Mobile and embedded platforms are increasingly required to efficiently execute computationally demanding DNNs across heterogeneous processing elements. At run-time, the available hardware resources can vary considerably due other concurrently running applications. The performance requirements of applications could also change under different scenarios. To achieve desired performance, dynamic have been proposed in which number channels/layers be scaled real time meet varying resource...

Recently, there has been growing interest in improving the efficiency and accuracy of Indoor Positioning System (IPS). The Received Signal Strength- (RSS-) based fingerprinting technique is essential for indoor localization. However, it challenging to estimate position on RSS’s measurement under complex environment. This paper evaluates three machine learning approaches Gaussian Process (GP) regression with different kernels get best positioning model. hyperparameter tuning used select...

Motion planning has been used in robotics research to make movement decisions under certain constraints. Deep Reinforcement Learning (DRL) approaches have applied the cases of motion with continuous state representations. However, current DRL suffer from reward sparsity and overestimation issues. It is also challenging train agents deal complex task specifications deep neural network approximations. This paper considers one fragments Linear Temporal Logic (LTL), Generalized Reactivity rank 1...

The Transformer architecture is widely used for machine translation tasks. However, its resource-intensive nature makes it challenging to implement on constrained embedded devices, particularly where available hardware resources can vary at run-time. We propose a dynamic model that scales the based any particular time. proposed approach, 'Dynamic-HAT', uses HAT SuperTransformer as backbone search SubTransformers with different accuracy-latency trade-offs design optimal are sampled from...

An improved dynamic Grid-based potential field method was proposed based on the consideration that goal, robot and obstacles in soccer compete are all dynamic. We combined advantages of grid method, set to represent environment, got function method. used form inspire search algorithm A* which is for adjacent nodes. The meets real-time planning requirements complex environment. And it has received very good results solving local minima problem traditional improving efficiency. It better...

The moving sofa problem, introduced by Leo Moser in 1966, seeks to determine the maximal area of a 2D shape that can navigate an L-shaped corridor unit width. Joseph Gerver’s 1992 solution, providing lower bound approximately 2.2195, is best known, though its global optimality remains unproven. This paper leverages neural networks’ approximation power and recent advances invexity optimization explore optimality. We propose two approaches supporting conjecture his unique maximum. first...

The Moving Sofa Problem, formally proposed by Leo Moser in 1966, seeks to determine the largest area of a two-dimensional shape that can navigate through an $L$-shaped corridor with unit width. current best lower bound is about 2.2195, achieved Joseph Gerver 1992, though its global optimality remains unproven. In this paper, we investigate problem leveraging universal approximation strength and computational efficiency neural networks. We report two approaches, both supporting Gerver's...

Reinforcement learning has been used to solve sequential decision-making problems in intelligent systems. However, current RL approaches suffer from slow convergence and reward sparsity, its mechanism is challenging deal with complex task specifications. As one of the software engineering practices, temporal logic can describe nonMarkovian specifications, synthesized strategy which could be as a priori knowledge train agents interact environment efficiently. This paper considers agent reacts...

Deep neural networks become more popular as its ability to solve very complex pattern recognition problems. However, deep often need massive computational and memory resources, which is main reason resulting them be difficult efficiently entirely running on embedded platforms. This work addresses this problem by saving the requirements of proposing a variance reduced (VR)-based optimization with regularization techniques compress models within fast training process. It shown theoretically...

In BOM product matching, there will be several solutions based on the matching mechanism owing to fuzziness and diversity of customers needs. To select optimal through process, a quantization method, which evaluates according intuitive estimate, is proposed. Customers needs are consisted by many parameters, each parameter has values. First, initialize satisfaction level every value in requirements assessment. Second, as parameter, get difference pairwise comparison. Third, revise got second...

Deep learning is becoming more widespread due to its power in solving complex classification problems. However, deep models often require large memory and energy consumption, which may prevent them from being deployed effectively on embedded platforms, limiting their application. This work addresses the problem of requirements by proposing a regularization approach compress footprint models. It shown that sparsity-inducing can be solved using an enhanced stochastic variance reduced gradient...

Nowadays, multi-relational classification has become a hotspot for research and application in the field of data mining. Compared to single table with simple structure, tables is more complicated. However, not all information good effects on classification. It may decrease accuracy algorithm when irrelevant relations are added. In this article, we optimized using usefulness backgrounds remove those which have little effect The results show that, method effective.

A number of optimization approaches have been proposed for optimizing nonconvex objectives (e.g. deep learning models), such as batch gradient descent, stochastic descent and variance reduced descent. Theory shows these methods can converge by using an unbiased estimator. However, in practice biased estimation allow more efficient convergence to the vicinity since approach is computationally expensive. To produce fast there are two trade-offs strategies which between stochastic/batch,...

In this paper, we proposed a new technique, {\em variance controlled stochastic gradient} (VCSG), to improve the performance of reduced gradient (SVRG) algorithm. To avoid over-reducing by SVRG, hyper-parameter $\lambda$ is introduced in VCSG that able control SVRG. Theory shows optimization method can converge using an unbiased estimator, but practice, biased estimation allow more efficient convergence vicinity since approach computationally expensive. also has effect balancing trade-off...

Mobile and embedded platforms are increasingly required to efficiently execute computationally demanding DNNs across heterogeneous processing elements. At runtime, the available hardware resources can vary considerably due other concurrently running applications. The performance requirements of applications could also change under different scenarios. To achieve desired performance, dynamic have been proposed in which number channels/layers be scaled real time meet varying resource...