The idea of combining the active query strategy and the passive-aggressive (PA) update strategy in online learning can be credited to the PA active (PAA) algorithm, which has proven to be effective in learning linear classifiers from datasets with a fixed feature space. We propose a novel family of online active learning algorithms, named PAA learning for trapezoidal data streams (PAATS) and multiclass PAATS (MPAATS) (and their variants), for binary and multiclass online classification tasks on trapezoidal data streams where the feature space may expand over time. Under the context of an ever-changing feature space, we provide the theoretical analysis of the mistake bounds for both PAATS and MPAATS. Our experiments on a wide variety of benchmark datasets have confirm that the combination of the instance-regulated active query strategy and the PA update strategy is much more effective in learning from trapezoidal data streams. We have also compared PAATS with online learning with streaming features (OLSF)-the state-of-the-art approach in learning linear classifiers from trapezoidal data streams. PAATS could achieve much better classification accuracy, especially for large-scale real-world data streams.

Non-intrusive sign language recognition has started to redefine the wellness of millions of hearing impaired people around the world. In this study we propose two models, named ContiNet and F-ContiNet. ContiNet is a 3D convolutional network with a constant number of temporal snapshots, and F-ContiNet is an evolution of ContiNet where the fusion idea is applied so that the spatial and temporal information from the early layers will have greater direct impact on the final recognition. We have conducted extensive experiments on a 500-class dataset and its 100-class subset, using both raw videos and mask videos produced by the skeletal information obtained from OpenPose. For all the models we have tested, those using mask videos performed significantly better and learned faster than those using raw videos, and F-ContiNet performed the best, with the best per-class top-1 accuracy of 96.6% over the 500-class dataset. More importantly, it confirms that keeping a constant temporal dimension in a deep network is a viable approach to SLR, especially when the temporal information from all the available snapshots can be fully exploited by a time-sensitive construct such as ConvLSTM.

This Innovate Practice Full Paper presents an orchestrated way to conducting agile sprint reviews in undergraduate capstone projects. An undergraduate capstone project typically involves a student development team as well as other stakeholders such as faculty advisors and industry mentors. Given that the agile development approach is adopted, a sprint review is conducted at the end of each sprint to demonstrate and critique the features that the development team have completed. Unlike the industrial settings, capstone design students can be challenged when conducting agile sprint reviews. On the one hand, the availability of industry mentors is unpredictable due to their own business commitments and schedules. On the other hand, it is extremely hard to bring all stakeholders together in one meeting. To overcome these difficulties, in our practice we have employed two strategies. First, we used a so-called agile buddy approach which allows each team to acquire an outsider student's opinion on their system. Second, we adopted a web-based collaboration platform such that students can collect critiques asynchronously from different stakeholders.

Many undergraduate engineering programs require senior students to take capstone design courses to work on industry-sponsored projects. Such a project often involves multiple parties of people playing various roles: students as project workers, faculty members as coordinator or advisors, as well as mentors from the sponsoring industry. In order to promote student success, highly demanded are platforms that can effectively facilitate multi-party collaboration during the development process of capstone projects. CapStone is such a platform that has been developed and deployed at the Behrend College, Pennsylvania State University. In this paper, we present the design rationale of CapStone, explaining the responsibilities of different roles involved in a capstone project and how those responsibilities are meshed into one common platform such that all parties could effectively orchestrate their activities. We have also conducted a study on how well CapStone could help student users in their project development. The responses showed that the adoption of CapStone has significantly improved student success in many ways. Particularly, over 80% of the students found CapStone helpful in project planning/scheduling, reporting weekly progress to advisors, collecting feedbacks on project works, and in writing their project reports.

Multi-agent systems research has been drawing innovative insights from other disciplines such as philosophy and cognitive science. While cognitive studies have shown that over 95% of human decisions conform to the recognition-primed decision making (

Diffusion geometry offers a general framework for multiscale analysis of massive data sets on manifold. However, its applicability is greatly limited due to the lack of work on distributed diffusion computing, as a data set expands over time, it can quickly exceed the processing capacity of a single agent. In this paper, we propose a multi-agent diffusion approach where a massive data set can be split into several subsets and each diffusion agent only needs to work with one subset in diffusion computation. We conduct an experiment by applying various splitting strategies to a large set of human decision-making experiences. The result indicates that the multi-agent diffusion approach is promising, and it is possible to benefit from using a large group of diffusion agents if their diffusion maps were constructed from subsets with shared data points (experiences). This study encourages the application of multi-agent diffusion approach to systems that rely on massive data analysis, and will stimulate further investigations on distributed diffusion computing.

Recent research on human-centered teamwork highly demands the design of cognitive agents that can model and exploit human partners' cognitive load to enhance team performance. In this paper, we focus on teams composed of human-agent pairs and develop a system called Shared Mental Models for all--SMMall. SMMall implements a hidden Markov model (HMM)-based cognitive load model for an agent to predict its human partner's instantaneous cognitive load status. It also implements a user interface (UI) concept called shared belief map, which offers a synergic representation of team members' information space and allows them to share beliefs. An experiment was conducted to evaluate the HMM-based load models. The results indicate that the HMM-based load models are effective in helping team members develop a shared mental model (SMM), and the benefit of load-based information sharing becomes more significant as communication capacity increases. It also suggests that multiparty communication plays an important role in forming/evolving team SMMs, and when a group of agents can be partitioned into subteams, splitting messages by their load status can be more effective for developing subteam SMMs.

In this article, an approach is proposed to evaluate the reliability of Web services, where three kinds of Web services are discussed, they are atomic service without the structural information, structural activity based composite service which is composed from other services by using a structural activity mechanism, and BPEL flow process based composite service which composes all kinds of atomic services and activity based composite services using BPEL language in an orchestration style. Firstly, the reliability of atomic service is evaluated based on an extended UDDI model, then, the reliability of activity based composite services is evaluated using BPEL code structure analysis, the reliability of atomic service, and function transition probability, finally, the reliability of BPEL flow process based composite web service was evaluated by a recursive algorithm. Case study and experimental results show the significance of the approach.

Naturalistic decision making (NDM) focuses on how people actually make decisions in realistic settings that typically involve ill-structured problems. Taking an experimental approach, we investigate the impacts of using an NDM-based software agent (R-CAST) on the performance of human decision-making teams in a simulated C3I (Communications, Command, Control and Intelligence) environment. We examined four types of decision-making teams with mixed human and agent members playing the roles of intelligence collection and command selection. The experiment also involved two within-group control variables: task complexity and context switching frequency. The result indicates that the use of an R-CAST agent in intelligence collection allows its team member to consider the latest situational information in decision making but might increase the team member's cognitive load. It also indicates that a human member playing the role of command selection should not rely too much on the agent serving as his or her decisi...

Multicontext team decision making under time stress is an extremely challenging issue faced by various real-world application domains. In this paper, we employ an experience-based cognitive agent architecture (called R-CAST) to address the informational challenges associated with military command and control (C2) decision-making teams, the performance of which can be significantly affected by dynamic context switching and tasking complexities. Using context switching frequency and task complexity as two factors, we conducted an experiment to evaluate whether the use of R-CAST agents as teammates and decision aids can benefit C2decision-making teams. Members from a U.S. Army Reserve Officer Training Corps organization were randomly recruited as human participants. They were grouped into ten human-human teams, each composed of two participants, and ten human-agent teams, each composed of one participant and two R-CAST agents, as teammates and decision aids. The statistical inference of experimental results indicates that R-CAST agents can significantly improve the performance of C2teams in multicontext decision making under varying time-stressed situations.

Helping behavior in effective teams is achieved via some overlapping "shared mental models" that are developed and maintained by members of the team. In this paper, we take the perspective that multiparty "proactive" communication is critical for establishing and maintaining such a shared mental model among teammates, which is the basis for agents to offer proactive help and to achieve coherent teamwork. We first provide formal semantics for multiparty proactive performatives within a team setting. We then examine how such performatives result in updates to mental model of teammates, and how such updates can trigger helpful behaviors from other teammates. We also provide conversation policies for multiparty proactive performatives.

MALLET, a multi-agent logic language for encoding teamwork, is intended to enable expression of teamwork emulating human teamwork, allowing experimentation with different levels and forms of inferred team intelligence. A consequence of this goal is that the actual teamwork behavior is determined by the level of intelligence built into the underlying system as well as the semantics of the language. In this paper, we give the design objectives, the syntax, and an operational semantics for MALLET in terms of a transition system. We show how the semantics can be used to reason about the behaviors of team-based agents. The semantics can also be used to guide the implementation of various MALLET interpreters emulating different forms of team intelligence, as well as formally study the properties of team-based agents specified in MALLET. We have explored various forms of proactive information exchange behavior embodied in human teamwork using the CAST system, which implements a built-in MALLET interpreter.

Multiparty communication and proactive communication have each been studied extensively, but separately. A combinatorial approach would be to investigate proactive communication in a multiparty settings, aiming to support effective collaboration by leveraging the benefits of both. In this paper, we take a first attempt along this direction, providing formal definitions for multiparty proactive performatives. Such performatives not only result in an exchange of knowledge among teammates, but also in an increased awareness about teammates' mental states which can initiate additional helpful behaviors.

Currently Web services composition problems are addressed using AI planning techniques. The team-based approach, with emphases on the sharing of mental models and proactive collaboration, provides an alternative to current static approaches to Web service composition. The approach provides clear advantages for proactive handling of failures that may be encountered during execution of a complex Web service. The paper proposes a generic framework for dynamic Web-service composition, and extends the CAST architecture to realize the framework.

Pointers are an important programming concept. They are used explicitly or implicitly in many programming languages. In particular, the semantics of object-oriented programming languages rely on pointers. We introduce a semantics for pointer structures. Pointers are seen as indexes and pointer fields are functions from these indexes to values. Using this semantics we turn all pointer operations into simple assignments and then we use refinement calculus techniques to construct a pointer-manipulating program that checks whether or not a single linked list has a loop. We also introduce an induction principle on pointer structures in order to reduce complexity of the proofs.

Psychological studies about human teamwork have shown that members of an effective team can often anticipate needs of other teammates, and choose to assist teammates through appropriate ways. However the fulfillment of help behaviors among teammates is subject to team members' understanding of the underpinning communicative acts performed by individual agents. By extending the mental state analysis of performatives initiated by Cohen and Levesque (1990), we focus this paper on analyzing a missing class of performatives -proactive performatives, which are prevalently used by individual agents to exchange information among teammates.

Agent oriented programming (AOP) is a special kind of object-oriented programming. It can be worked out best for open systems and has the potentials to become a very attractive technique in the future. In this paper, we describe a specification and programming language-SPLAW. The syntax and operational semantics of SPLAW are presented, and by means of labeled transition system, the proof theory is also provided. SPLAW is based on KQML, the standard inter-agent communication language, which makes it possible for agents written in SPLAW to interoperate with other agents obeying KQML.