I detta arbete jämfördes ett LSTM-nätverk med ett feedforward-nätverk för generellt Atari 2600 spelande. Prestandan definierades som poängen agenten får för ett visst spel. Hypotesen var att LSTM skulle prestera minst lika bra som feedforward och förhoppningsvis mycket bättre. För att svara på frågeställningen skapades två olika agenter, en med ett LSTM-nätverk och en med ett feedforward-nätverk. Experimenten utfördes på Stella emulatorn med hjälp av ramverket the Arcade Learning Environment (ALE). Hänsyn togs till Machado råd om inställningar för användning av ALE och hur agenter borde tränas och evalueras samtidigt. Agenterna utvecklades med hjälp av en genetisk algoritm. Resultaten visade att LSTM var minst lika bra som feedforward men båda metoderna blev slagna av Machados metoder. Toppoängen i varje spel jämfördes med Granfelts arbete som har varit en utgångspunkt för detta arbete.

Generellt spelande är ett forskningsområde fokuserat på att skapa AI som kan spela spel utan någon domänspecifik information. Detta arbete har undersökt elman-nätverks potential för generellt Atari-spelande genom att testa ett elman-nätverk och ett feedforward-nätverk via the Arcade Learning Environment. Nätverken använder en pixelrepresentation för att representera spelmiljön och baserar sina handlingar endast på den informationen. Agenterna testades på fyra spel varav två anses kräva en mer avancerad struktur än feedforward. Agenterna evalueras via deras toppoäng i spelen som testas och tränas via en genetisk algoritm. Resultaten visade att elman-strukturen inte presterar bättre än feedforward, dessutom erhölls ingen poäng i de avancerade spelen vilket tyder på att ett korttidsminne inte är tillräckligt för att spela dessa spel. Jämfört med tidigare forskning sågs en liten förbättring över liknande struktur vilket tyder på en förbättrad representation. För att förbättra resultaten i detta arbete borde ett större antal spel testas.

Ten subjects ranging from 9 to 16 years in age wi.th IQ's ranging from 23 to 62 were randomly selected as contingent or noncontingent subjects for two experiments. Five subjects received contingent access to two electronic games for performance within a specified learning session, while five subjects received noncontingent access to the games. These experiments were designed to determine the effect on performance, attending, and compliance skills in the classroom, when contingent access to the electronic games was based on performance. The development of fine motor skills and/or eye-hand coordination skills as a result of game usage was examined. The generalization of any effect to the remainder of the classroom day was also evaluated. The experimental design for these experiments was a single subject multiple baseline design for data on performance with the additional collection of attending and compliance data in a multiple baseline fashion. Probes were utilized to assess generalization effects. A change in performance related to experimental manipulation was noted in three of five of the contingent subjects, while support for subsequent change in attending and compliance was demonstrated by fewer subjects (one subject in regard to attending; three subjects in regard to compliance) . No changes in performance, attending, or compliance related to experimental manipulation were demonstrated by subjects receiving noncontingent access to the games. Nine of ten subjects (contingent and noncontingent) demonstrated gains in age equivalencies on the Upper Limb Coordination subtest of the Bruininks-Oseretsky Test of Motor Proficiency in excess of the duration of the experiment. In addition, six of ten subjects demonstrated gains on the Fine Motor Composite of this test.

This research focuses on developing AI agents that play arbitrary Atari 2600 console games without having any game-specific assumptions or prior knowledge. Two main approaches are considered: reinforcement learning based methods and search based methods. The RL-based methods use feature vectors generated from the game screen as well as the console RAM to learn to play a given game. The search-based methods use the emulator to simulate the consequence of actions into the future, aiming to play as well as possible by only exploring a very small fraction of the state-space. To insure the generic nature of our methods, all agents are designed and tuned using four specific games. Once the development and parameter selection is complete, the performance of the agents is evaluated on a set of 50 randomly selected games. Significant learning is reported for the RL-based methods on most games. Additionally, some instances of human-level performance is achieved by the search-based methods.

No description available.

Thesis (M.Sc.)--University of Alberta, 2010. / Title from PDF file main screen (viewed on July 15, 2010). A thesis submitted to the Faculty of Graduate Studies and Research in partial fulfillment of the requirements for the degree of Master of Science, Department of Computing Science, University of Alberta. Includes bibliographical references.

Deep reinforcement learning algorithms typically require vast amounts of data to train to a useful level of performance. Each time new data is encountered, the network must inefficiently update all of its parameters. Auxiliary memory units can help deep neural networks train more efficiently by separating computation from storage, and providing a means to rapidly store and retrieve precise information. We present four deep reinforcement learning models augmented with external memory, and benchmark their performance on ten tasks from the Arcade Learning Environment. Our discussion and insights will be helpful for future RL researchers developing their own memory agents.

Current communication and network systems are not built for delay-sensitive applications. The most obvious fact is that the communication capacity is only achievable in theory with infinitely long codes, which means infinitely long delays. One remedy for this is to use shorter codes. Conceptually, there is a deeper reason for the difficulties in such solutions: in Shannon's original 1948 paper, he started out by stating that the "semantic aspects" of information is "irrelevant" to communications. Hence, in Shannon's communication system, as well as every network built after him, we put all information into a uniform bit-stream, regardless what meanings they carry, and we transmit these bits over the network as a single type of commodity. Consequently, the network system can only provide a uniform level of error protection and latency control to all these bits. We argue that such a single measure of latency, or Age of Information (AoI), is insufficient for military Internet of Things (IoT) applications that inherently connect the communication network with a cyber-physical system. For example, a self-driving military vehicle might send to the controller a front-view image. Clearly, not everything in the image is equally important for the purpose of steering the vehicle: an approaching vehicle is a much more urgent piece of information than a tree in the background. Similar examples can be seen for other military IoT devices, such as drones and sensors. In this work, we present a new approach that inherently extracts the most critical information in a Military Battlefield IoT scenario by using a metric - called H-Score. This ensures the neural network to only concentrate on the most important information and ignore all background information. We then carry out extensive evaluation of this a by testing it against various inputs, ranging from a vector of numbers to a 1000x1000 pixel image. Next, we introduce the concept of Manual Marginalization, which helps us to make independent decisions for each object in the image. We also develop a video game that captures the essence of a military battlefield scenario and test our developed algorithm here. Finally, we apply our approach on a simple Atari Space Invaders video game to shoot down enemies before they fire at us. / Master of Science / The IoT is transforming military and civilian environments into truly integrated cyberphysical systems (CPS), in which the dynamic physical world is tightly embedded with communication capabilities. This CPS nature of the military IoT will enable it to integrate a plethora of devices, ranging from small sensors to autonomous aerial, ground, and naval vehicles. This results in huge amount of information being transferred between the devices. However, not all the information is equally important. Broadly we can categorize information into two types: Critical and Non-Critical. For example in a military battlefield, the information about enemies is critical and information abouut the background trees is not so important. Therefore, it is essential to isolate the critical information from non-critical informaiton. This is the focus of our work. We use neural networks and some domain knowledge about the enemies to extract the critical information and use the extracted information to take control decisions. We then evalue the performance of this approach by testing it against various kinds of synthetic data sets. Finally we use an Atari Space Invaders video game to demonstrate how the extracted information can be used to make crucial decisions about enemies.

Thesis (B.S.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 1982. / MICROFICHE COPY AVAILABLE IN ARCHIVES AND ENGINEERING / Bibliography: leaf 22. / by Inwhan Choi. / B.S.

Graphical user interfaces for high-assurance systems must fulfill a range of security requirements such as protected and reliable presentation, prevention of unauthorized cross-domain talk, and prevention of user-input eavesdropping. Additionally, it is desirable to support legacy applications running in confined compartments. Standard isolation methods such as virtual-machine monitors provide one frame buffer per security domain, where each frame buffer is managed by one legacy window system. This raises the question of how to safely integrate multiple (legacy) window systems and protect the displayed data while preserving the usability of modern user interfaces. Our paper describes the OverlayWindow System, a general mechanism for multiplexing windows of multiple distinct window systems into the host frame buffer. Thus, each legacy window appears to the user as one corresponding host window that can be moved and resized. To achieve this, only slight modifications of the legacy window system are required whereby, the source code does not have to be available. Our implementation of an Overlay Window System successfully multiplexes Linux, GEM and native L4 applications.