MAD Dispersion Measure Makes Extremal Queue Analysis Simple

A notorious problem in queueing theory is to compute the worst possible performance of the GI/G/1 queue under mean-dispersion constraints for the interarrival- and service-time distributions. We address this extremal queue problem by measuring dispersion in terms of mean absolute deviation (MAD) instead of the more conventional variance, making available methods for distribution-free analysis. Combined with random walk theory, we obtain explicit expressions for the extremal interarrival- and service-time distributions and, hence, the best possible upper bounds for all moments of the waiting time. We also obtain tight lower bounds that, together with the upper bounds, provide robust performance intervals. We show that all bounds are computationally tractable and remain sharp also when the mean and MAD are not known precisely but are estimated based on available data instead. Summary of Contribution: Queueing theory is a classic OR topic with a central role for the GI/G/1 queue. Although this queueing system is conceptually simple, it is notoriously hard to determine the worst-case expected waiting time when only knowing the first two moments of the interarrival- and service-time distributions. In this setting, the exact form of the extremal distribution can only be determined numerically as the solution to a nonconvex nonlinear optimization problem. Our paper demonstrates that using mean absolute deviation (MAD) instead of variance alleviates the computational intractability of the extremal GI/G/1 queue problem, enabling us to state the worst-case distributions explicitly.

Chapter 5. Answering Natural-Language Questions with Neuro-Symbolic Knowledge Bases

Symbolic reasoning systems based on first-order logics are computationally powerful, and feedforward neural networks are computationally efficient, so unless P=NP, neural networks cannot, in general, emulate symbolic logics. Hence bridging the gap between neural and symbolic methods requires achieving a delicate balance: one needs to incorporate just enough of symbolic reasoning to be useful for a task, but not so much as to cause computational intractability. In this chapter we first present results that make this claim precise, and then use these formal results to inform the choice of a neuro-symbolic knowledge-based reasoning system, based on a set-based dataflow query language. We then present experimental results with a number of variants of this neuro-symbolic reasoner, and also show that this neuro-symbolic reasoner can be closely integrated into modern neural language models.

THE ROLE OF MACHINE LEARNING FOR FLEXIBILITY AND REAL OPTIONS ANALYSIS IN ENGINEERING SYSTEMS DESIGN

Flexibility analysis helps improve the expected value of engineering systems under uncertainty (economic and/or social). Designing for flexibility, however, can be challenging as a large number of design variables, parameters, uncertainty drivers, decision making possibilities and metrics must be considered. Many available techniques either rely on assumptions that are not suitable for an engineering setting, or may be limited due to computational intractability. This paper makes the case for an increased integration of Machine Learning into flexibility and real options analysis in engineering systems design to complement existing design methods. Several synergies are found and discussed critically between the fields in order to explore better solutions that may exist by analyzing the data, which may not be intuitive to domain experts. Reinforcement Learning is particularly promising as a result of the theoretical common grounds with latest methodological developments e.g. decision-rule based real options analysis. Relevance to the field of computational creativity is examined, and potential avenues for further research are identified. The proposed concepts are illustrated through the design of an example infrastructure system.

Security and privacy issues related to the workplace-based security robot system

BACKGROUND: An isolated robot must take account of uncertainty in its world model and adapt its activities to take into account such as uncertainty. In the same way, a robot interaction with security and privacy issues (RISAPI) with people has to account for its confusion about the human internal state, as well as how this state will shift as humans respond to the robot. OBJECTIVES: This paper discusses RISAPI of our original work in the field, which shows how probabilistic planning and system theory algorithms in workplace robotic systems that work with people can allow for that reasoning using a security robot system. The problem is a general way as an incomplete knowledge 2- player game. RESULTS: In this general framework, the various hypotheses and these contribute to thrilling and complex robot behavior through real-time interaction, which transforms actual human subjects into a spectrum of production systems, robots, and care facilities. CONCLUSION: The models of the internal human situation, in which robots can be designed efficiently, are limited, and achieve optimal computational intractability in large, high-dimensional spaces. To achieve this, versatile, lightweight portrayals of the human inner state and modern algorithms offer great hope for reasoning.

Attribute Noise Robust Binary Classification (Student Abstract)

We consider the problem of learning linear classifiers when both features and labels are binary. In addition, the features are noisy, i.e., they could be flipped with an unknown probability. In Sy-De attribute noise model, where all features could be noisy together with same probability, we show that 0-1 loss (l0−1) need not be robust but a popular surrogate, squared loss (lsq) is. In Asy-In attribute noise model, we prove that l0−1 is robust for any distribution over 2 dimensional feature space. However, due to computational intractability of l0−1, we resort to lsq and observe that it need not be Asy-In noise robust. Our empirical results support Sy-De robustness of squared loss for low to moderate noise rates.

Candidate for practical post-quantum signature scheme

A new criterion of post-quantum security is used to design a practical signature scheme based on the computational complexity of the hidden discrete logarithm problem. A 4-dimensional finite non-commutative associative algebra is applied as algebraic support of the cryptoscheme. The criterion is formulated as computational intractability of the task of constructing a periodic function containing a period depending on the discrete logarithm value. To meet the criterion, the hidden commutative group possessing the 2-dimensional cyclicity is exploited in the developed signature scheme. The public-key elements are computed depending on two vectors that are generators of two different cyclic groups contained in the hidden group. When computing the public key two types of masking operations are used: i) possessing the property of mutual commutativity with the exponentiation operation and ii) being free of such property. The signature represents two integers and one vector S used as a multiplier in the verification equation. To prevent attacks using the value S as a fitting element the signature verification equation is doubled.

On resampling schemes for polytopes

The convex hull of a sample is used to approximate the support of the underlying distribution. This approximation has many practical implications in real life. To approximate the distribution of the functionals of convex hulls, asymptotic theory plays a crucial role. Unfortunately most of the asymptotic results are computationally intractable. To address this computational intractability, we consider consistent bootstrapping schemes for certain cases. Let $S_n=\{X_i\}_{i=1}^{n}$ be a sequence of independent and identically distributed random points uniformly distributed on an unknown convex set in $\mathbb{R}^{d}$ ($d\ge 2$ ). We suggest a bootstrapping scheme that relies on resampling uniformly from the convex hull of $S_n$ . Moreover, the resampling asymptotic consistency of certain functionals of convex hulls is derived under this bootstrapping scheme. In particular, we apply our bootstrapping technique to the Hausdorff distance between the actual convex set and its estimator. For $d=2$ , we investigate the asymptotic consistency of the suggested bootstrapping scheme for the area of the symmetric difference and the perimeter difference between the actual convex set and its estimate. In all cases the consistency allows us to rely on the suggested resampling scheme to study the actual distributions, which are not computationally tractable.

A Novel Authentication Protocol with Strong Security for Roaming Service in Global Mobile Networks

In today’s society, Global Mobile Networks (GLOMONETs) have become an important network infrastructure that provides seamless roaming service for mobile users when they leave their home network. Authentication is an essential mechanism for secure communication among the mobile user, home network, and foreign network in GLOMONET. Recently, Madhusudhan and Shashidhara presented a lightweight authentication protocol for roaming application in GLOMONET. However, we found their protocol not only has design flaws, but is also vulnerable to many attacks. To address these weaknesses, this paper proposes a novel authentication protocol with strong security for GLOMONET based on previous work. The fuzzy verifier technique makes the protocol free from smart card breach attack, while achieving the feature of local password change. Moreover, the computational intractability of the Discrete Logarithm Problem (DLP) guarantees the security of the session key. The security of the protocol is verified by the ProVerif tool. Compared with other related protocols, our protocol achieves a higher level of security at the expense of small increases in computational cost and communication cost. Therefore, it is more suitable for securing the roaming application in GLOMONET.

Robust Multi-Agent Reinforcement Learning via Minimax Deep Deterministic Policy Gradient

Despite the recent advances of deep reinforcement learning (DRL), agents trained by DRL tend to be brittle and sensitive to the training environment, especially in the multi-agent scenarios. In the multi-agent setting, a DRL agent’s policy can easily get stuck in a poor local optima w.r.t. its training partners – the learned policy may be only locally optimal to other agents’ current policies. In this paper, we focus on the problem of training robust DRL agents with continuous actions in the multi-agent learning setting so that the trained agents can still generalize when its opponents’ policies alter. To tackle this problem, we proposed a new algorithm, MiniMax Multi-agent Deep Deterministic Policy Gradient (M3DDPG) with the following contributions: (1) we introduce a minimax extension of the popular multi-agent deep deterministic policy gradient algorithm (MADDPG), for robust policy learning; (2) since the continuous action space leads to computational intractability in our minimax learning objective, we propose Multi-Agent Adversarial Learning (MAAL) to efficiently solve our proposed formulation. We empirically evaluate our M3DDPG algorithm in four mixed cooperative and competitive multi-agent environments and the agents trained by our method significantly outperforms existing baselines.