论文检索

Bin Wang, Chunyu Xie, Dawei Leng, Yuhui Yin

In the field of multimodal large language models (MLLMs), common methods typically involve unfreezing the language model during training to foster profound visual understanding. However, the fine-tuning of such models with vision-language data often leads to a diminution of their natural language processing (NLP) capabilities. To avoid this performance degradation, a straightforward solution is to freeze the language model while developing multimodal competencies. Unfortunately, previous works have not attained satisfactory outcomes. Building on the strategy of freezing the language model, we conduct thorough structural exploration and introduce the Inner-Adaptor Architecture (IAA). Specifically, the architecture incorporates multiple multimodal adaptors at varying depths within the large language model to facilitate direct interaction with the inherently text-oriented transformer layers, thereby enabling the frozen language model to acquire multimodal capabilities. Unlike previous approaches of freezing language models that require large-scale aligned data, our proposed architecture is able to achieve superior performance on small-scale datasets. We conduct extensive experiments to improve the general multimodal capabilities and visual grounding abilities of the MLLM. Our approach remarkably outperforms previous state-of-the-art methods across various vision-language benchmarks without sacrificing performance on NLP tasks. Code and models will be released.

Benjie Wang, Guy Van den Broeck

Probabilistic circuits are a unifying representation of functions as computation graphs of weighted sums and products. Their primary application is in probabilistic modeling, where circuits with non-negative weights (monotone circuits) can be used to represent and learn density/mass functions, with tractable marginal inference. Recently, it was proposed to instead represent densities as the square of the circuit function (squared circuits); this allows the use of negative weights while retaining tractability, and can be exponentially more expressive efficient than monotone circuits. Unfortunately, we show the reverse also holds, meaning that monotone circuits and squared circuits are incomparable in general. This raises the question of whether we can reconcile, and indeed improve upon the two modeling approaches. We answer in the positive by proposing Inception PCs, a novel type of circuit that naturally encompasses both monotone circuits and squared circuits as special cases, and employs complex parameters. Empirically, we validate that Inception PCs can outperform both monotone and squared circuits on a range of tabular and image datasets.

Benfeng Wang, Chao Huang, Jie Wen, Wei Wang, Yabo Liu, Yong Xu

Video anomaly detection (VAD) aims at locating the abnormal events in videos. Recently, the Weakly Supervised VAD has made great progress, which only requires video-level annotations when training. In practical applications, different institutions may have different types of abnormal videos. However, the abnormal videos cannot be circulated on the internet due to privacy protection. To train a more generalized anomaly detector that can identify various anomalies, it is reasonable to introduce federated learning into WSVAD. In this paper, we propose Global and Local Context-driven Federated Learning, a new paradigm for privacy protected weakly supervised video anomaly detection. Specifically, we utilize the vision-language association of CLIP to detect whether the video frame is abnormal. Instead of leveraging handcrafted text prompts for CLIP, we propose a text prompt generator. The generated prompt is simultaneously influenced by text and visual. On the one hand, the text provides global context related to anomaly, which improves the model's ability of generalization. On the other hand, the visual provides personalized local context because different clients may have videos with different types of anomalies or scenes. The generated prompt ensures global generalization while processing personalized data from different clients. Extensive experiments show that the proposed method achieves remarkable performance.

Baozhen Wang, Xingye Qiao

In an era where diverse and complex data are increasingly accessible, the precise prediction of individual treatment effects (ITE) becomes crucial across fields such as healthcare, economics, and social policy. Current state-of-the-art approaches, while providing valid prediction intervals through Conformal Quantile Regression (CQR) and related techniques, often yield overly conservative prediction intervals. In this work, we introduce a conformal inference approach to ITE using the conditional density of the outcome given the covariates. We leverage the reference distribution technique to efficiently estimate the conditional densities as the score functions under a two-stage conformal ITE framework. We show that our prediction intervals are not only marginally valid but are narrower than existing methods. Experimental results further validate the usefulness of our method.

Przemysław Andrzej Wałęga, Michael Rawson

Graph neural networks (GNNs) have recently been adapted to temporal settings, often employing temporal versions of the message-passing mechanism known from GNNs. We divide temporal message passing mechanisms from literature into two main types: global and local, and establish Weisfeiler-Leman characterisations for both. This allows us to formally analyse expressive power of temporal message-passing models. We show that global and local temporal message-passing mechanisms have incomparable expressive power when applied to arbitrary temporal graphs. However, the local mechanism is strictly more expressive than the global mechanism when applied to colour-persistent temporal graphs, whose node colours are initially the same in all time points. Our theoretical findings are supported by experimental evidence, underlining practical implications of our analysis.

Dariush Wahdany, Matthew Jagielski, Adam Dziedzic, Franziska Boenisch

Machine learning (ML) models have been shown to leak private information from their training datasets. Differential Privacy (DP), typically implemented through the differential private stochastic gradient descent algorithm (DP-SGD), has become the standard solution to bound leakage from the models. Despite recent improvements, DP-SGD-based approaches for private learning still usually struggle in the high privacy (ε≤1) and low data regimes, and when the private training datasets are imbalanced. To overcome these limitations, we propose Differentially Private Prototype Learning (DPPL) as a new paradigm for private transfer learning. DPPL leverages publicly pre-trained encoders to extract features from private data and generates DP prototypes that represent each private class in the embedding space and can be publicly released for inference. Since our DP prototypes can be obtained from only a few private training data points and without iterative noise addition, they offer high-utility predictions and strong privacy guarantees even under the notion of pure DP. We additionally show that privacy-utility trade-offs can be further improved when leveraging the public data beyond pre-training of the encoder: in particular, we can privately sample our DP prototypes from the publicly available data points used to train the encoder. Our experimental evaluation with four state-of-the-art encoders, four vision datasets, and under different data and imbalancedness regimes demonstrate DPPL's high performance under strong privacy guarantees in challenging private learning setups.

Mátyás Vincze, Laura Ferrarotti, Leonardo Lucio Custode, Bruno Lepri, Giovanni Iacca

Continuous control tasks often involve high-dimensional, dynamic, and non-linear environments. State-of-the-art performance in these tasks is achieved through complex closed-box policies that are effective, but suffer from an inherent opacity. Interpretable policies, while generally underperforming compared to their closed-box counterparts, advantageously facilitate transparent decision-making within automated systems. Hence, their usage is often essential for diagnosing and mitigating errors, supporting ethical and legal accountability, and fostering trust among stakeholders. In this paper, we propose SMoSE, a novel method to train sparsely activated interpretable controllers, based on a top-1 Mixture-of-Experts architecture. SMoSE combines a set of interpretable decision-makers, trained to be experts in different basic skills, and an interpretable router that assigns tasks among the experts. The training is carried out via state-of-the-art Reinforcement Learning algorithms, exploiting load-balancing techniques to ensure fair expert usage. We then distill decision trees from the weights of the router, significantly improving the ease of interpretation. We evaluate SMoSE on six benchmark environments from MuJoCo: our method outperforms recent interpretable baselines and narrows the gap with non-interpretable state-of-the-art algorithms.

Pascal R. Van der Vaart, Matthijs T. J. Spaan, Neil Yorke-Smith

Uncertainty quantification remains a difficult challenge in reinforcement learning. Several algorithms exist that successfully quantify uncertainty in a practical setting. However it is unclear whether these algorithms are theoretically sound and can be expected to converge. Furthermore, they seem to treat the uncertainty in the target parameters in different ways. In this work, we unify several practical algorithms into one theoretical framework by defining a new Bellman operator on distributions, and show that this Bellman operator is a contraction. We highlight use cases of our framework by analyzing an existing Bayesian Q-learning algorithm, and also introduce a novel uncertainty-aware variant of PPO that adaptively sets its clipping hyperparameter.

Bastien Van Delft, Tommaso Martorella, Alexandre Alahi

Conditioning image generation facilitates seamless editing and the creation of photorealistic images. However, conditioning on noisy or Out-of-Distribution (OoD) images poses significant challenges, particularly in balancing fidelity to the input and realism of the output. We introduce Confident Ordinary Differential Editing (CODE), a novel approach for image synthesis that effectively handles OoD guidance images. Utilizing a diffusion model as a generative prior, CODE enhances images through score-based updates along the probability-flow Ordinary Differential Equation (ODE) trajectory. This method requires no task-specific training, handcrafted modules, or assumptions, and is compatible with any diffusion model. Positioned at the intersection of conditional image generation and blind image restoration, CODE operates in a fully blind manner, relying solely on a pre-trained generative model. Our method introduces an alternative approach to blind restoration: instead of targeting a specific ground truth image based on assumptions about the underlying corruption, CODE aims to increase the likelihood of the input image while maintaining fidelity. This results in the most probable in-distribution image around the input. Our contributions are twofold. First, CODE introduces a novel editing method based on ODE providing enhanced control, realism, and fidelity compared to SDE-based counterpart. Second, we introduce a confidence interval-based clipping method, which improves CODE’s effectiveness by allowing it to disregard certain pixels or information, thus enhancing the restoration process in a blind manner. Experimental results demonstrate CODE’s effectiveness over existing methods, particularly in scenarios involving severe degradation or OoD inputs.

David Valencia, Henry Williams, Yuning Xing, Trevor Gee, Bruce A MacDonald, Minas Liarokapis

Categorical Distributional Reinforcement Learning (CDRL) has demonstrated superior sample efficiency in learning complex tasks compared to conventional Reinforcement Learning (RL) approaches. However, the practical application of CDRL is encumbered by challenging projection steps, detailed parameter tuning, and domain knowledge. This paper addresses these challenges by introducing a pioneering Continuous Distributional Model-Free RL algorithm tailored for continuous action spaces. The proposed algorithm simplifies the implementation of distributional RL, adopting an actor-critic architecture wherein the critic outputs a continuous probability distribution. Additionally, we propose an ensemble of multiple critics fused through a Kalman fusion mechanism to mitigate overestimation bias. Through a series of experiments, we validate that our proposed method provides a sample-efficient solution for executing complex continuous-control tasks.

Sharlin Utke, Jeremie Houssineau, Giovanni Montana

This paper explores the impact of relational state abstraction on sample efficiency and performance in collaborative Multi-Agent Reinforcement Learning. The proposed abstraction is based on spatial relationships in environments where direct communication between agents is not allowed, leveraging the ubiquity of spatial reasoning in real-world multi-agent scenarios. We introduce MARC (Multi-Agent Relational Critic), a simple yet effective critic architecture incorporating spatial relational inductive biases by transforming the state into a spatial graph and processing it through a relational graph neural network. The performance of MARC is evaluated across four collaborative tasks, including a novel environment with heterogeneous agents. We conduct a comprehensive empirical analysis, comparing MARC against state-of-the-art MARL baselines, demonstrating improvements in both sample efficiency and asymptotic performance, as well as its potential for generalization. Our findings suggest that a minimal integration of spatial relational inductive biases as abstraction can yield substantial benefits without requiring complex designs or task-specific engineering. This work provides insights into the potential of relational state abstraction to address sample efficiency, a key challenge in MARL, offering a promising direction for developing more efficient algorithms in spatially complex environments.

Alexander Tyurin

We focus on the classification problem with a separable dataset, one of the most important and classical problems from machine learning. The standard approach to this task is logistic regression with gradient descent (LR+GD). Recent studies have observed that LR+GD can find a solution with arbitrarily large step sizes, defying conventional optimization theory. Our work investigates this phenomenon and makes three interconnected key observations about LR+GD with large step sizes. First, we find a remarkably simple explanation of why LR+GD with large step sizes solves the classification problem: LR+GD reduces to a batch version of the celebrated perceptron algorithm when the step size tends to infinity. Second, we observe that larger step sizes lead LR+GD to higher logistic losses when it tends to the perceptron algorithm, but larger step sizes also lead to faster convergence to a solution for the classification problem, meaning that logistic loss is an unreliable metric of the proximity to a solution. Surprisingly, high loss values can actually indicate faster convergence. Third, since the convergence rate in terms of loss function values of LR+GD is unreliable, we examine the iteration complexity required by LR+GD with large step sizes to solve the classification problem and prove that this complexity is suboptimal. To address this, we propose a new method, Normalized LR+GD – based on the connection between LR+GD and the perceptron algorithm – with much better theoretical guarantees.

Songjun Tu, Jingbo Sun, Qichao Zhang, Yaocheng Zhang, Jia Liu, Ke Chen, Dongbin Zhao

Offline preference-based reinforcement learning (PbRL) typically operates in two phases: first, use human preferences to learn a reward model and annotate rewards for a reward-free offline dataset; second, learn a policy by optimizing the learned reward via offline RL. However, accurately modeling step-wise rewards from trajectory-level preference feedback presents inherent challenges. The reward bias introduced, particularly the overestimation of predicted rewards, leads to optimistic trajectory stitching, which undermines the pessimism mechanism critical to the offline RL phase. To address this challenge, we propose In-Dataset Trajectory Return Regularization (DTR) for offline PbRL, which leverages conditional sequence modeling to mitigate the risk of learning inaccurate trajectory stitching under reward bias. Specifically, DTR employs Decision Transformer and TD-Learning to strike a balance between maintaining fidelity to the behavior policy with high in-dataset trajectory returns and selecting optimal actions based on high reward labels. Additionally, we introduce an ensemble normalization technique that effectively integrates multiple reward models, balancing the trade-off between reward differentiation and accuracy. Empirical evaluations on various benchmarks demonstrate the superiority of DTR over other state-of-the-art baselines.

Viet-Hung Tran, Zichi Zhang, Tuan Dung Pham, Ngoc Phu Doan, Anh-Tuan Hoang, Peixin Li, Hans Vandierendonck, Ira Assent, Son T. Mai

Explanation for deep learning models on time series classification (TSC) tasks is an important and challenging problem. Most existing approaches use attribution maps to explain outcomes. However, they have limitations in generating explanations that are well-aligned with humans's perceptions. Recently LIME-based approaches provide a more meaningful explanation via segmenting the data. However, these approaches are still suffering from the processes of segment generations and evaluations. In this paper, we propose a novel time series explanation approach called InteDisUX to overcome these problems. Our technique utilizes the segment-level integrated gradient (SIG) for calculating importance scores for an initial set of small and equal segments before iteratively merge two consecutive ones to create better explanations under a unique greedy strategy guided by two new proposed metrics including discrimination and faithfulness gains. By this way, our method does not depend on predefined segments like others while being robusts to instability, poor local fidelity and data imbalance like LIME-based methods. Furthermore, InteDisUX is the first work to use the model's information to improve the set of segments} for time series explanation. Extensive experiments show that our method outperforms LIME-based ones in 12 datasets in terms of faithfulness and 8/12 datasets in terms of robustness.

Suxin Tong, Jingling Yuan

Long-term series forecasting aims to predict future data over long horizons based on historical information. However, existing methods struggle to effectively utilize long lookback windows due to overfitting, computational resource constraints, or information extraction challenges, thereby limiting them to using limited lookback windows for predicting long-term future series. To address these issues, this paper introduces the Input Refinement and Prediction Auxiliary (IRPA) framework, a lightweight model consisting of four linear layers designed to extract key information from ultra-long lookback windows to enhance limited lookback windows and assist prediction processes. IRPA comprises an Input Refinement Module (IRM) and a Prediction Auxiliary Module (PAM), each constructed from two linear layer sub-modules. The IRM performs effective decomposition and patching of ultra-long series, refining seasonal and trend features to increase the information density in limited lookback windows and mitigate overfitting and parameter inflation. The PAM extracts historical similarities and seasonal patterns from ultra-long lookback windows to significantly improve prediction accuracy. IRPA substantially extends the utilization of lookback windows, offering a lightweight and efficient solution with broad applicability. Experimental results on eight datasets show IRPA reduces the Mean Squared Error (MSE) by an average of 16.1% for various models.

Tomas Tokar, Scott Sanner

Finding high-quality representations of heterogeneous tabular datasets is crucial for their effective use in downstream machine learning tasks. Contrastive representation learning (CRL) methods have been previously shown to provide a straightforward way to learn such representations across various data domains. Current tabular CRL methods learn joint embeddings of data instances (tabular rows) by minimizing a contrastive loss between the original instance and its perturbations. Unlike existing tabular CRL methods, we propose leveraging frameworks established in multimodal representation learning, treating each tabular column as a distinct modality. A naive approach that applies a contrastive loss pairwise to tabular columns is not only prohibitively expensive as the number of columns increases, but as we demonstrate, it also fails to capture interactions between variables. Instead, we propose a novel method called ICE-T that learns each columnar embedding by contrasting it with aggregate embeddings of the complementary part of the table, thus capturing interactions and scaling linearly with the number of columns. Unlike existing tabular CRL methods, ICE-T allows for column-specific embeddings to be obtained independently of the rest of the table, enabling the inference of missing values and translation between columnar variables. We provide a comprehensive evaluation of ICE-T across diverse datasets, demonstrating that it generally surpasses the performance of the state-of-the-art alternatives.

Haoxiang Tian, Xingshuo Han, Guoquan Wu, An Guo, Yuan Zhou, Jie Zhang, Shuo Li, Jun Wei, Tianwei Zhang

Multi-objective evolutionary algorithms (MOEAs) are widely used for searching optimal solutions in complex multi-component applications. Traditional MOEAs for multi-component deep learning (MCDL) systems face challenges in enhancing the search efficiency while maintaining the diversity. To combat these, this paper proposes the first LLM-empowered adaptive evolutionary search algorithm to detect safety violations in MCDL systems. Inspired by the context-understanding ability of Large Language Models (LLMs), our approach promotes the LLM to comprehend the optimization problem and generate an initial population tailed to evolutionary objectives. Subsequently, it employs adaptive selection and variation to iteratively produce offspring, balancing the evolutionary efficiency and diversity. During the evolutionary process, to navigate away from the local optima, our approach integrates the evolutionary experience back into the LLM. This utilization harnesses the LLM's quantitative reasoning prowess to generate differential seeds, breaking away from current optimal solutions. We evaluate our approach in finding safety violations of MCDL systems, and compare its performance with state-of-the-art MOEA methods. Experimental results show that our approach can significantly improve the efficiency and diversity of the evolutionary search.

Changyuan Tian, Zhicong Lu, Zequn Zhang, Heming Yang, Wei Cao, Zhi Guo, Xian Sun, Li Jin

Long-term Multivariate Time Series (LMTS) forecasting aims to predict extended future trends based on channel-interrelated historical data. Considering the elusive channel correlations, most existing methods compromise by treating channels as independent or tentatively modeling pairwise channel interactions, making it challenging to handle the characteristics of both higher-order interactions and time variation in channel correlations. In this paper, we propose HyperMixer, a novel specializable hypergraph channel mixing plugin which introduces versatile hypergraph structures to capture group channel interactions and time-varying patterns for long-term multivariate time series forecasting. Specifically, to encode the higher-order channel interactions, we structure multiple channels into a hypergraph, achieving a two-phase message-passing mechanism: channel-to-group and group-to-channel. Moreover, the functionally specializable hypergraph structures are presented to boost the capability of hypergraph to capture the time-varying patterns across periods, further refining modeling of channel correlations. Extensive experimental results on seven available benchmark datasets demonstrate the effectiveness and generalization of our plugin in LMTS forecasting. The visual analysis further illustrates that HyperMixer with specializable hypergraphs tailors channel interactions specific to certain periods.

Aleksandar Terzic, Michael Hersche, Giacomo Camposampiero, Thomas Hofmann, Abu Sebastian, Abbas Rahimi

Selective state-space models (SSMs) are an emerging alternative to the Transformer, offering the unique advantage of parallel training and sequential inference. Although these models have shown promising performance on a variety of tasks, their formal expressiveness and length generalization properties remain underexplored. In this work, we provide insight into the workings of selective SSMs by analyzing their expressiveness and length generalization performance on regular language tasks, i.e., finite-state automaton (FSA) emulation. We address certain limitations of modern SSM-based architectures by introducing the Selective Dense State-Space Model (SD-SSM), the first selective SSM that exhibits perfect length generalization on a set of various regular language tasks using a single layer. It utilizes a dictionary of dense transition matrices, a softmax selection mechanism that creates a convex combination of dictionary matrices at each time step, and a readout consisting of layer normalization followed by a linear map. We then proceed to evaluate variants of diagonal selective SSMs by considering their empirical performance on commutative and non-commutative automata. We explain the experimental results with theoretical considerations.

Georgios Tertytchny, Georgios L. Stavrinides, Maria K. Michael

To address the challenges of imbalanced multi-class datasets typically used for rare event detection in critical cyber-physical systems, we propose an optimal, efficient, and adaptable mixed integer programming (MIP) ensemble weighting scheme. Our approach leverages the diverse capabilities of the classifier ensemble on a granular per class basis, while optimizing the weights of classifier-class pairs using elastic net regularization for improved robustness and generalization. Additionally, it seamlessly and optimally selects a predefined number of classifiers from a given set. We evaluate and compare our MIP-based method against six well-established weighting schemes, using representative datasets and suitable metrics, under various ensemble sizes. The experimental results reveal that MIP outperforms all existing approaches, achieving an improvement in balanced accuracy ranging from 0.99% to 7.31%, with an overall average of 4.53% across all datasets and ensemble sizes. Furthermore, it attains an overall average increase of 4.63%, 4.60%, and 4.61% in macro-averaged precision, recall, and F1-score, respectively, while maintaining computational efficiency.