Recent empirical works show that large deep neural networks are often highly redundant and one can find much smaller subnetworks without a significant drop of accuracy. However, most existing methods of network pruning are empirical and heuristic, leaving it open whether good subnetworks provably exist, how to find them efficiently, and if network pruning can be provably better than direct training using gradient descent. We answer these problems positively by proposing a simple greedy selection approach for finding good subnetworks, which starts from an empty network and greedily adds important neurons from the large network. This differs from the existing methods based on backward elimination, which remove redundant neurons from the large network. Theoretically, applying our greedy selection strategy on sufficiently large pre-trained networks guarantees to find small subnetworks with lower loss than networks directly trained with gradient descent. Practically, we improve prior arts of network pruning on learning compact neural architectures on ImageNet, including ResNet, MobilenetV2/V3, and ProxylessNet. Our theory and empirical results on MobileNet suggest that we should fine-tune the pruned subnetworks to leverage the information from the large model, instead of re-training from new random initialization as suggested in \citet{liu2018rethinking}.

In recent decades, attention has been directed at anemia classification for various medical purposes, such as thalassemia screening and predicting iron deficiency anemia (IDA). In this study, a new method has been successfully tested for discrimination between IDA and \b{eta}-thalassemia trait (\b{eta}-TT). The method is based on a Dynamic Harmony Search (DHS). Complete blood count (CBC), a fast and inexpensive laboratory test, is used as the input of the system. Other models, such as a genetic programming method called structured representation on genetic algorithm in non-linear function fitting (STROGANOFF), an artificial neural network (ANN), an adaptive neuro-fuzzy inference system (ANFIS), a support vector machine (SVM), k-nearest neighbor (KNN), and certain traditional methods, are compared with the proposed method.

Image interpolation is a special case of image super-resolution, where the low-resolution image is directly down-sampled from its high-resolution counterpart without blurring and noise. Therefore, assumptions adopted in super-resolution models are not valid for image interpolation. To address this problem, we propose a novel image interpolation model based on sparse representation. Two widely used priors including sparsity and nonlocal self-similarity are used as the regularization terms to enhance the stability of interpolation model. Meanwhile, we incorporate the nonlocal linear regression into this model since nonlocal similar patches could provide a better approximation to a given patch. Moreover, we propose a new approach to learn adaptive sub-dictionary online instead of clustering. For each patch, similar patches are grouped to learn adaptive sub-dictionary, generating a more sparse and accurate representation. Finally, the weighted encoding is introduced to suppress tailing of fitting residuals in data fidelity. Abundant experimental results demonstrate that our proposed method outperforms several state-of-the-art methods in terms of quantitative measures and visual quality.

The ability to learn new concepts with small amounts of data is a crucial aspect of intelligence that has proven challenging for deep learning methods. Meta-learning for few-shot learning offers a potential solution to this problem: by learning to learn across data from many previous tasks, few-shot learning algorithms can discover the structure among tasks to enable fast learning of new tasks. However, a critical challenge in few-shot learning is task ambiguity: even when a powerful prior can be meta-learned from a large number of prior tasks, a small dataset for a new task can simply be very ambiguous to acquire a single model for that task. The Bayesian meta-learning models can naturally resolve this problem by putting a sophisticated prior distribution and let the posterior well regularized through Bayesian decision theory. However, currently known Bayesian meta-learning procedures such as VERSA suffer from the so-called {\it information preference problem}, that is, the posterior distribution is degenerated to one point and is far from the exact one. To address this challenge, we design a novel meta-regularization objective using {\it cyclical annealing schedule} and {\it maximum mean discrepancy} (MMD) criterion. The cyclical annealing schedule is quite effective at avoiding such degenerate solutions. This procedure includes a difficult KL-divergence estimation, but we resolve the issue by employing MMD instead of KL-divergence. The experimental results show that our approach substantially outperforms standard meta-learning algorithms.

The emergence of specialized optimization hardware such as CMOS annealers and adiabatic quantum computers carries the promise of solving hard combinatorial optimization problems more efficiently in hardware. Recent work has focused on formulating different combinatorial optimization problems as Ising models, the core mathematical abstraction used by a large number of these hardware platforms, and evaluating the performance of these models when solved on specialized hardware. An interesting area of application is data mining, where combinatorial optimization problems underlie many core tasks. In this work, we focus on consensus clustering (clustering aggregation), an important combinatorial problem that has received much attention over the last two decades. We present two Ising models for consensus clustering and evaluate them using the Fujitsu Digital Annealer, a quantum-inspired CMOS annealer. Our empirical evaluation shows that our approach outperforms existing techniques and is a promising direction for future research.

Decades of global research and development initiatives such as Integrated Computational Materials Engineering (ICME) [2][3] and the Materials Genome Initiative (MGI) [4] have demonstrated the ability for both physics-based and data-driven computations to accelerate the discovery and deployment of new alloys. It is established that machine learning (ML) can model process-structure-property relationships of alloys [5][6]. Of equal or greater impact, ML can greatly reduce the number of physics-based experiments and calculations needed to discover and design new materials with optimal properties [7][8][9]. However, the robust prediction of a new alloy and its processing designed to meet a desired, yet not previously achieved performance remains an open challenge; one that is met in this work. In other sects of materials science and engineering where new materials have been successfully predicted, the formulation of effective data descriptors, or "feature engineering," has emerged as a critical data pre-processing step to enable better performances from ML. Most such studies have focused on using high-throughput physics-based calculations together with chemical element descriptors to assist ML prediction [7][9].

Estimating the gradients of stochastic nodes is one of the crucial research questions in the deep generative modeling community. This estimation problem becomes further complex when we regard the stochastic nodes to be discrete because pathwise derivative techniques can not be applied. Hence, the gradient estimation requires the score function methods or the continuous relaxation of the discrete random variables. This paper proposes a general version of the Gumbel-Softmax estimator with continuous relaxation, and this estimator is able to relax the discreteness of probability distributions, including broader types than the current practice. In detail, we utilize the truncation of discrete random variables and the Gumbel-Softmax trick with a linear transformation for the relaxation. The proposed approach enables the relaxed discrete random variable to be reparameterized and to backpropagate through a large scale stochastic neural network. Our experiments consist of synthetic data analyses, which show the efficacy of our methods, and topic model analyses, which demonstrates the value of the proposed estimation in practices.

We study the problem of out-of-sample risk estimation in the high dimensional regime where both the sample size $n$ and number of features $p$ are large, and $n/p$ can be less than one. Extensive empirical evidence confirms the accuracy of leave-one-out cross validation (LO) for out-of-sample risk estimation. Yet, a unifying theoretical evaluation of the accuracy of LO in high-dimensional problems has remained an open problem. This paper aims to fill this gap for penalized regression in the generalized linear family. With minor assumptions about the data generating process, and without any sparsity assumptions on the regression coefficients, our theoretical analysis obtains finite sample upper bounds on the expected squared error of LO in estimating the out-of-sample error. Our bounds show that the error goes to zero as $n,p \rightarrow \infty$, even when the dimension $p$ of the feature vectors is comparable with or greater than the sample size $n$. One technical advantage of the theory is that it can be used to clarify and connect some results from the recent literature on scalable approximate LO.

How to accurately and efficiently label data on a mobile device is critical for the success of training machine learning models on mobile devices. Auto-labeling data on mobile devices is challenging, because data is usually incrementally generated and there is possibility of having unknown labels. Furthermore, the rich hardware heterogeneity on mobile devices creates challenges on efficiently executing auto-labeling workloads. In this paper, we introduce Flame, an auto-labeling system that can label non-stationary data with unknown labels. Flame includes a runtime system that efficiently schedules and executes auto-labeling workloads on heterogeneous mobile processors. Evaluating Flame with eight datasets on a smartphone, we demonstrate that Flame enables auto-labeling with high labeling accuracy and high performance.

Activity recognition in wearable computing faces two key challenges: i) activity characteristics may be context-dependent and change under different contexts or situations; ii) unknown contexts and activities may occur from time to time, requiring flexibility and adaptability of the algorithm. We develop a context-aware mixture of deep models termed the {\alpha}-\b{eta} network coupled with uncertainty quantification (UQ) based upon maximum entropy to enhance human activity recognition performance. We improve accuracy and F score by 10% by identifying high-level contexts in a data-driven way to guide model development. In order to ensure training stability, we have used a clustering-based pre-training in both public and in-house datasets, demonstrating improved accuracy through unknown context discovery.