High-resolution mapping of cells and tissue structures provides a foundation for developing interpretable machine-learning models for computational pathology. Deep learning algorithms can provide accurate mappings given large numbers of labeled instances for training and validation. Generating adequate volume of quality labels has emerged as a critical barrier in computational pathology given the time and effort required from pathologists. In this paper we describe an approach for engaging crowds of medical students and pathologists that was used to produce a dataset of over 220,000 annotations of cell nuclei in breast cancers. We show how suggested annotations generated by a weak algorithm can improve the accuracy of annotations generated by non-experts and can yield useful data for training segmentation algorithms without laborious manual tracing. We systematically examine interrater agreement and describe modifications to the MaskRCNN model to improve cell mapping. We also describe a technique we call Decision Tree Approximation of Learned Embeddings (DTALE) that leverages nucleus segmentations and morphologic features to improve the transparency of nucleus classification models. The annotation data produced in this study are freely available for algorithm development and benchmarking at: https://sites.google.com/view/nucls .

Federated Learning (FL) provides both model performance and data privacy for machine learning tasks where samples or features are distributed among different parties. In the training process of FL, no party has a global view of data distributions or model architectures of other parties. Thus the manually-designed architectures may not be optimal. In the past, Neural Architecture Search (NAS) has been applied to FL to address this critical issue. However, existing Federated NAS approaches require prohibitive communication and computation effort, as well as the availability of high-quality labels. In this work, we present Self-supervised Vertical Federated Neural Architecture Search (SS-VFNAS) for automating FL where participants hold feature-partitioned data, a common cross-silo scenario called Vertical Federated Learning (VFL). In the proposed framework, each party first conducts NAS using self-supervised approach to find a local optimal architecture with its own data. Then, parties collaboratively improve the local optimal architecture in a VFL framework with supervision. We demonstrate experimentally that our approach has superior performance, communication efficiency and privacy compared to Federated NAS and is capable of generating high-performance and highly-transferable heterogeneous architectures even with insufficient overlapping samples, providing automation for those parties without deep learning expertise.

Recent advances in machine learning have led to an acceleration of interest in research on artificial intelligence (AI). This fostered the exploration of possible applications of AI in various domains and also prompted critical discussions addressing the lack of interpretability, the limits of machine intelligence, potential risks and social challenges. In the exploration of the settings of the "human versus AI" relationship, perhaps the most elusive domain of interest is the creation and understanding of art. Many interesting initiatives are emerging at the intersection of AI and art, however comprehension and appreciation of art is still considered to be an exclusively human capability. Rooted in the idea that the existence and meaning of art is indeed inseparable from human-to-human interaction, the motivation behind this paper is to explore how bringing AI in the loop can foster not only advances in the fields of digital art and art history, but also inspire our perspectives on the future of art. The variety of activities and research initiatives related to "AI and Art" can generally be divided into two categories: 1) AI is used in the process of analyzing existing art; or 2) AI is used in the process of creating new art. In this paper, relevant aspects and contributions of these two categories are discussed, with a particular focus on the relation of AI to visual arts. In recent years, there has been a surge of interest among artists, technologists and researchers in exploring the creative potential of AI technologies. The use of AI in the process of creating visual art was significantly accelerated with the emergence of Generative Adversarial Networks (GAN) [56].

Online prediction for streaming time series data has practical use for many real-world applications where downstream decisions depend on accurate forecasts for the future. Deployment in dynamic environments requires models to adapt quickly to changing data distributions without overfitting. We propose POLA (Predicting Online by Learning rate Adaptation) to automatically regulate the learning rate of recurrent neural network models to adapt to changing time series patterns across time. POLA meta-learns the learning rate of the stochastic gradient descent (SGD) algorithm by assimilating the prequential or interleaved-test-then-train evaluation scheme for online prediction. We evaluate POLA on two real-world datasets across three commonly-used recurrent neural network models. POLA demonstrates overall comparable or better predictive performance over other online prediction methods.

While deep neural networks demonstrate state-of-the-art performance on a variety of learning tasks, their performance relies on the assumption that train and test distributions are the same, which may not hold in real-world applications. Domain generalization addresses this issue by employing multiple source domains to build robust models that can generalize to unseen target domains subject to shifts in data distribution. In this paper, we propose Domain-Free Domain Generalization (DFDG), a model-agnostic method to achieve better generalization performance on the unseen test domain without the need for source domain labels. DFDG uses novel strategies to learn domain-invariant class-discriminative features. It aligns class relationships of samples through class-conditional soft labels, and uses saliency maps, traditionally developed for post-hoc analysis of image classification networks, to remove superficial observations from training inputs. DFDG obtains competitive performance on both time series sensor and image classification public datasets.

Adversarial robustness is an open challenge in deep learning, most often tackled using adversarial training. Adversarial training is computationally costly, involving alternated optimization with a trade-off between standard generalization and adversarial robustness. We explore training robust models without adversarial training by revisiting a known result linking maximally robust classifiers and minimum norm solutions, and combining it with recent results on the implicit bias of optimizers. First, we show that, under certain conditions, it is possible to achieve both perfect standard accuracy and a certain degree of robustness without a trade-off, simply by training an overparameterized model using the implicit bias of the optimization. In that regime, there is a direct relationship between the type of the optimizer and the attack to which the model is robust. Second, we investigate the role of the architecture in designing robust models. In particular, we characterize the robustness of linear convolutional models, showing that they resist attacks subject to a constraint on the Fourier-$\ell_\infty$ norm. This result explains the property of $\ell_p$-bounded adversarial perturbations that tend to be concentrated in the Fourier domain. This leads us to a novel attack in the Fourier domain that is inspired by the well-known frequency-dependent sensitivity of human perception. We evaluate Fourier-$\ell_\infty$ robustness of recent CIFAR-10 models with robust training and visualize adversarial perturbations.

We present the group equivariant conditional neural process (EquivCNP), a metalearning method with permutation invariance in a data set as in conventional conditional neural processes (CNPs), and it also has transformation equivariance in data space. Incorporating group equivariance, such as rotation and scaling equivariance, provides a way to consider the symmetry of real-world data. We give a decomposition theorem for permutation-invariant and group-equivariant maps, which leads us to construct EquivCNPs with an infinite-dimensional latent space to handle group symmetries. In this paper, we build architecture using Lie group convolutional layers for practical implementation. We show that EquivCNP with translation equivariance achieves comparable performance to conventional CNPs in a 1D regression task. Moreover, we demonstrate that incorporating an appropriate Lie group equivariance, EquivCNP is capable of zero-shot generalization for an image-completion task by selecting an appropriate Lie group equivariance. Data symmetry has played a significant role in the deep neural networks. In particular, a convolutional neural network, which play an important part in the recent achievements of deep neural networks, has translation equivariance that preserves the symmetry of the translation group. From the same point of view, many studies have aimed to incorporate various group symmetries into neural networks, especially convolutional operation (Cohen et al., 2019; Defferrard et al., 2019; Finzi et al., 2020).

In the past decade, artificial neural networks have experienced an unprecedented renaissance. However, the current theory has yet to catch-up with the practice and cannot explain their impressive performance. Particularly intriguing is the fact that over-parameterized models do not tend to over-fit, even when trained to zero error on the training set. Owing to this seemingly paradoxical fact, researchers have focused on understanding the infinite-width limit of neural networks. This line of research has led to many important discoveries such as the'lazy-training' regime [9,32] which is governed by the limiting'neural tangent kernel' (see [19]), as well as the'mean-field' limit approach (see [22,25,26] for some examples) to study the training dynamics and loss landscape. The first to study the limiting distribution of a neural network at (a random) initialization was Neal [28], who proved a Central Limit Theorem (CLT) for two-layered wide neural networks. According to Neal's result, when initialized with random weights, as the width of the network goes to infinity, its law converges, in distribution, to a Gaussian process. Subsequent works have generalized this result to deeper networks and other architectures ([13, 17, 24, 30, 33, 35, 36]). This correspondence between Gaussian processes and neural networks has proved to be highly influential and has inspired many new models (see [35] for a thorough review of these models).

Self-training is a standard approach to semi-supervised learning where the learner's own predictions on unlabeled data are used as supervision during training. In this paper, we reinterpret this label assignment process as an optimal transportation problem between examples and classes, wherein the cost of assigning an example to a class is mediated by the current predictions of the classifier. This formulation facilitates a practical annealing strategy for label assignment and allows for the inclusion of prior knowledge on class proportions via flexible upper bound constraints. The solutions to these assignment problems can be efficiently approximated using Sinkhorn iteration, thus enabling their use in the inner loop of standard stochastic optimization algorithms. We demonstrate the effectiveness of our algorithm on the CIFAR-10, CIFAR-100, and SVHN datasets in comparison with FixMatch, a state-of-the-art self-training algorithm. Additionally, we elucidate connections between our proposed algorithm and existing confidence thresholded self-training approaches in the context of homotopy methods in optimization. Our code is available at https://github.com/stanford-futuredata/sinkhorn-label-allocation.

Self-attention, as the key block of transformers, is a powerful mechanism for extracting features from the inputs. In essence, what self-attention does to infer the pairwise relations between the elements of the inputs, and modify the inputs by propagating information between input pairs. As a result, it maps inputs to N outputs and casts a quadratic $O(N^2)$ memory and time complexity. We propose centroid attention, a generalization of self-attention that maps N inputs to M outputs $(M\leq N)$, such that the key information in the inputs are summarized in the smaller number of outputs (called centroids). We design centroid attention by amortizing the gradient descent update rule of a clustering objective function on the inputs, which reveals an underlying connection between attention and clustering. By compressing the inputs to the centroids, we extract the key information useful for prediction and also reduce the computation of the attention module and the subsequent layers. We apply our method to various applications, including abstractive text summarization, 3D vision, and image processing. Empirical results demonstrate the effectiveness of our method over the standard transformers.