Search

In stock trading, a common phenomenon is that the trends of stocks in the same industry are very similar. In contrast, the movements of stocks in different industries are often different. Therefore, applying the same model to all stock trading is inappropriate without distinguishing the industries in which the stocks belong. However, recommending an optimal industry stock trading model is very challenging based on performance evaluation indicators. First, the indicators of the trading model are diverse. Second, the ranking of multiple indicators is often inconsistent. In the paper, we model the problem to be solved as a multi-criteria decision-making process. Therefore, we first divide stock dataset into nine industries according to their main business. Then, we apply several machine learning algorithms as candidate models to generate trading signals. Second, we conduct daily trading backtesting based on the trading signals to obtain multiple performance evaluation indicators. Third, we propose an optimal recommendation algorithm for the industry stock trading model with TODIM. The experimental results in the US stock market and China's A-share market show that the proposed algorithm can get a better trading model out-of-sample industry stock. Moreover, we effectively evaluate the generalization ability of the algorithm based on the proposed metrics. Finally, the proposed long–short portfolios based on the algorithm have achieved returns exceeding the benchmark on most out-of-sample datasets. [ABSTRACT FROM AUTHOR]

This paper is concerned with the ℓ 1 , ∞ -norm ball constrained multi-task learning problem, which has received extensive attention in many research areas such as machine learning, cognitive neuroscience, and signal processing. To address the challenges of solving large-scale multi-task Lasso problems, this paper develops an inexact semismooth Newton-based augmented Lagrangian (Ssnal) algorithm. When solving the inner problems in the Ssnal algorithm, the semismooth Newton (Ssn) algorithm with superlinear or even quadratic convergence is applied. Theoretically, this paper presents the global and asymptotically superlinear local convergence of the Ssnal algorithm under standard conditions. Computationally, we derive an efficient procedure to construct the generalized Jacobian of the projector onto ℓ 1 , ∞ -norm ball, which is an important component of the Ssnal algorithm, making the computational cost in the Ssn algorithm very cheap. Comprehensive numerical experiments on the multi-task Lasso problems demonstrate that the Ssnal algorithm is more efficient and robust than several existing state-of-the-art first-order algorithms. [ABSTRACT FROM AUTHOR]

To solve the problem of low accuracy in the traditional FunkSVD recommendation algorithm, an improved FunkSVD algorithm (RM-FS) is proposed. RM-FS is an improvement of the traditional FunkSVD algorithm, using RMSProp, a deep learning optimization algorithm. The RM-FS algorithm can not only solve the problem of reduced accuracy of the traditional FunkSVD algorithm because of iterative oscillations but also alleviate the impact of data sparseness on the accuracy of the algorithm, achieving the effect of improving the accuracy of the traditional algorithm. The experimental results show that the RM-FS algorithm proposed in this paper effectively improves the accuracy of the recommendation algorithm, which is better than the traditional FunkSVD recommendation algorithm and other improved FunkSVD algorithms. [ABSTRACT FROM AUTHOR]

In reinforcement learning (RL), the exploration/exploitation (E/E) dilemma is a very crucial issue, which can be described as searching between the exploration of the environment to find more profitable actions, and the exploitation of the best empirical actions for the current state. We focus on the single trajectory RL problem where an agent is interacting with a partially unknown MDP over single trajectories, and try to deal with the E/E in this setting. Given the reward function, we try to find a good E/E strategy to address the MDPs under some MDP distribution. This is achieved by selecting the best strategy in mean over a potential MDP distribution from a large set of candidate strategies, which is done by exploiting single trajectories drawn from plenty of MDPs. In this paper, we mainly make the following contributions: (1) We discuss the strategy-selector algorithm based on formula set and polynomial function. (2) We provide the theoretical and experimental regret analysis of the learned strategy under an given MDP distribution. (3) We compare these methods with the 'state-of-the-art' Bayesian RL method experimentally. [ABSTRACT FROM AUTHOR]

This paper studies the online hierarchical scheduling problem on two uniform machines with rejection. Two uniform machines M1, M2 run at the speeds of s ∈ (0, +∞), 1 separately; and they are provided with different capabilities. Each machine has a certain GOS level 1 or 2 and every job is also associated with a hierarchy 1 or 2. The job can only be assigned to the machine whose GOS level does not exceed the job's hierarchy. Preemption is permitted but idle is not introduced. Jobs come one by one over list. When a job arrives, it can be accepted and scheduled on some machine or rejected by paying its penalty. The objective is to minimize the sum of makespan yielded by accepted jobs and total penalties of all rejected jobs. For this problem, we propose a family of several online algorithms according to the range of s and the related lower bound is also obtained. These algorithms achieve optimal competitive ratio when s ∈ (0, 1) ∪ [1.618, +∞), but have a small gap between upper bound and lower bound in interval [1, 1.618). [ABSTRACT FROM AUTHOR]

Job learning and deterioration coexist in many realistic machine-job scheduling situations. However, in literature, the constructed forms of the machine scheduling models with job learning and/or deteriorating effects were specific types of functions, which constrained their applicability in practice. This paper introduces a new single-machine scheduling model, where the actual processing time of a job is a general function of its starting time as well as scheduled position, which shows a broad generalization in contrast to that of certain existing models. For three objectives corresponding to the single-machine scheduling problem-total weighted completion time, discounted total weighted completion time, and maximum lateness - this paper presents their respective approximation result on the basis of the worst-case bound analysis from the optimal algorithm. The results demonstrate that under our proposed model, minimization of scheduling operations such as the makespan, sum of the kth power of completion times, and total lateness are polynomially solvable. Moreover, under some feasible conditions for the scheduling parameters, the minimum optimization problems of the total weighted completion time, discounted total weighted completion time, maximum lateness, and total tardiness are all recognized as polynomial forms and their solutions are provided. [ABSTRACT FROM AUTHOR]

In the paper, single machine scheduling problems with a learning effect and a rate-modifying activity are considered. Under the learning effect, the processing time of a job is a decreasing function of its position in the sequence. The rate-modifying activity is an event that can change the speed of the machine, and hence the processing time of jobs after the activity. The following objective functions are considered: the makespan, the total earliness, tardiness and completion time penalty, and the total earliness, tardiness, due-window starting time and due-window size penalty. Polynomial time algorithms are proposed to optimally solve the problems. [ABSTRACT FROM AUTHOR]