The recent application of deep learning (DL) to various tasks has seen the performance of classical techniques surpassed by their DL-based counterparts. As a result, DL has equally seen application in the removal of noise from images. In particular, the use of deep feed-forward convolutional neural networks (DnCNNs) has been investigated for denoising. It utilizes advances in DL techniques such as deep architecture, residual learning, and batch normalization to achieve better denoising performance when compared with the other classical state-of-the-art denoising algorithms. However, its deep architecture resulted in a huge set of trainable parameters. Meta-optimization is a training approach of enabling algorithms to learn to train themselves by themselves. Training algorithms using meta-optimizers have been shown to enable algorithms to achieve better performance when compared to the classical gradient descent-based training approach. In this work, we investigate the application of the meta-optimization training approach to the DnCNN denoising algorithm to enhance its denoising capability. Our preliminary experiments on simpler algorithms reveal the prospects of utilizing the meta-optimization training approach towards the enhancement of the DnCNN denoising capability.

One of the most pressing problems in the automated analysis of historical documents is the availability of annotated training data. In this paper, we propose a novel method for the synthesis of training data for semantic segmentation of document images. We utilize clusters found in intermediate features of a StyleGAN generator for the synthesis of RGB and label images at the same time. Our model can be applied to any dataset of scanned documents without the need for manual annotation of individual images, as each model is custom-fit to the dataset. In our experiments, we show that models trained on our synthetic data can reach competitive performance on open benchmark datasets for line segmentation.

Artificial intelligence (AI) has significant potential to positively impact and advance medical imaging, including positron emission tomography (PET) imaging applications. AI has the ability to enhance and optimize all aspects of the PET imaging chain from patient scheduling, patient setup, protocoling, data acquisition, detector signal processing, reconstruction, image processing and interpretation. AI poses industry-specific challenges which will need to be addressed and overcome to maximize the future potentials of AI in PET. This paper provides an overview of these industry-specific challenges for the development, standardization, commercialization, and clinical adoption of AI, and explores the potential enhancements to PET imaging brought on by AI in the near future. In particular, the combination of on-demand image reconstruction, AI, and custom designed data processing workflows may open new possibilities for innovation which would positively impact the industry and ultimately patients.

Random exploration is one of the main mechanisms through which reinforcement learning (RL) finds well-performing policies. However, it can lead to undesirable or catastrophic outcomes when learning online in safety-critical environments. In fact, safe learning is one of the major obstacles towards real-world agents that can learn during deployment. One way of ensuring that agents respect hard limitations is to explicitly configure boundaries in which they can operate. While this might work in some cases, we do not always have clear a-priori information which states and actions can lead dangerously close to hazardous states. Here, we present an approach where an additional policy can override the main policy and offer a safer alternative action. In our instinct-regulated RL (IR^2L) approach, an "instinctual" network is trained to recognize undesirable situations, while guarding the learning policy against entering them. The instinct network is pre-trained on a single task where it is safe to make mistakes, and transferred to environments in which learning a new task safely is critical. We demonstrate IR^2L in the OpenAI Safety gym domain, in which it receives a significantly lower number of safety violations during training than a baseline RL approach while reaching similar task performance.

The M5 competition uncertainty track aims for probabilistic forecasting of sales of thousands of Walmart retail goods. We show that the M5 competition data faces strong overdispersion and sporadic demand, especially zero demand. We discuss resulting modeling issues concerning adequate probabilistic forecasting of such count data processes. Unfortunately, the majority of popular prediction methods used in the M5 competition (e.g. lightgbm and xgboost GBMs) fails to address the data characteristics due to the considered objective functions. The distributional forecasting provides a suitable modeling approach for to the overcome those problems. The GAMLSS framework allows flexible probabilistic forecasting using low dimensional distributions. We illustrate, how the GAMLSS approach can be applied for the M5 competition data by modeling the location and scale parameter of various distributions, e.g. the negative binomial distribution. Finally, we discuss software packages for distributional modeling and their drawback, like the R package gamlss with its package extensions, and (deep) distributional forecasting libraries such as TensorFlow Probability.

Existing analyses of optimization in deep learning are either continuous, focusing on (variants of) gradient flow, or discrete, directly treating (variants of) gradient descent. Gradient flow is amenable to theoretical analysis, but is stylized and disregards computational efficiency. The extent to which it represents gradient descent is an open question in deep learning theory. The current paper studies this question. Viewing gradient descent as an approximate numerical solution to the initial value problem of gradient flow, we find that the degree of approximation depends on the curvature along the latter's trajectory. We then show that over deep neural networks with homogeneous activations, gradient flow trajectories enjoy favorable curvature, suggesting they are well approximated by gradient descent. This finding allows us to translate an analysis of gradient flow over deep linear neural networks into a guarantee that gradient descent efficiently converges to global minimum almost surely under random initialization. Experiments suggest that over simple deep neural networks, gradient descent with conventional step size is indeed close to the continuous limit. We hypothesize that the theory of gradient flows will be central to unraveling mysteries behind deep learning.

Recurrent neural networks for language models like long short-term memory (LSTM) have been utilized as a tool for modeling and predicting long term dynamics of complex stochastic molecular systems. Recently successful examples on learning slow dynamics by LSTM are given with simulation data of low dimensional reaction coordinate. However, in this report we show that the following three key factors significantly affect the performance of language model learning, namely dimensionality of reaction coordinates, temporal resolution and state partition. When applying recurrent neural networks to molecular dynamics simulation trajectories of high dimensionality, we find that rare events corresponding to the slow dynamics might be obscured by other faster dynamics of the system, and cannot be efficiently learned. Under such conditions, we find that coarse graining the conformational space into metastable states and removing recrossing events when estimating transition probabilities between states could greatly help improve the accuracy of slow dynamics learning in molecular dynamics. Moreover, we also explore other models like Transformer, which do not show superior performance than LSTM in overcoming these issues. Therefore, to learn rare events of slow molecular dynamics by LSTM and Transformer, it is critical to choose proper temporal resolution (i.e., saving intervals of MD simulation trajectories) and state partition in high resolution data, since deep neural network models might not automatically disentangle slow dynamics from fast dynamics when both are present in data influencing each other.

Face anti-spoofing (FAS) plays an important role in protecting face recognition systems from face representation attacks. Many recent studies in FAS have approached this problem with domain generalization technique. Domain generalization aims to increase generalization performance to better detect various types of attacks and unseen attacks. However, previous studies in this area have defined each domain simply as an anti-spoofing datasets and focused on developing learning techniques. In this paper, we proposed a method that enables network to judge its domain by itself with the clustered convolutional feature statistics from intermediate layers of the network, without labeling domains as datasets. We obtained pseudo-domain labels by not only using the network extracting features, but also using depth estimators, which were previously used only as an auxiliary task in FAS. In our experiments, we trained with three datasets and evaluated the performance with the remaining one dataset to demonstrate the effectiveness of the proposed method by conducting a total of four sets of experiments.

Distributed training with synchronous stochastic gradient descent (SGD) on GPU clusters has been widely used to accelerate the training process of deep models. However, SGD only utilizes the first-order gradient in model parameter updates, which may take days or weeks. Recent studies have successfully exploited approximate second-order information to speed up the training process, in which the Kronecker-Factored Approximate Curvature (KFAC) emerges as one of the most efficient approximation algorithms for training deep models. Yet, when leveraging GPU clusters to train models with distributed KFAC (D-KFAC), it incurs extensive computation as well as introduces extra communications during each iteration. In this work, we propose D-KFAC (SPD-KFAC) with smart parallelism of computing and communication tasks to reduce the iteration time. Specifically, 1) we first characterize the performance bottlenecks of D-KFAC, 2) we design and implement a pipelining mechanism for Kronecker factors computation and communication with dynamic tensor fusion, and 3) we develop a load balancing placement for inverting multiple matrices on GPU clusters. We conduct real-world experiments on a 64-GPU cluster with 100Gb/s InfiniBand interconnect. Experimental results show that our proposed SPD-KFAC training scheme can achieve 10%-35% improvement over state-of-the-art algorithms.

Neural sequence models exhibit limited compositional generalization ability in semantic parsing tasks. Compositional generalization requires algebraic recombination, i.e., dynamically recombining structured expressions in a recursive manner. However, most previous studies mainly concentrate on recombining lexical units, which is an important but not sufficient part of algebraic recombination. In this paper, we propose LeAR, an end-to-end neural model to learn algebraic recombination for compositional generalization. The key insight is to model the semantic parsing task as a homomorphism between a latent syntactic algebra and a semantic algebra, thus encouraging algebraic recombination. Specifically, we learn two modules jointly: a Composer for producing latent syntax, and an Interpreter for assigning semantic operations. Experiments on two realistic and comprehensive compositional generalization benchmarks demonstrate the effectiveness of our model. The source code is publicly available at https://github.com/microsoft/ContextualSP.