In this paper, we propose to train deep neural networks with biomechanical simulations, to predict the prostate motion encountered during ultrasound-guided interventions. In this application, unstructured points are sampled from segmented pre-operative MR images to represent the anatomical regions of interest. The point sets are then assigned with point-specific material properties and displacement loads, forming the un-ordered input feature vectors. An adapted PointNet can be trained to predict the nodal displacements, using finite element (FE) simulations as ground-truth data. Furthermore, a versatile bootstrap aggregating mechanism is validated to accommodate the variable number of feature vectors due to different patient geometries, comprised of a training-time bootstrap sampling and a model averaging inference. This results in a fast and accurate approximation to the FE solutions without requiring subject-specific solid meshing. Based on 160,000 nonlinear FE simulations on clinical imaging data from 320 patients, we demonstrate that the trained networks generalise to unstructured point sets sampled directly from holdout patient segmentation, yielding a near real-time inference and an expected error of 0.017 mm in predicted nodal displacement.

Neural architecture search (NAS) has been extensively studied in the past few years. A popular approach is to represent each neural architecture in the search space as a directed acyclic graph (DAG), and then search over all DAGs by encoding the adjacency matrix and list of operations as a set of hyperparameters. Recent work has demonstrated that even small changes to the way each architecture is encoded can have a significant effect on the performance of NAS algorithms. In this work, we present the first formal study on the effect of architecture encodings for NAS, including a theoretical grounding and an empirical study. First we formally define architecture encodings and give a theoretical characterization on the scalability of the encodings we study Then we identify the main encoding-dependent subroutines which NAS algorithms employ, running experiments to show which encodings work best with each subroutine for many popular algorithms. The experiments act as an ablation study for prior work, disentangling the algorithmic and encoding-based contributions, as well as a guideline for future work. Our results demonstrate that NAS encodings are an important design decision which can have a significant impact on overall performance. Our code is available at https://github.com/naszilla/nas-encodings.

Deep generative models have been successfully applied to Zero-Shot Learning (ZSL) recently. However, the underlying drawbacks of GANs and VAEs (e.g., the hardness of training with ZSL-oriented regularizers and the limited generation quality) hinder the existing generative ZSL models from fully bypassing the seen-unseen bias. To tackle the above limitations, for the first time, this work incorporates a new family of generative models (i.e., flow-based models) into ZSL. The proposed Invertible Zero-shot Flow (IZF) learns factorized data embeddings (i.e., the semantic factors and the non-semantic ones) with the forward pass of an invertible flow network, while the reverse pass generates data samples. This procedure theoretically extends conventional generative flows to a factorized conditional scheme. To explicitly solve the bias problem, our model enlarges the seen-unseen distributional discrepancy based on a negative sample-based distance measurement. Notably, IZF works flexibly with either a naive Bayesian classifier or a held-out trainable one for zero-shot recognition. Experiments on widely-adopted ZSL benchmarks demonstrate the significant performance gain of IZF over existing methods, in both classic and generalized settings.

We present a federated learning approach for learning a client adaptable, robust model when data is non-identically and non-independently distributed (non-IID) across clients. By simulating heterogeneous clients, we show that adding learned client-specific conditioning improves model performance, and the approach is shown to work on balanced and imbalanced data set from both audio and image domains. The client adaptation is implemented by a conditional gated activation unit and is particularly beneficial when there are large differences between the data distribution for each client, a common scenario in federated learning.

In this review paper, we give a comprehensive overview of the large variety of approximation results for neural networks. Approximation rates for classical function spaces as well as benefits of deep neural networks over shallow ones for specifically structured function classes are discussed. While the mainbody of existing results is for general feedforward architectures, we also depict approximation results for convolutional, residual and recurrent neural networks.

An agent in a non-stationary contextual bandit problem should balance between exploration and the exploitation of (periodic or structured) patterns present in its previous experiences. Handcrafting an appropriate historical context is an attractive alternative to transform a non-stationary problem into a stationary problem that can be solved efficiently. However, even a carefully designed historical context may introduce spurious relationships or lack a convenient representation of crucial information. In order to address these issues, we propose an approach that learns to represent the relevant context for a decision based solely on the raw history of interactions between the agent and the environment. This approach relies on a combination of features extracted by recurrent neural networks with a contextual linear bandit algorithm based on posterior sampling. Our experiments on a diverse selection of contextual and non-contextual non-stationary problems show that our recurrent approach consistently outperforms its feedforward counterpart, which requires handcrafted historical contexts, while being more widely applicable than conventional non-stationary bandit algorithms.

Restricted Boltzmann machines (RBMs) with low-precision synapses are much appealing with high energy efficiency. However, training RBMs with binary synapses is challenging due to the discrete nature of synapses. Recently Huang proposed one efficient method to train RBMs with binary synapses by using a combination of gradient ascent and the message passing algorithm under the variational inference framework. However, additional heuristic clipping operation is needed. In this technical note, inspired from Huang's work , we propose one alternative optimization method using the Bayesian learning rule, which is one natural gradient variational inference method. As opposed to Huang's method, we update the natural parameters of the variational symmetric Bernoulli distribution rather than the expectation parameters. Since the natural parameters take values in the entire real domain, no additional clipping is needed. Interestingly, the algorithm in \cite{huang2019data} could be viewed as one first-order approximation of the proposed algorithm, which justifies its efficacy with heuristic clipping.

Machine teaching [42, 37] uses a meta/teacher model to guide the training of a student model (which will be used in real tasks) through training data selection, loss function design, etc. Previously, the teacher model only takes shallow/surface information as inputs (e.g., training iteration number, loss and accuracy from training/validation sets) while ignoring the internal states of the student model, which limits the potential of learning to teach. In this work, we propose an improved data teaching algorithm, where the teacher model deeply interacts with the student model by accessing its internal states. The teacher model is jointly trained with the student model using meta gradients propagated from a validation set. We conduct experiments on image classification with clean/noisy labels and empirically demonstrate that our algorithm makes significant improvement over previous data teaching methods.

While successful in many fields, deep neural networks (DNNs) still suffer from some open problems such as bad local minima and unsatisfactory generalization performance. In this work, we propose a novel architecture called Maximum-and-Concatenation Networks (MCN) to try eliminating bad local minima and improving generalization ability as well. Remarkably, we prove that MCN has a very nice property; that is, \emph{every local minimum of an $(l+1)$-layer MCN can be better than, at least as good as, the global minima of the network consisting of its first $l$ layers}. In other words, by increasing the network depth, MCN can autonomously improve its local minima's goodness, what is more, \emph{it is easy to plug MCN into an existing deep model to make it also have this property}. Finally, under mild conditions, we show that MCN can approximate certain continuous functions arbitrarily well with \emph{high efficiency}; that is, the covering number of MCN is much smaller than most existing DNNs such as deep ReLU. Based on this, we further provide a tight generalization bound to guarantee the inference ability of MCN when dealing with testing samples.

Graph Convolutional Network (GCN) has experienced great success in graph analysis tasks. It works by smoothing the node features across the graph. The current GCN models overwhelmingly assume that node feature information is complete. However, real-world graph data are often incomplete and containing missing features. Traditionally, people have to estimate and fill in the unknown features based on imputation techniques and then apply GCN. However, the process of feature filling and graph learning are separated, resulting in degraded and unstable performance. This problem becomes more serious when a large number of features are missing. We propose an approach that adapts GCN to graphs containing missing features. In contrast to traditional strategy, our approach integrates the processing of missing features and graph learning within the same neural network architecture. Our idea is to represent the missing data by Gaussian Mixture Model (GMM) and calculate the expected activation of neurons in the first hidden layer of GCN, while keeping the other layers of the network unchanged. This enables us to learn the GMM parameters and network weight parameters in an end-to-end manner. Notably, our approach does not increase the computational complexity of GCN and it is consistent with GCN when the features are complete. We conduct experiments on the node label classification task and demonstrate that our approach significantly outperforms the best imputation based methods by up to 99.43%, 102.96%, 6.97%, 35.36% in four benchmark graphs when a large portion of features are missing. The performance of our approach for the case with a low level of missing features is even superior to GCN for the case with complete features.