当客户具有不同的数据分布时，最新的联合学习方法的性能比其集中式同行差得多。对于神经网络，即使集中式SGD可以轻松找到同时执行所有客户端的解决方案，当前联合优化方法也无法收敛到可比的解决方案。我们表明，这种性能差异很大程度上可以归因于非概念性提出的优化挑战。具体来说，我们发现网络的早期层确实学习了有用的功能，但是最后一层无法使用它们。也就是说，适用于此非凸问题的联合优化扭曲了最终层的学习。利用这一观察结果，我们提出了一个火车征征训练（TCT）程序来避开此问题：首先，使用现成方法（例如FedAvg）学习功能；然后，优化从网络的经验神经切线核近似获得的共透性问题。当客户具有不同的数据时，我们的技术可在FMNIST上的准确性提高高达36％，而CIFAR10的准确性提高了 +37％。

联合学习通常被认为是一种有益的技术，它允许多个代理人相互协作，提高模型的准确性，并解决这些问题，这些问题否则这些问题是数据密集型 /昂贵而无法单独解决的。但是，在预期其他代理商将共享其数据的情况下，理性的代理人可能会很想从事有害行为，例如自由骑行的行为，他们在哪里贡献了数据，但仍然享有改进的模型。在这项工作中，我们提出了一个框架来分析此类合理数据生成器的行为。我们首先展示了幼稚的方案如何导致灾难性的自由骑行水平，其中数据共享的好处被完全侵蚀。然后，使用合同理论的想法，我们介绍基于准确性的机制，以最大程度地提高每个代理生成的数据量。这些可以防止自由骑行而无需任何付款机制。

联合学习的个性化可以通过交易模型的偏差来提高用户模型的准确性（通过使用来自可能不同）的数据引入的数据来抵消其方差（由于任何单个用户的数据量有限）。为了开发最佳地平衡此权衡的培训算法，有必要扩展我们的理论基础。在这项工作中，我们将个性化协作学习问题正式，作为用户目标$ f_0（x）$的随机优化，同时获得对N $相关但其他用户的不同目标$ \ {f_1（x），\ dots，f_n （x）\} $。我们在此设置中为两个算法提供收敛保证 - 一种名称的个性化方法，称为\ emph {加权梯度平均}，以及一种新颖的\ emph {偏压校正}方法 - 以及我们可以最佳地折衷的条件偏差减少方差并实现线性加速WRT \用户数量$ N $。此外，我们还经验验证他们的表现，证实了我们的理论见解。

Federated Averaging (FEDAVG) has emerged as the algorithm of choice for federated learning due to its simplicity and low communication cost. However, in spite of recent research efforts, its performance is not fully understood. We obtain tight convergence rates for FEDAVG and prove that it suffers from 'client-drift' when the data is heterogeneous (non-iid), resulting in unstable and slow convergence.As a solution, we propose a new algorithm (SCAFFOLD) which uses control variates (variance reduction) to correct for the 'client-drift' in its local updates. We prove that SCAFFOLD requires significantly fewer communication rounds and is not affected by data heterogeneity or client sampling. Further, we show that (for quadratics) SCAFFOLD can take advantage of similarity in the client's data yielding even faster convergence. The latter is the first result to quantify the usefulness of local-steps in distributed optimization.

Sign-based algorithms (e.g. SIGNSGD) have been proposed as a biased gradient compression technique to alleviate the communication bottleneck in training large neural networks across multiple workers. We show simple convex counter-examples where signSGD does not converge to the optimum. Further, even when it does converge, signSGD may generalize poorly when compared with SGD. These issues arise because of the biased nature of the sign compression operator.We then show that using error-feedback, i.e. incorporating the error made by the compression operator into the next step, overcomes these issues. We prove that our algorithm (EF-SGD) with arbitrary compression operator achieves the same rate of convergence as SGD without any additional assumptions. Thus EF-SGD achieves gradient compression for free. Our experiments thoroughly substantiate the theory and show that error-feedback improves both convergence and generalization. Code can be found at https://github.com/epfml/error-feedback-SGD.

Recent advances in neural radiance fields have enabled the high-fidelity 3D reconstruction of complex scenes for novel view synthesis. However, it remains underexplored how the appearance of such representations can be efficiently edited while maintaining photorealism. In this work, we present PaletteNeRF, a novel method for photorealistic appearance editing of neural radiance fields (NeRF) based on 3D color decomposition. Our method decomposes the appearance of each 3D point into a linear combination of palette-based bases (i.e., 3D segmentations defined by a group of NeRF-type functions) that are shared across the scene. While our palette-based bases are view-independent, we also predict a view-dependent function to capture the color residual (e.g., specular shading). During training, we jointly optimize the basis functions and the color palettes, and we also introduce novel regularizers to encourage the spatial coherence of the decomposition. Our method allows users to efficiently edit the appearance of the 3D scene by modifying the color palettes. We also extend our framework with compressed semantic features for semantic-aware appearance editing. We demonstrate that our technique is superior to baseline methods both quantitatively and qualitatively for appearance editing of complex real-world scenes.

The rapid growth of machine translation (MT) systems has necessitated comprehensive studies to meta-evaluate evaluation metrics being used, which enables a better selection of metrics that best reflect MT quality. Unfortunately, most of the research focuses on high-resource languages, mainly English, the observations for which may not always apply to other languages. Indian languages, having over a billion speakers, are linguistically different from English, and to date, there has not been a systematic study of evaluating MT systems from English into Indian languages. In this paper, we fill this gap by creating an MQM dataset consisting of 7000 fine-grained annotations, spanning 5 Indian languages and 7 MT systems, and use it to establish correlations between annotator scores and scores obtained using existing automatic metrics. Our results show that pre-trained metrics, such as COMET, have the highest correlations with annotator scores. Additionally, we find that the metrics do not adequately capture fluency-based errors in Indian languages, and there is a need to develop metrics focused on Indian languages. We hope that our dataset and analysis will help promote further research in this area.

Flooding is one of the most disastrous natural hazards, responsible for substantial economic losses. A predictive model for flood-induced financial damages is useful for many applications such as climate change adaptation planning and insurance underwriting. This research assesses the predictive capability of regressors constructed on the National Flood Insurance Program (NFIP) dataset using neural networks (Conditional Generative Adversarial Networks), decision trees (Extreme Gradient Boosting), and kernel-based regressors (Gaussian Process). The assessment highlights the most informative predictors for regression. The distribution for claims amount inference is modeled with a Burr distribution permitting the introduction of a bias correction scheme and increasing the regressor's predictive capability. Aiming to study the interaction with physical variables, we incorporate Daymet rainfall estimation to NFIP as an additional predictor. A study on the coastal counties in the eight US South-West states resulted in an $R^2=0.807$. Further analysis of 11 counties with a significant number of claims in the NFIP dataset reveals that Extreme Gradient Boosting provides the best results, that bias correction significantly improves the similarity with the reference distribution, and that the rainfall predictor strengthens the regressor performance.

Recent advancements in sensing and communication facilitate obtaining high-frequency real-time data from various physical systems like power networks, climate systems, biological networks, etc. However, since the data are recorded by physical sensors, it is natural that the obtained data is corrupted by measurement noise. In this paper, we present a novel algorithm for online real-time learning of dynamical systems from noisy time-series data, which employs the Robust Koopman operator framework to mitigate the effect of measurement noise. The proposed algorithm has three main advantages: a) it allows for online real-time monitoring of a dynamical system; b) it obtains a linear representation of the underlying dynamical system, thus enabling the user to use linear systems theory for analysis and control of the system; c) it is computationally fast and less intensive than the popular Extended Dynamic Mode Decomposition (EDMD) algorithm. We illustrate the efficiency of the proposed algorithm by applying it to identify the Van der Pol oscillator, the IEEE 68 bus system, and a ring network of Van der Pol oscillators.

Understanding the ambient scene is imperative for several applications such as autonomous driving and navigation. While obtaining real-world image data with per-pixel labels is challenging, existing accurate synthetic image datasets primarily focus on indoor spaces with fixed lighting and scene participants, thereby severely limiting their application to outdoor scenarios. In this work we introduce OmniHorizon, a synthetic dataset with 24,335 omnidirectional views comprising of a broad range of indoor and outdoor spaces consisting of buildings, streets, and diverse vegetation. Our dataset also accounts for dynamic scene components including lighting, different times of a day settings, pedestrians, and vehicles. Furthermore, we also demonstrate a learned synthetic-to-real cross-domain inference method for in-the-wild 3D scene depth and normal estimation method using our dataset. To this end, we propose UBotNet, an architecture based on a UNet and a Bottleneck Transformer, to estimate scene-consistent normals. We show that UBotNet achieves significantly improved depth accuracy (4.6%) and normal estimation (5.75%) compared to several existing networks such as U-Net with skip-connections. Finally, we demonstrate in-the-wild depth and normal estimation on real-world images with UBotNet trained purely on our OmniHorizon dataset, showing the promise of proposed dataset and network for scene understanding.