论文检索

Ruiqian Nai, Zixin Wen, Ji Li, Yuanzhi Li, Yang Gao

In representation learning, a disentangled representation is highly desirable as it encodes generative factors of data in a separable and compact pattern. Researchers have advocated leveraging disentangled representations to complete downstream tasks with encouraging empirical evidence. This paper further investigates the necessity of disentangled representation in downstream applications. Specifically, we show that dimension-wise disentangled representations are unnecessary on a fundamental downstream task, abstract visual reasoning. We provide extensive empirical evidence against the necessity of disentanglement, covering multiple datasets, representation learning methods, and downstream network architectures. Furthermore, our findings suggest that the informativeness of representations is a better indicator of downstream performance than disentanglement. Finally, the positive correlation between informativeness and disentanglement explains the claimed usefulness of disentangled representations in previous works. The source code is available at https://github.com/Richard-coder-Nai/disentanglement-lib-necessity.git

Lokesh Nagalapatti, Akshay Iyer, Abir De, Sunita Sarawagi

We address the Individualized continuous treatment effect (ICTE) estimation problem where we predict the effect of any continuous valued treatment on an individual using ob- servational data. The main challenge in this estimation task is the potential confounding of treatment assignment with in- dividual’s covariates in the training data, whereas during in- ference ICTE requires prediction on independently sampled treatments. In contrast to prior work that relied on regularizers or unstable GAN training, we advocate the direct approach of augmenting training individuals with independently sam- pled treatments and inferred counterfactual outcomes. We in- fer counterfactual outcomes using a two-pronged strategy: a Gradient Interpolation for close-to-observed treatments, and a Gaussian Process based Kernel Smoothing which allows us to down weigh high variance inferences. We evaluate our method on five benchmarks and show that our method out- performs six state-of-the-art methods on the counterfactual estimation error. We analyze the superior performance of our method by showing that (1) our inferred counterfactual re- sponses are more accurate, and (2) adding them to the train- ing data reduces the distributional distance between the con- founded training distribution and test distribution where treat- ment is independent of covariates. Our proposed method is model-agnostic and we show that it improves ICTE accuracy of several existing models.

Maximilian Muschalik, Fabian Fumagalli, Barbara Hammer, Eyke Hüllermeier

While shallow decision trees may be interpretable, larger ensemble models like gradient-boosted trees, which often set the state of the art in machine learning problems involving tabular data, still remain black box models. As a remedy, the Shapley value (SV) is a well-known concept in explainable artificial intelligence (XAI) research for quantifying additive feature attributions of predictions. The model-specific TreeSHAP methodology solves the exponential complexity for retrieving exact SVs from tree-based models. Expanding beyond individual feature attribution, Shapley interactions reveal the impact of intricate feature interactions of any order. In this work, we present TreeSHAP-IQ, an efficient method to compute any-order additive Shapley interactions for predictions of tree-based models. TreeSHAP-IQ is supported by a mathematical framework that exploits polynomial arithmetic to compute the interaction scores in a single recursive traversal of the tree, akin to Linear TreeSHAP. We apply TreeSHAP-IQ on state-of-the-art tree ensembles and explore interactions on well-established benchmark datasets.

Coenraad Mouton, Marthinus Wilhelmus Theunissen, Marelie H Davel

Understanding generalization in deep neural networks is an active area of research. A promising avenue of exploration has been that of margin measurements: the shortest distance to the decision boundary for a given sample or its representation internal to the network. While margins have been shown to be correlated with the generalization ability of a model when measured at its hidden representations (hidden margins), no such link between large margins and generalization has been established for input margins. We show that while input margins are not generally predictive of generalization, they can be if the search space is appropriately constrained. We develop such a measure based on input margins, which we refer to as 'constrained margins'. The predictive power of this new measure is demonstrated on the 'Predicting Generalization in Deep Learning' (PGDL) dataset and contrasted with hidden representation margins. We find that constrained margins achieve highly competitive scores and outperform other margin measurements in general. This provides a novel insight on the relationship between generalization and classification margins, and highlights the importance of considering the data manifold for investigations of generalization in DNNs.

Zeping Min, Jinfeng Bai, Chengfei Li

Semi-supervised learning algorithms that use pseudo-labeling have become increasingly popular for improving model performance by utilizing both labeled and unlabeled data. In this paper, we offer a fresh perspective on the selection of pseudo-labels, inspired by theoretical insights. We suggest that pseudo-labels with a high degree of local variance are more prone to inaccuracies. Based on this premise, we introduce the Local Variance Match (LVM) method, which aims to optimize the selection of pseudo-labels in semi-supervised learning (SSL) tasks. Our methodology is validated through a series of experiments on widely-used image classification datasets, such as CIFAR-10, CIFAR-100, and SVHN, spanning various labeled data quantity scenarios. The empirical findings show that the LVM method substantially outpaces current SSL techniques, achieving state-of-the-art results in many of these scenarios. For instance, we observed an error rate of 5.41% on CIFAR-10 with a single label for each class, 35.87% on CIFAR-100 when using four labels per class, and 1.94% on SVHN with four labels for each class. Notably, the standout error rate of 5.41% is less than 1% shy of the performance in a fully-supervised learning environment. In experiments on ImageNet with 100k labeled data, the LVM also reached state-of-the-art outcomes. Additionally, the efficacy of the LVM method is further validated by its stellar performance in speech recognition experiments.

Umberto Michieli, Mete Ozay

Continual Learning (CL) aims to learn a sequence of problems (i.e., tasks and domains) by transferring knowledge acquired on previous problems, whilst avoiding forgetting of past ones. Different from previous approaches which focused on CL for one NLP task or domain in a specific use-case, in this paper, we address a more general CL setting to learn from a sequence of problems in a unique framework. Our method, HOP, permits to hop across tasks and domains by addressing the CL problem along three directions: (i) we employ a set of adapters to generalize a large pre-trained model to unseen problems, (ii) we compute high-order moments over the distribution of embedded representations to distinguish independent and correlated statistics across different tasks and domains, (iii) we process this enriched information with auxiliary heads specialized for each end problem. Extensive experimental campaign on 4 NLP applications, 5 benchmarks and 2 CL setups demonstrates the effectiveness of our HOP.

Nicolas Michel, Giovanni Chierchia, Romain Negrel, Jean-François Bercher

We use the maximum a posteriori estimation principle for learning representations distributed on the unit sphere. We propose to use the angular Gaussian distribution, which corresponds to a Gaussian projected on the unit-sphere and derive the associated loss function. We also consider the von Mises-Fisher distribution, which is the conditional of a Gaussian in the unit-sphere. The learned representations are pushed toward fixed directions, which are the prior means of the Gaussians; allowing for a learning strategy that is resilient to data drift. This makes it suitable for online continual learning, which is the problem of training neural networks on a continuous data stream, where multiple classification tasks are presented sequentially so that data from past tasks are no longer accessible, and data from the current task can be seen only once. To address this challenging scenario, we propose a memory-based representation learning technique equipped with our new loss functions. Our approach does not require negative data or knowledge of task boundaries and performs well with smaller batch sizes while being computationally efficient. We demonstrate with extensive experiments that the proposed method outperforms the current state-of-the-art methods on both standard evaluation scenarios and realistic scenarios with blurry task boundaries. For reproducibility, we use the same training pipeline for every compared method and share the code at https://github.com/Nicolas1203/ocl-fd.

Xiangming Meng, Yoshiyuki Kabashima

In practical compressed sensing (CS), the obtained measurements typically necessitate quantization to a limited number of bits prior to transmission or storage. This nonlinear quantization process poses significant recovery challenges, particularly with extreme coarse quantization such as 1-bit. Recently, an efficient algorithm called QCS-SGM was proposed for quantized CS (QCS) which utilizes score-based generative models (SGM) as an implicit prior. Due to the adeptness of SGM in capturing the intricate structures of natural signals, QCS-SGM substantially outperforms previous QCS methods. However, QCS-SGM is constrained to (approximately) row-orthogonal sensing matrices as the computation of the likelihood score becomes intractable otherwise. To address this limitation, we introduce an advanced variant of QCS-SGM, termed QCS-SGM+, capable of handling general matrices effectively. The key idea is a Bayesian inference perspective on the likelihood score computation, wherein expectation propagation is employed for its approximate computation. Extensive experiments are conducted, demonstrating the substantial superiority of QCS-SGM+ over QCS-SGM for general sensing matrices beyond mere row-orthogonality.

Jeremy McMahan, Young Wu, Xiaojin Zhu, Qiaomin Xie

To ensure the usefulness of Reinforcement Learning (RL) in real systems, it is crucial to ensure they are robust to noise and adversarial attacks. In adversarial RL, an external attacker has the power to manipulate the victim agent's interaction with the environment. We study the full class of online manipulation attacks, which include (i) state attacks, (ii) observation attacks (which are a generalization of perceived-state attacks), (iii) action attacks, and (iv) reward attacks. We show the attacker's problem of designing a stealthy attack that maximizes its own expected reward, which often corresponds to minimizing the victim's value, is captured by a Markov Decision Process (MDP) that we call a meta-MDP since it is not the true environment but a higher level environment induced by the attacked interaction. We show that the attacker can derive optimal attacks by planning in polynomial time or learning with polynomial sample complexity using standard RL techniques. We argue that the optimal defense policy for the victim can be computed as the solution to a stochastic Stackelberg game, which can be further simplified into a partially-observable turn-based stochastic game (POTBSG). Neither the attacker nor the victim would benefit from deviating from their respective optimal policies, thus such solutions are truly robust. Although the defense problem is NP-hard, we show that optimal Markovian defenses can be computed (learned) in polynomial time (sample complexity) in many scenarios.

Sascha Marton, Stefan Lüdtke, Christian Bartelt, Heiner Stuckenschmidt

Decision Trees (DTs) are commonly used for many machine learning tasks due to their high degree of interpretability. However, learning a DT from data is a difficult optimization problem, as it is non-convex and non-differentiable. Therefore, common approaches learn DTs using a greedy growth algorithm that minimizes the impurity locally at each internal node. Unfortunately, this greedy procedure can lead to inaccurate trees. In this paper, we present a novel approach for learning hard, axis-aligned DTs with gradient descent. The proposed method uses backpropagation with a straight-through operator on a dense DT representation, to jointly optimize all tree parameters. Our approach outperforms existing methods on binary classification benchmarks and achieves competitive results for multi-class tasks. The implementation is available under: https://github.com/s-marton/GradTree

Davide Maran, Pierriccardo Olivieri, Francesco Emanuele Stradi, Giuseppe Urso, Nicola Gatti, Marcello Restelli

In this paper, we investigate the optimal online configuration of episodic Markov decision processes when the space of the possible configurations is continuous. Specifically, we study the interaction between a learner (referred to as the configurator) and an agent with a fixed, unknown policy, when the learner aims to minimize her losses by choosing transition functions in online fashion. The losses may be unrelated to the agent's rewards. This problem applies to many real-world scenarios where the learner seeks to manipulate the Markov decision process to her advantage. We study both deterministic and stochastic settings, where the losses are either fixed or sampled from an unknown probability distribution. We design two algorithms whose peculiarity is to rely on occupancy measures to explore with optimism the continuous space of transition functions, achieving constant regret in deterministic settings and sublinear regret in stochastic settings, respectively. Moreover, we prove that the regret bound is tight with respect to any constant factor in deterministic settings. Finally, we compare the empiric performance of our algorithms with a baseline in synthetic experiments.

Francesca Mandel, Ian Barnett

Neural networks are powerful predictive models, but they provide little insight into the nature of relationships between predictors and outcomes. Although numerous methods have been proposed to quantify the relative contributions of input features, statistical inference and hypothesis testing of feature associations remain largely unexplored. We propose a permutation-based approach to testing that uses the partial derivatives of the network output with respect to specific inputs to assess both the significance of input features and whether significant features are linearly associated with the network output. These tests, which can be flexibly applied to a variety of network architectures, enhance the explanatory power of neural networks, and combined with powerful predictive capability, extend the applicability of these models.

Mikołaj Małkiński, Jacek Mańdziuk

Abstract Visual Reasoning (AVR) comprises a wide selection of various problems similar to those used in human IQ tests. Recent years have brought dynamic progress in solving particular AVR tasks, however, in the contemporary literature AVR problems are largely dealt with in isolation, leading to highly specialized task-specific methods. With the aim of developing universal learning systems in the AVR domain, we propose the unified model for solving Single-Choice Abstract visual Reasoning tasks (SCAR), capable of solving various single-choice AVR tasks, without making any a priori assumptions about the task structure, in particular the number and location of panels. The proposed model relies on a novel Structure-Aware dynamic Layer (SAL), which adapts its weights to the structure of the considered AVR problem. Experiments conducted on Raven's Progressive Matrices, Visual Analogy Problems, and Odd One Out problems show that SCAR (SAL-based models, in general) effectively solves diverse AVR tasks, and its performance is on par with the state-of-the-art task-specific baselines. What is more, SCAR demonstrates effective knowledge reuse in multi-task and transfer learning settings. To our knowledge, this work is the first successful attempt to construct a general single-choice AVR solver relying on self-configurable architecture and unified solving method. With this work we aim to stimulate and foster progress on task-independent research paths in the AVR domain, with the long-term goal of development of a general AVR solver.

Jayesh Malaviya, Anirban Dasgupta, Rachit Chhaya

While coresets have been growing in terms of their application, barring few exceptions, they have mostly been limited to unsupervised settings. We consider supervised classification problems, and non-decomposable evaluation measures in such settings. We show that stratified uniform sampling based coresets have excellent empirical performance that are backed by theoretical guarantees too. We focus on the F1 score and Matthews Correlation Coefficient, two widely used non-decomposable objective functions that are nontrivial to optimize for and show that uniform coresets attain a lower bound for coreset size, and have good empirical performance, comparable with ``smarter'' coreset construction strategies.

Louis Mahon, Thomas Lukasiewicz

Online deep clustering refers to the joint use of a feature extraction network and a clustering model to assign cluster labels to each new data point or batch as it is processed. While faster and more versatile than offline methods, online clustering can easily reach the collapsed solution where the encoder maps all inputs to the same point and all are put into a single cluster. Successful existing models have employed various techniques to avoid this problem, most of which require data augmentation or which aim to make the average soft assignment across the dataset the same for each cluster. We propose a method that does not require data augmentation, and that, differently from existing methods, regularizes the hard assignments. Using a Bayesian framework, we derive an intuitive optimization objective that can be straightforwardly included in the training of the encoder network. Tested on four image datasets, it consistently avoids collapse more robustly than other methods and leads to more accurate clustering. We also conduct further experiments and analyses justifying our choice to regularize the hard cluster assignments. Code is available at https://github.com/Lou1sM/online_hard_clustering.

Yuting Ma, Yuanzhi Yao, Xiaohua Xu

Federated learning (FL) has attracted growing attention since it allows for privacy-preserving collaborative training on decentralized clients without explicitly uploading sensitive data to the central server. However, recent works have revealed that it still has the risk of exposing private data to adversaries. In this paper, we conduct reconstruction attacks and enhance inference attacks on various datasets to better understand that sharing trained classification model parameters to a central server is the main problem of privacy leakage in FL. To tackle this problem, a privacy-preserving image distribution sharing scheme with GAN (PPIDSG) is proposed, which consists of a block scrambling-based encryption algorithm, an image distribution sharing method, and local classification training. Specifically, our method can capture the distribution of a target image domain which is transformed by the block encryption algorithm, and upload generator parameters to avoid classifier sharing with negligible influence on model performance. Furthermore, we apply a feature extractor to motivate model utility and train it separately from the classifier. The extensive experimental results and security analyses demonstrate the superiority of our proposed scheme compared to other state-of-the-art defense methods. The code is available at https://github.com/ytingma/PPIDSG.

Yingfan Ma, Xiaoyuan Luo, Kexue Fu, Manning Wang

Pathological images play a vital role in clinical cancer diagnosis. Computer-aided diagnosis utilized on digital Whole Slide Images (WSIs) has been widely studied. The major challenge of using deep learning models for WSI analysis is the huge size of WSI images and existing methods struggle between end-to-end learning and proper modeling of contextual information. Most state-of-the-art methods utilize a two-stage strategy, in which they use a pre-trained model to extract features of small patches cut from a WSI and then input these features into a classification model. These methods can not perform end-to-end learning and consider contextual information at the same time. To solve this problem, we propose a framework that models a WSI as a pathologist's observing video and utilizes Transformer to process video clips with a divide-and-conquer strategy, which helps achieve both context-awareness and end-to-end learning. Extensive experiments on three public WSI datasets show that our proposed method outperforms existing SOTA methods in both WSI classification and positive region detection.

Xiang Ma, Xuemei Li, Lexin Fang, Tianlong Zhao, Caiming Zhang

Time series forecasting is a crucial task in various domains. Caused by factors such as trends, seasonality, or irregular fluctuations, time series often exhibits non-stationary. It obstructs stable feature propagation through deep layers, disrupts feature distributions, and complicates learning data distribution changes. As a result, many existing models struggle to capture the underlying patterns, leading to degraded forecasting performance. In this study, we tackle the challenge of non-stationarity in time series forecasting with our proposed framework called U-Mixer. By combining Unet and Mixer, U-Mixer effectively captures local temporal dependencies between different patches and channels separately to avoid the influence of distribution variations among channels, and merge low- and high-levels features to obtain comprehensive data representations. The key contribution is a novel stationarity correction method, explicitly restoring data distribution by constraining the difference in stationarity between the data before and after model processing to restore the non-stationarity information, while ensuring the temporal dependencies are preserved. Through extensive experiments on various real-world time series datasets, U-Mixer demonstrates its effectiveness and robustness, and achieves 14.5% and 7.7% improvements over state-of-the-art (SOTA) methods.

Lianbo Ma, Yuee Zhou, Jianlun Ma, Guo Yu, Qing Li

Weight quantization is an effective technique to compress deep neural networks for their deployment on edge devices with limited resources. Traditional loss-aware quantization methods commonly use the quantized gradient to replace the full-precision gradient. However, we discover that the gradient error will lead to an unexpected zig-zagging-like issue in the gradient descent learning procedures, where the gradient directions rapidly oscillate or zig-zag, and such issue seriously slows down the model convergence. Accordingly, this paper proposes a one-step forward and backtrack way for loss-aware quantization to get more accurate and stable gradient direction to defy this issue. During the gradient descent learning, a one-step forward search is designed to find the trial gradient of the next-step, which is adopted to adjust the gradient of current step towards the direction of fast convergence. After that, we backtrack the current step to update the full-precision and quantized weights through the current-step gradient and the trial gradient. A series of theoretical analysis and experiments on benchmark deep models have demonstrated the effectiveness and competitiveness of the proposed method, and our method especially outperforms others on the convergence performance.

Jianfei Ma

Temporal difference learning (TD) is a foundational concept in reinforcement learning (RL), aimed at efficiently assessing a policy's value function. TD(λ), a potent variant, incorporates a memory trace to distribute the prediction error into the historical context. However, this approach often neglects the significance of historical states and the relative importance of propagating the TD error, influenced by challenges such as visitation imbalance or outcome noise. To address this, we propose a novel TD algorithm named discerning TD learning (DTD), which allows flexible emphasis functions—predetermined or adapted during training—to allocate efforts effectively across states. We establish the convergence properties of our method within a specific class of emphasis functions and showcase its promising potential for adaptation to deep RL contexts. Empirical results underscore that employing a judicious emphasis function not only improves value estimation but also expedites learning across diverse scenarios.