scholarly journals Dynamic Decision-Making for Stabilized Deep Learning Software Platforms

2020 ◽  
Author(s):  
Soohyun Park ◽  
Dohyun Kim ◽  
Joongheon Kim
Keyword(s):  
Decision Making ◽  
Deep Learning ◽  
Super Resolution ◽  
Low Complexity ◽  
Dynamic Decision Making ◽  
Sequential Decision ◽  
Real World Data ◽  
Lyapunov Optimization ◽  
Learning Platforms ◽  
Learning Software

This chapter introduces a dynamic and low-complexity decision-making algorithm which aims at time-average utility maximization in real-time deep learning platforms, inspired by Lyapunov optimization. In deep learning computation, large delays can happen due to the fact that it is computationally expensive. Thus, handling the delays is an important issue for the commercialization of deep learning algorithms. In this chapter, the proposed algorithm observes system delays at first formulated by queue-backlog, and then it dynamically conducts sequential decision-making under the tradeoff between utility (i.e., deep learning performance) and system delays. In order to evaluate the proposed decision-making algorithm, the performance evaluation results with real-world data are presented under the applications of super-resolution frameworks. Lastly, this chapter summarizes that the Lyapunov optimization algorithm can be used in various emerging applications.

The Black Box Myth

2019 ◽  
Vol 1 (1) ◽  
pp. 1-3
Author(s):  
Abraham Rudnick
Keyword(s):  
Artificial Intelligence ◽  
Health Care ◽  
Decision Making ◽  
Deep Learning ◽  
Black Box ◽  
Dynamic Decision Making ◽  
Empirical Knowledge ◽  
Negative Consequences ◽  
Extreme Caution ◽  
Scientific Facts

Artificial intelligence (AI) and its correlates, such as machine and deep learning, are changing health care, where complex matters such as comoribidity call for dynamic decision-making. Yet, some people argue for extreme caution, referring to AI and its correlates as a black box. This brief article uses philosophy and science to address the black box argument about knowledge as a myth, concluding that this argument is misleading as it ignores a fundamental tenet of science, i.e., that no empirical knowledge is certain, and that scientific facts – as well as methods – often change. Instead, control of the technology of AI and its correlates has to be addressed to mitigate such unexpected negative consequences.

scholarly journals Preventing Disparate Treatment in Sequential Decision Making

2018 ◽  
Author(s):  
Hoda Heidari ◽  
Andreas Krause
Keyword(s):  
Decision Making ◽  
Learning Algorithm ◽  
Feature Space ◽  
Sequential Decision Making ◽  
Data Sets ◽  
Sequential Decision ◽  
Real World Data ◽  
Time Step ◽  
Job Application ◽  
Disparate Treatment

We study fairness in sequential decision making environments, where at each time step a learning algorithm receives data corresponding to a new individual (e.g. a new job application) and must make an irrevocable decision about him/her (e.g. whether to hire the applicant) based on observations made so far. In order to prevent cases of disparate treatment, our time-dependent notion of fairness requires algorithmic decisions to be consistent: if two individuals are similar in the feature space and arrive during the same time epoch, the algorithm must assign them to similar outcomes. We propose a general framework for post-processing predictions made by a black-box learning model, that guarantees the resulting sequence of outcomes is consistent. We show theoretically that imposing consistency will not significantly slow down learning. Our experiments on two real-world data sets illustrate and confirm this finding in practice.

scholarly journals Delay-Tolerant Sequential Decision Making for Task Offloading in Mobile Edge Computing Environments

Information ◽  
2019 ◽  
Vol 10 (10) ◽  
pp. 312 ◽  
Author(s):  
Ibrahim Alghamdi ◽  
Christos Anagnostopoulos ◽  
Dimitrios P. Pezaros
Keyword(s):  
Decision Making ◽  
Cloud Computing ◽  
Edge Computing ◽  
Cloud Services ◽  
Mobile Edge Computing ◽  
Sequential Decision ◽  
Mobile Nodes ◽  
Real World Data ◽  
Optimal Stopping Theory ◽  
Execution Delay

In recent years, there has been a significant increase in the use of mobile devices and their applications. Meanwhile, cloud computing has been considered as the latest generation of computing infrastructure. There has also been a transformation in cloud computing ideas and their implementation so as to meet the demand for the latest applications. mobile edge computing (MEC) is a computing paradigm that provides cloud services near to the users at the edge of the network. Given the movement of mobile nodes between different MEC servers, the main aim would be the connection to the best server and at the right time in terms of the load of the server in order to optimize the quality of service (QoS) of the mobile nodes. We tackle the offloading decision making problem by adopting the principles of optimal stopping theory (OST) to minimize the execution delay in a sequential decision manner. A performance evaluation is provided using real world data sets with baseline deterministic and stochastic offloading models. The results show that our approach significantly minimizes the execution delay for task execution and the results are closer to the optimal solution than other offloading methods.

scholarly journals UCBoost: A Boosting Approach to Tame Complexity and Optimality for Stochastic Bandits

2018 ◽  
Author(s):  
Fang Liu ◽  
Sinong Wang ◽  
Swapna Buccapatnam ◽  
Ness Shroff
Keyword(s):  
Decision Making ◽  
Computational Complexity ◽  
Open Problem ◽  
Computational Cost ◽  
Low Complexity ◽  
Numerical Results ◽  
Sequential Decision Making ◽  
Confidence Bound ◽  
Sequential Decision ◽  
Upper Confidence Bound

In this work, we address the open problem of finding low-complexity near-optimal multi-armed bandit algorithms for sequential decision making problems. Existing bandit algorithms are either sub-optimal and computationally simple (e.g., UCB1) or optimal and computationally complex (e.g., kl-UCB). We propose a boosting approach to Upper Confidence Bound based algorithms for stochastic bandits, that we call UCBoost. Specifically, we propose two types of UCBoost algorithms. We show that UCBoost(D) enjoys O(1) complexity for each arm per round as well as regret guarantee that is 1/e-close to that of the kl-UCB algorithm. We propose an approximation-based UCBoost algorithm, UCBoost(epsilon), that enjoys a regret guarantee epsilon-close to that of kl-UCB as well as O(log(1/epsilon)) complexity for each arm per round. Hence, our algorithms provide practitioners a practical way to trade optimality with computational complexity. Finally, we present numerical results which show that UCBoost(epsilon) can achieve the same regret performance as the standard kl-UCB while incurring only 1% of the computational cost of kl-UCB.

Sign in / Sign up

Export Citation Format

Share Document