Multiple instance learning (MIL) has been extensively applied to whole slide histopathology image (WSI) analysis. The existing aggregation strategy in MIL, which primarily relies on the first-order distance (e.g., mean difference) between instances, fails to accurately approximate the true feature distribution of each instance, leading to biased slide-level representations. Moreover, the scarcity of WSI observations easily leads to model overfitting, resulting in unstable testing performance and limited generalizability. To tackle these challenges, we propose a new Bayesian nonparametric framework for multiple instance learning, which adopts a cascade of Dirichlet processes (cDP) to incorporate the instance-to-bag characteristic of the WSIs. We perform feature aggregation based on the latent clusters formed by the Dirichlet process, which incorporates the covariances of the patch features and forms more representative clusters. We then perform bag-level prediction with another Dirichlet process model on the bags, which imposes a natural regularization on learning to prevent overfitting and enhance generalizability. Moreover, as a Bayesian nonparametric method, the cDP model can accurately generate posterior uncertainty, which allows for the detection of outlier samples and tumor localization. Extensive experiments on five WSI benchmarks validate the superior performance of our method, as well as its generalizability and ability to estimate uncertainties. Codes are anonymously available at https://anonymous.4open.science/r/cDPMIL-F130.
论文检索
输入标题、作者或关键词，从 701 篇学术成果中精准定位
Recovering the camera motion and scene geometry from visual data is a fundamental problem in the field of computer vision. Its success in standard vision is attributed to the maturity of feature extraction, data association and multi-view geometry. The recent emergence of neuromorphic event-based cameras places great demands on approaches that use raw event data as input to solve this fundamental problem. Existing state-of-the-art solutions typically infer implicitly data association by iteratively reversing the event data generation process. However, the nonlinear nature of these methods limits their applicability in real-time tasks, and the constant-motion assumption leads to unstable results under agile motion. To this end, we rethink the problem formulation in a way that aligns better with the differential working principle of event cameras. We show that the event-based normal flow can be used, via the proposed geometric error term, as an alternative to the full flow in solving a family of geometric problems that involve instantaneous first-order kinematics and scene geometry. Furthermore, we develop a fast linear solver and a continuous-time nonlinear solver on top of the proposed geometric error term. Experiments on both synthetic and real data demonstrate the superiority of our linear solver in terms of accuracy and efficiency, and indicate its complementary feature as an initialization method for existing nonlinear solvers. Besides, our continuous-time non-linear solver exhibits exceptional capability in accommodating sudden variations in motion since it does not rely on the constant-motion assumption.
This paper introduces HINER, a novel neural representation for compressing HSI and ensuring high-quality downstream tasks on compressed HSI. HINER fully exploits inter-spectral correlations by explicitly encoding of spectral wavelengths and achieves a compact representation of the input HSI sample through joint optimization with a learnable decoder. By additionally incorporating the Content Angle Mapper with the L1 loss, we can supervise the global and local information within each spectral band, thereby enhancing the overall reconstruction quality. For downstream classification on compressed HSI, we theoretically demonstrate the task accuracy is not only related to the classification loss but also to the reconstruction fidelity through a first-order expansion of the accuracy degradation, and accordingly adapt the reconstruction by introducing Adaptive Spectral Weighting. Owing to the monotonic mapping of HINER between wavelengths and spectral bands, we propose Implicit Spectral Interpolation for data augmentation by adding random variables to input wavelengths during classification model training. Experimental results on various HSI datasets demonstrate the superior compression performance of our HINER compared to the existing learned methods and also the traditional codecs. Our model is lightweight and computationally efficient, which maintains high accuracy for downstream classification task even on decoded HSIs at high compression ratios. Our materials will be released at https://github.com/Eric-qi/HINER.
Incomplete Multi-View Clustering (IMVC) is crucial for multi-media data analysis. While graph learning-based IMVC methods have shown promise, they still have limitations. The prevalent first-order affinity graph often misclassifies out-neighborhood intra-cluster and in-neighbor inter-cluster samples, worsened by data incompleteness. These inaccuracies, combined with high computational demands, restrict their suitability for large-scale IMVC tasks. To address these issues, we propose a novel Fast and Scalable IMVC with duality Optimal graph Filtering (FSIMVC-OF). Specifically, we refine the clustering-friendly structure of the bipartite graph by learning an optimal filter within a consensus clustering framework. Instead of learning a sample-side filter, we optimize an anchor-side graph filter and apply it to the anchor side, ensuring computational efficiency with linear complexity, supported by the provable equivalence between these two types of graph filters. We present an alternative optimization algorithm with linear complexity. Extensive experimental analysis demonstrates the superior performance of FSIMVC-OF over current IMVC methods. The codes of this article are released in https://github.com/sroytik/FSIMVC-OF.
Mainstream multimodal recommender systems are designed to learn user interest by analyzing user-item interaction graphs. However, what they learn about user interest needs to be completed because historical interactions only record items that best match user interest (i.e., the first-order interest), while suboptimal items are absent. To fully exploit user interest, we propose a Second-Order Interest Learning (SOIL) framework to retrieve second-order interest from unrecorded suboptimal items. In this framework, we build a user-item interaction graph augmented by second-order interest, an interest-aware item-item graph for the visual modality, and a similar graph for the textual modality. In our work, all three graphs are constructed from user-item interaction records and multimodal feature similarity. Similarly to other graph-based approaches, we apply graph convolutional networks to each of the three graphs to learn representations of users and items. To improve the exploitation of both first-order and second-order interest, we optimize the model by implementing contrastive learning modules for user and item representations at both the user-item and item-item levels. The proposed framework is evaluated on three real-world public datasets in online shopping scenarios. Experimental results verify that our method is able to significantly improve prediction performance. For instance, our method outperforms the previous state-of-the-art method MGCN by an average of 8.1% in terms of Recall@10. Code: https://github.com/TL-UESTC/SOIL.
Because high-quality data is like oxygen for AI systems, effectively eliciting information from crowdsourcing workers has become a first-order problem for developing high-performance machine learning algorithms. Two prevalent paradigms, spot-checking and peer prediction, enable the design of mechanisms to evaluate and incentivize high-quality data from human labelers. So far, at least three metrics have been proposed to compare the performances of these techniques \citepzhang2022high,gao2016incentivizing,burrell2021measurement. However, different metrics lead to divergent and even contradictory results in various contexts. In this paper, we harmonize these divergent stories, showing that two of these metrics are actually the same within certain contexts and explain the divergence of the third. Moreover, we unify these different contexts by introducingSpot Check Equivalence, which offers an interpretable metric for the effectiveness of a peer prediction mechanism. Finally, we present two approaches to compute spot check equivalence in various contexts, where simulation results verify the effectiveness of our proposed metric.
Dynamic graph learning equips the edges with time attributes and allows multiple links between two nodes, which is a crucial technology for understanding evolving data scenarios like traffic prediction and recommendation systems. Existing works obtain the evolving patterns mainly depending on the most recent neighbor sequences. However, we argue that whether two nodes will have interaction with each other in the future is highly correlated with the same interaction that happened in the past. Only considering the recent neighbors overlooks the phenomenon of repeat behavior and fails to accurately capture the temporal evolution of interactions. To fill this gap, this paper presents RepeatMixer, which considers evolving patterns of first and high-order repeat behavior in the neighbor sampling strategy and temporal information learning. Firstly, we define the first-order repeat-aware nodes of the source node as the destination nodes that have interacted historically and extend this concept to high orders as nodes in the destination node's high-order neighbors. Then, we extract neighbors of the source node that interacted before the appearance of repeat-aware nodes with a slide window strategy as its neighbor sequence. Next, we leverage both the first and high-order neighbor sequences of source and destination nodes to learn temporal patterns of interactions via an MLP-based encoder. Furthermore, considering the varying temporal patterns on different orders, we introduce a time-aware aggregation mechanism that adaptively aggregates the temporal representations from different orders based on the significance of their interaction time sequences. Experimental results demonstrate the superiority of RepeatMixer over state-of-the-art models in link prediction tasks, underscoring the effectiveness of the proposed repeat-aware neighbor sampling strategy.
The confluence of Federated Learning (FL) and Large Language Models (LLMs) is ushering in a new era in privacy-preserving natural language processing. However, the intensive memory requirements for fine-tuning LLMs pose significant challenges, especially when deploying on clients with limited computational resources. To circumvent this, we explore the novel integration of Memory-efficient Zeroth-Order Optimization within a federated setting, a synergy we term as FedMeZO. Our study is the first to examine the theoretical underpinnings of FedMeZO in the context of LLMs, tackling key questions regarding the influence of large parameter spaces on optimization behavior, the establishment of convergence properties, and the identification of critical parameters for convergence to inform personalized federated strategies. Our extensive empirical evidence supports the theory, showing that FedMeZO not only converges faster than traditional first-order methods such as FedAvg but also significantly reduces GPU memory usage during training to levels comparable to those during inference. Moreover, the proposed personalized FL strategy that is built upon the theoretical insights to customize the client-wise learning rate can effectively accelerate loss reduction. We hope our work can help to bridge theoretical and practical aspects of federated fine-tuning for LLMs, thereby stimulating further advancements and research in this area.
Multitask learning is a widely used paradigm for training models on diverse tasks, with applications ranging from graph neural networks to language model fine-tuning. Since tasks may interfere with each other, a key notion for modeling their relationships is task affinity. This includes pairwise task affinity, computed among pairs of tasks, and higher-order affinity, computed among subsets of tasks. Naively computing either of them requires repeatedly training on data pooled from various task combinations, which is computationally intensive. We present a new algorithm Grad-TAG that can estimate task affinities without this repeated training. The key idea of Grad-TAG is to train a "base" model for all tasks and then use a linearization technique to estimate the loss of any other model with a specific task combination. The linearization works by computing a gradient-based first-order approximation of the loss, using low-dimensional projections of gradients as features in a logistic regression trained to predict labels for the specific task combination. We show theoretically that the linearized model can provably approximate the loss when the gradient-based approximation is accurate, and also empirically verify that on several large models. Then, given the estimated task affinity matrix, we design a semi-definite program for clustering to group similar tasks that maximize the average density of clusters. We evaluate Grad-TAG's performance across seven datasets, including multi-label classification on graphs, and instruction fine-tuning of language models. Our results show that our task affinity estimates are within 2.7% distance of the true affinities while needing only 3% of FLOPs compared to full training. On our largest graph with 21M edges and 500 labeling tasks, our algorithm delivers an estimate accurate to within 5% of the true affinities, while using only 112.3 GPU hours. Our results show that Grad-TAG achieves excellent performance and runtime tradeoffs compared to existing approaches.
Learning first-order logic programs from relational facts yields intuitive insights into the data. Inductive logic programming (ILP) models are effective in learning first-order logic programs from observed relational data. Symbolic ILP models support rule learning in a data-ecient manner. However, symbolic ILP models are not robust to learn from noisy data. Neuro-symbolic ILP models utilize neural networks to learn logic programs in a differentiable manner which improves the robustness of ILP models. However, most neuro-symbolic methods need a strong language bias to learn logic programs, which reduces the usability and flexibility of ILP models and limits the logic program formats. In addition, most neuro-symbolic ILP methods cannot learn logic programs effectively from both small-size datasets and large-size datasets such as knowledge graphs. In the paper, we introduce a novel differentiable ILP model called differentiable first-order rule learner (DFORL), which is scalable to learn rules from both smaller and larger datasets. Besides, DFORL only needs the number of variables in the learned logic programs as input. Hence, DFORL is easy to use and does not need a strong language bias. We demonstrate that DFORL can perform well on several standard ILP datasets, knowledge graphs, and probabilistic relation facts and outperform several well-known differentiable ILP models. Experimental results indicate that DFORL is a precise, robust, scalable, and computationally cheap differentiable ILP model.
Lifted planning is usually defined as planning directly over a first-order representation. From the mid-1990s until the late 2010s, lifted planning was sidelined, as most of the state-of-the-art planners first ground the task and then solve it using a propositional representation. Moreover, it was unclear whether lifted planners could scale. But as planning problems become harder, they also become infeasible to ground. Recently, lifted planners came back into play, aiming at problems where grounding is a bottleneck. In this work, we survey recent advances in lifted planning. The main techniques rely either on state-space search or logic satisfiability. For lifted search-based planners, we show the direct connections to other areas of computer science, such as constraint satisfaction problems and databases. For lifted planners based on satisfiability, the advances in modeling are crucial to their scalability. We briefly describe the main planners available in the literature and their techniques.
Integrating Intent Understanding and Optimal Behavior Planning for Behavior Tree Generation from Human Instructions
PDF ↗Robots executing tasks following human instructions in domestic or industrial environments essentially require both adaptability and reliability. Behavior Tree (BT) emerges as an appropriate control architecture for these scenarios due to its modularity and reactivity. Existing BT generation methods, however, either do not involve interpreting natural language or cannot theoretically guarantee the BTs' success. This paper proposes a two-stage framework for BT generation, which first employs large language models (LLMs) to interpret goals from high-level instructions, then constructs an efficient goal-specific BT through the Optimal Behavior Tree Expansion Algorithm (OBTEA). We represent goals as well-formed formulas in first-order logic, effectively bridging intent understanding and optimal behavior planning. Experiments in the service robot validate the proficiency of LLMs in producing grammatically correct and accurately interpreted goals, demonstrate OBTEA's superiority over the baseline BT Expansion algorithm in various metrics, and finally confirm the practical deployability of our framework. The project website is https://dids-ei.github.io/Project/LLM-OBTEA.
Stochastic bilevel optimization (SBO) has been integrated into many machine learning paradigms recently including hyperparameter optimization, meta learning, reinforcement learning, etc. Along with the wide range of applications, there have been abundant studies on concerning the computing behaviors of SBO. However, the generalization guarantees of SBO methods are far less understood from the lens of statistical learning theory. In this paper, we provide a systematical generalization analysis of the first-order gradient-based bilevel optimization methods. Firstly, we establish the quantitative connections between the on-average argument stability and the generalization gap of SBO methods. Then, we derive the upper bounds of on-average argument stability for single timescale stochastic gradient descent (SGD) and two timescale SGD, where three settings (nonconvex-nonconvex (NC-NC), convex-convex (C-C) and strongly-convex-strongly-convex (SC-SC)) are considered respectively. Experimental analysis validates our theoretical findings. Compared with the previous algorithmic stability analysis, our results do not require the re-initialization of the inner-level parameters before each iteration and are suit for more general objective functions.
Provable Acceleration of Nesterov’s Accelerated Gradient Method over Heavy Ball Method in Training Over-Parameterized Neural Networks
PDF ↗Due to its simplicity and efficiency, the first-order gradient method has been extensively employed in training neural networks. Although the optimization problem of the neural network is non-convex, recent research has proved that the first-order method is capable of attaining a global minimum during training over-parameterized neural networks, where the number of parameters is significantly larger than that of training instances. Momentum methods, including the heavy ball (HB) method and Nesterov's accelerated gradient (NAG) method, are the workhorse of first-order gradient methods owning to their accelerated convergence. In practice, NAG often exhibits superior performance than HB. However, current theoretical works fail to distinguish their convergence difference in training neural networks. To fill this gap, we consider the training problem of the two-layer ReLU neural network under over-parameterization and random initialization. Leveraging high-resolution dynamical systems and neural tangent kernel (NTK) theory, our result not only establishes tighter upper bounds of the convergence rate for both HB and NAG, but also provides the first theoretical guarantee for the acceleration of NAG over HB in training neural networks. Finally, we validate our theoretical results on three benchmark datasets.
In Constrained Reinforcement Learning (CRL), agents explore the environment to learn the optimal policy while satisfying constraints. The penalty function method has recently been studied as an effective approach for handling constraints, which imposes constraints penalties on the objective to transform the constrained problem into an unconstrained one. However, it is challenging to choose appropriate penalties that balance policy performance and constraint satisfaction efficiently. In this paper, we propose a theoretically guaranteed penalty function method, Exterior Penalty Policy Optimization (EPO), with adaptive penalties generated by a Penalty Metric Network (PMN). PMN responds appropriately to varying degrees of constraint violations, enabling efficient constraint satisfaction and safe exploration. We theoretically prove that EPO consistently improves constraint satisfaction with a convergence guarantee. We propose a new surrogate function and provide worst-case constraint violation and approximation error. In practice, we propose an effective smooth penalty function, which can be easily implemented with a first-order optimizer. Extensive experiments are conducted, showing that EPO outperforms the baselines in terms of policy performance and constraint satisfaction with a stable training process, particularly on complex tasks.
One of the fundamental problems in reasoning about action is progression, which is to update a knowledge base according to the effects of an action into another knowledge base that retains all proper information. The problem is notoriously challenging, as in general, it requires second-order logic. Efforts have been made to find fragments where progression is first-order definable. Liu and Lakemeyer showed that for actions that have only local effects, progression is always first-order definable. They also generalized the result to so-called normal actions, that allow for non-local effects, as long as the affected fluent predicates only depend on local-effect ones, under certain restrictions on the knowledge base. In addition, they showed that for so-called proper+ knowledge bases, progression for normal actions can be efficient under reasonable assumptions. In this paper, we consider a larger class of theories, called the acyclic ones, that strictly subsumes normal actions. In such theories, dependencies between non-local effect fluent predicates are allowed, as long as they do not contain any cycles. We prove progression to be equally first-order definable for this class. Furthermore, under similar but stronger assumptions than those made by Liu and Lakemeyer, we show that progression is efficient as well.
In Community Question Answering (CQA) websites, most current expert finding methods often model expert embeddings from textual features and optimize them with expert-question first-order interactions, i.e., this expert has answered this question. In this paper, we try to address the limitation of current models that typically neglect the intrinsic high-order connectivity within expert-question interactions, which is pivotal for collaborative effects. We introduce an innovative and simple approach: by conceptualizing expert-question interactions as a bipartite graph, and then we propose a novel graph-based expert finding method based on contrastive learning to effectively capture both first-order and intricate high-order connectivity, named CGEF. Specifically, we employ a question encoder to model questions from titles and employ the graph attention network to recursively propagate embeddings. Besides, to alleviate the problem of sparse interactions, we devise two auxiliary tasks to enhance expert modeling. First, we generate multiple views of one expert, including: 1) behavior-level augmentation drops interaction edges randomly in the graph; 2) interest-level augmentation randomly replaces question titles with tags in the graph. Then we maximize the agreement between one expert and the corresponding augmented expert on a specific view. In this way, the model can effectively inject collaborative signals into expert modeling. Extensive experiments on six CQA datasets demonstrate significant improvements compared with recent methods.
Planning is the act of deliberative thinking before acting. It is based on a symbolic model of the world and the options to act in it, usually defined in function-free first-order logic. The task is to find a sequence of actions (a plan) that leads from a given current state to a desired goal state. The basic, purely physical description may be augmented with a partially ordered grammar-like structure (a Hierarchical Task Network or HTN), which can describe expert knowledge, or practical, legal, or operational requirements. In this talk, I will survey a variety of methods for automatically deriving plans using symbolic methods for planning -- from both my past and future research. These symbolic methods -- in some sense -- translate planning problems into other, simpler symbolic representations and reason over them to find plans. As a basis for these methods, I will firstly introduce relevant theoretical results on planning. First, I will discuss the expressive power of planning formalisms (ECAI'14, ICAPS'16) and second, the computational complexity of HTN planning and related tasks such as HTN plan verification, plan modification, and plan recognition (ICAPS'15, ICAPS'16). Based on these theoretical results, I will develop why SAT-based HTN planning is possible and how it can be implemented. To this end, I will survey several of my publications at top-tier conferences, including papers at ICAPS'17, AAAI'18, AAAI'19, IJCAI'19, AAAI'20, and ICAPS'21 -- in which I developed an highly SAT-based planner for HTN problems including the ability to find optimal plans as well as the grounding as a preprocessing step. Here I will also give an outlook on future developments and new ideas that I propose for SAT-based planning -- including the exploitation of structures in plan (e.g.\ landmarks or operator-counting constraints). Next, I will present the idea of expressing lifted classical planning as SAT (ICAPS'22). The resulting planner LiSAT was the first lifted SAT-based planner -- and proved highly efficient and outperformed all other lifted planners at the time of publication. Notably, LiSAT was the first planner (lifted or grounded) and still is the only one to solve the challenging OrganicSynthesis benchmark -- and could even prove optimality for all plans. I will also outline future ideas to further improve the efficiency of LiSAT. Lastly, I introduce the notion of planning with symbolic symbolic representations (AAAI'21 and ICAPS'23). Here one uses Binary Decision Diagrams to encode large sets of states efficiently. For expressing the additional structure encoded by HTNs, I show how BDDs can be suitably integrated into finite automata. Based on this representation, an efficient and optimal planning algorithm can be derived. Additionally, I show how this algorithm can be extended to also cover oversubscription planning.
Unlearnable example attacks are data poisoning attacks aiming to degrade the clean test accuracy of deep learning by adding imperceptible perturbations to the training samples, which can be formulated as a bi-level optimization problem. However, directly solving this optimization problem is intractable for deep neural networks. In this paper, we investigate unlearnable example attacks from a game-theoretic perspective, by formulating the attack as a nonzero sum Stackelberg game. First, the existence of game equilibria is proved under the normal setting and the adversarial training setting. It is shown that the game equilibrium gives the most powerful poison attack in that the victim has the lowest test accuracy among all networks within the same hypothesis space when certain loss functions are used. Second, we propose a novel attack method, called the Game Unlearnable Example (GUE), which has three main gradients. (1) The poisons are obtained by directly solving the equilibrium of the Stackelberg game with a first-order algorithm. (2) We employ an autoencoder-like generative network model as the poison attacker. (3) A novel payoff function is introduced to evaluate the performance of the poison. Comprehensive experiments demonstrate that GUE can effectively poison the model in various scenarios. Furthermore, the GUE still works by using a relatively small percentage of the training data to train the generator, and the poison generator can generalize to unseen data well. Our implementation code can be found at https://github.com/hong-xian/gue.
Jointly Improving the Sample and Communication Complexities in Decentralized Stochastic Minimax Optimization
PDF ↗We propose a novel single-loop decentralized algorithm, DGDA-VR, for solving the stochastic nonconvex strongly-concave minimax problems over a connected network of agents, which are equipped with stochastic first-order oracles to estimate their local gradients. DGDA-VR, incorporating variance reduction, achieves O(ε^−3) oracle complexity and O(ε^−2) communication complexity without resorting to multi-communication rounds – both are optimal, i.e., matching the lower bounds for this class of problems. Since DGDA-VR does not require multiple communication rounds, it is applicable to a broader range of decentralized computational environments. To the best of our knowledge, this is the first distributed method using a single communication round in each iteration to jointly optimize the oracle and communication complexities for the problem considered here.