The huge number of parameters deep neural network makes it difficult to deploy on embedded devices with limited hardware, computation, storage and energy resources. In this paper, we shall propose a log-sum minimization approach prune trained layer by thereby improving the compression ratio. Specifically, is achieved enhancing sparsity for such that output after pruning consistent original one. We further present an iteratively reweighted algorithm solve nonconvex nonsmooth problem general...

Enhanced-index-funds have attracted considerable attention from investors over the last decade, which aims at outperforming a benchmark index while maintaining similar risk level. In this article, we investigate an enhanced indexation methodology using Conditional Value-at-Risk (CVaR). particular, adopt CVaR of excess returns as measurement subject to cardinality constraint for controlling tracking portfolio scale precisely and tunable short-selling constraints adjusting margin each risky...

This paper considers the problem of minimizing sum a smooth function and Schatten-$p$ norm matrix. Our contribution involves proposing accelerated iteratively reweighted nuclear methods designed for solving nonconvex low-rank minimization problem. Two major novelties characterize our approach. Firstly, proposed method possesses rank identification property, enabling provable "correct" stationary point within finite number iterations. Secondly, we introduce an adaptive updating strategy...

We propose a simulation-based value iteration algorithm for approximately solving infinite horizon discounted MDPs with continuous state spaces and finite actions. At each time step, the employs shrinking ball method to estimate function at sampled states uses historical estimates in an interpolation-based fitting strategy build approximator of optimal function. Under moderate conditions, we prove that sequence approximators generated by converges uniformly probability one. Simple numerical...

Recently, the demand on performing intelligent tasks for mobile devices with low-latency is ever- increasing, while requirements of intensive computation and large storage size impedes deployment deep learning models directly devices. Fog radio access network (Fog-RAN) offers a promising solution by integrating computing power edge processing nodes (e.g., base stations). In this paper, we propose joint task selection downlink transmit beamforming approach to improve communication efficiency...

The huge size of deep neural networks makes it difficult to deploy on the embedded platforms with limited computation resources directly. In this article, we propose a novel trimming approach determine redundant parameters trained network in layer-wise manner produce compact network. This is achieved by minimizing nonconvex sparsity-inducing term while maintaining response close original one. We present proximal iteratively reweighted method resolve resulting model, which approximates...

This paper primarily focuses on computing the Euclidean projection of a vector onto $\ell_{p}$ ball in which $p\in(0,1)$. Such problem emerges as core building block statistical machine learning and signal processing tasks because its ability to promote sparsity desired solution. However, efficient numerical algorithms for finding projections are still not available, particularly large-scale optimization. To meet this challenge, we first derive first-order necessary optimality conditions...

We propose an online algorithm for solving a class of average-reward Markov decision processes with continuous state spaces in model-free setting. The combines the classical relative Q-learning asynchronous averaging procedure, which permits Q-value estimate at state-action pair to be updated based on observations other neighboring pairs sampled subsequent iterations. These point estimates are then retained and used constructing interpolation-based function approximator that predicts...

We investigate a class of nonconvex optimization problems characterized by feasible set consisting level-bounded regularizers, with continuously differentiable objective. propose novel hybrid approach to tackle such structured within first-order algorithmic framework combining the Frank-Wolfe method and gradient projection method. The step is amenable closed-form solution, while can be efficiently performed in reduced subspace. A notable characteristic our lies its independence from...

With the rapid upsurge of deep learning tasks at network edge, effective edge artificial intelligence (AI) inference becomes critical to provide low-latency intelligent services for mobile users via leveraging computing capability. In such scenarios, energy efficiency a primary concern. this paper, we present joint task selection and downlink beamforming strategy achieve energy-efficient AI through minimizing overall power consumption consisting both computation transmission consumption,...

A nonuniform quantization method based on A-law compression is proposed to mitigate noise of learned digital backpropagation. The Effective SNR improved about 0.41 dB over uniform under 8-bit quantization.