many robotics or controls applications that would benefit from collaboration operate in settings with changing environments and goals, whereas most guarantees for representation learning are stated for static settings. Toward rigorously establishing the benefit of representation learning in dynamic settings, we analyze the regret of multi-task representation learning for linear-quadratic control. This setting introduces unique challenges. Firstly, we must account for and balance the misspecification introduced by an approximate representation. Secondly, we cannot rely on the parameter update schemes of single-task online LQR, for which least-squares often suffices, and must devise a novel scheme to ensure sufficient improvement. We demonstrate that for settings where exploration is "benign", the regret of any agent after T timesteps scales with the square root of T/H, where H is the number of agents. In settings with "difficult" exploration, the regret scales as the square root of the input dimension times the parameter dimension multiplied by T, plus a term which scales with T to the three quarters divided by H to the one fifth. In both cases, by comparing to the minimax single-task regret, we see a benefit of a large number of agents. Notably, in the difficult exploration case, by sharing a representation across tasks, the effective task-specific parameter count can often be small. Lastly, we validate the trends we predict.

Anton Lee, Heitor Murilo Gomes, Yaqian Zhang, W. Bastiaan Kleijn

Catastrophic forgetting is when a neural network loses previously learnt information after learning a new task sequentially. Avoiding catastrophic forgetting could reduce the resources necessary to update neural networks. Recently, Kolmogorov–Arnold Networks (KAN) gained the community's attention as preliminary experiments suggest KAN avoid catastrophic forgetting. KAN replace neural network edges with learnable B-splines and sum incoming edges in nodes. Proponents of KAN argue they avoid forgetting, are more accurate, are interpretable, and use fewer parameters. Our work investigates the claims that KAN avoid catastrophic forgetting, finding that they fail to do so on more complex datasets containing features that overlap between tasks. We give a simple explanation as to why and how KAN catastrophically forget. Motivated by evidence suggesting KAN are superior for symbolic regression, we augment KAN in the same ways as multilayer perceptron (MLP) to perform continual learning tasks, making special accommodations to support KAN. Our experiments found that unmodified KAN often forget more than MLP, but KAN can be better than MLP when combined with continual learning strategies. We aim to highlight some of the current shortcomings and strengths associated with KAN for continual learning.

Antoine Ledent, Peng Liu

Balancing predictive power and interpretability has long been a challenging research area, particularly in powerful yet complex models like neural networks, where nonlinearity obstructs direct interpretation. This paper introduces a novel approach to constructing an explainable neural network that harmonizes predictiveness and explainability. Our model is designed as a linear combination of a sparse set of jointly learned features, each derived from a different trainable function applied to a single 1-dimensional input feature. Leveraging the ability to learn arbitrarily complex relationships, our neural network architecture enables automatic selection of a sparse set of important features, with the final prediction being a sum of rescaled versions of these features. We demonstrate the ability to select significant features while maintaining comparable predictive performance and direct interpretability through extensive experiments on synthetic and real-world datasets. We also provide theoretical analysis on the generalization bounds of our framework, which is favorably linear in the number of selected features and only logarithmic in the number of input features. We further lift any dependence of sample complexity on the number of parameters or the architectural details under very mild conditions. Our research paves the way for further research on sparse and explainable neural networks with guarantees.

Jian Lang, Zhangtao Cheng, Ting Zhong, Fan Zhou

Multimodal learning with incomplete modality is practical and challenging. Recently, researchers have focused on enhancing the robustness of pre-trained MultiModal Transformers (MMTs) under missing modality conditions by applying learnable prompts. However, these prompt-based methods face several limitations: (1) incomplete modalities provide restricted modal cues for task-specific inference, (2) dummy imputation for missing content causes information loss and introduces noise, and (3) static prompts are instance-agnostic, offering limited knowledge for instances with various missing conditions. To address these issues, we propose RAGPT, a novel Retrieval-AuGmented dynamic Prompt Tuning framework. RAGPT comprises three modules: (I) the multi-channel retriever, which identifies similar instances through a within-modality retrieval strategy, (II) the missing modality generator, which recovers missing information using retrieved contexts, and (III) the context-aware prompter, which captures contextual knowledge from relevant instances and generates dynamic prompts to largely enhance the MMT’s robustness. Extensive experiments conducted on three real-world datasets show that RAGPT consistently outperforms all competitive baselines in handling incomplete modality problems.

Zhao-Rong Lai, Xiaotian Wu, Liangda Fang, Ziliang Chen, Cheng Li

The Weber location problem is widely used in several artificial intelligence scenarios. However, the gradient of the objective does not exist at a considerable set of singular points. Recently, a de-singularity subgradient method has been proposed to fix this problem, but it can only handle the q-th-powered l_2-norm case (1

Yuanhao Lai, Pengfei Zheng, Chenpeng Ji, Cheng Qiu, Tingkai Wang, Songhan Zhang, Zhengang Wang, Yunfei Du

Bayesian optimization (BO) is a key technique for solving black-box optimization problems. This study extends the scope of BO from conventional applications (e.g., AutoML and robotics learning) to automated tuning of software systems. Despite GP (Gaussian Process) implementing a foundation formalism for exploitation and exploration in BO, its limited predictive power and unrealistic assumptions (e.g., continuity and Gaussianity) can severely affect its effectiveness and efficiency in tuning complex software systems. To overcome these limitations, we propose a BO framework CoffeeBoost, which implements exploitation and exploration with a GBDT-native distribution-free probabilistic surrogate model. CoffeeBoost constructs surrogate models via stochastic gradient boosting ensembles (SGBE) and quantifies probabilistic distributions via distribution-free conformal predictive systems. Moreover, CoffeeBoost leverages the residual paths in SGBE to improve the local adaptiveness of the resulting predictive distributions in a GBDT-native manner. Across eight auto-tuning benchmarks for database management systems (DBMS), we evaluate CoffeeBoost and show its superior learnability and optimizability against existing GP-based and tree-ensemble-based BO schemes. Detailed analysis further shows CoffeeBoost's predictive distributions excel in both coverage and tightness.

Song Lai, Zhe Zhao, Fei Zhu, Xi Lin, Qingfu Zhang, Gaofeng Meng

Continual learning aims to learn multiple tasks sequentially. A key challenge in continual learning is balancing between two objectives: retaining knowledge from old tasks (stability) and adapting to new tasks (plasticity). Experience replay methods, which store and replay past data alongside new data, have become a widely adopted approach to mitigate catastrophic forgetting. However, these methods neglect the dynamic nature of the stability-plasticity trade-off and aim to find a fixed and unchanging balance, resulting in suboptimal adaptation during training and inference. In this paper, we propose Pareto Continual Learning (ParetoCL), a novel framework that reformulates the stability-plasticity trade-off in continual learning as a multi-objective optimization (MOO) problem. ParetoCL introduces a preference-conditioned model to efficiently learn a set of Pareto optimal solutions representing different trade-offs and enables dynamic adaptation during inference. From a generalization perspective, ParetoCL can be seen as an objective augmentation approach that learns from different objective combinations of stability and plasticity. Extensive experiments across multiple datasets and settings demonstrate that ParetoCL outperforms state-of-the-art methods and adapts to diverse continual learning scenarios.

Jiahao Lai, Jiaqi Li, Jian Xu, Yanru Wu, Boshi Tang, Siqi Chen, Yongfeng Huang, Wenbo Ding, Yang Li

Federated Learning (FL) offers a decentralized approach to model training, where data remains local and only model parameters are shared between the clients and the central server. Traditional methods, such as Federated Averaging (FedAvg), linearly aggregate these parameters which are usually trained on heterogeneous data distributions, potentially overlooking the complex, high-dimensional nature of the parameter space. This can result in degraded performance of the aggregated model. While personalized FL approaches can mitigate the heterogeneous data issue to some extent, the limitation of linear aggregation remains unresolved. To alleviate this issue, we investigate the generative approach of diffusion model and propose a novel generative parameter aggregation framework for personalized FL, pFedGPA. In this framework, we deploy a diffusion model on the server to integrate the diverse parameter distributions and propose a parameter inversion method to efficiently generate a set of personalized parameters for each client. This inversion method transforms the uploaded parameters into a latent code, which is then aggregated through denoising sampling to produce the final personalized parameters. By encoding the dependence of a client's model parameters on the specific data distribution using the high-capacity diffusion model, pFedGPA can effectively decouple the complexity of the overall distribution of all clients' model parameters from the complexity of each individual client's parameter distribution. Our experimental results consistently demonstrate the superior performance of the proposed method across multiple datasets, surpassing baseline approaches.