Results for “representation learning” — Page 2

NASA NTRS1991

On the integration of reinforcement learning and approximate reasoning for control

The author discusses the importance of strengthening the knowledge representation characteristic of reinforcement learning techniques using methods such as approximate reasoning. The ARIC (approximate reasoning-based intelligent control) architecture is an example of such a hybrid approach in which the fuzzy control rules are modified (fine-tuned) using reinforcement learning. ARIC also demonstrates that it is possible to start with an approximately correct control knowledge base and learn to refine this knowledge through further experience. On the other hand, techniques such as the TD (temporal difference) algorithm and Q-learning establish stronger theoretical foundations for their use in adaptive control and also in stability analysis of hybrid reinforcement learning and approximate reasoning-based controllers.

Berenji, Hamid R.↗

NASA NTRS1988

A survey of planning and scheduling research at the NASA Ames Research Center

NASA Ames Research Center has a diverse program in planning and scheduling. This paper highlights some of our research projects as well as some of our applications. Topics addressed include machine learning techniques, action representations and constraint-based scheduling systems. The applications discussed are planetary rovers, Hubble Space Telescope scheduling, and Pioneer Venus orbit scheduling.

Zweben, Monte↗

NASA NTRS1989

A survey of planning and scheduling research at the NASA Ames Research Center

NASA Ames Research Center has a diverse program in planning and scheduling. Some research projects as well as some applications are highlighted. Topics addressed include machine learning techniques, action representations and constraint-based scheduling systems. The applications discussed are planetary rovers, Hubble Space Telescope scheduling, and Pioneer Venus orbit scheduling.

Zweben, Monte↗

NASA NTRS1998

Preliminary Work for Examining the Scalability of Reinforcement Learning

Researchers began studying automated agents that learn to perform multiple-step tasks early in the history of artificial intelligence (Samuel, 1963; Samuel, 1967; Waterman, 1970; Fikes, Hart & Nilsonn, 1972). Multiple-step tasks are tasks that can only be solved via a sequence of decisions, such as control problems, robotics problems, classic problem-solving, and game-playing. The objective of agents attempting to learn such tasks is to use the resources they have available in order to become more proficient at the tasks. In particular, each agent attempts to develop a good policy, a mapping from states to actions, that allows it to select actions that optimize a measure of its performance on the task; for example, reducing the number of steps necessary to complete the task successfully. Our study focuses on reinforcement learning, a set of learning techniques where the learner performs trial-and-error experiments in the task and adapts its policy based on the outcome of those experiments. Much of the work in reinforcement learning has focused on a particular, simple representation, where every problem state is represented explicitly in a table, and associated with each state are the actions that can be chosen in that state. A major advantage of this table lookup representation is that one can prove that certain reinforcement learning techniques will develop an optimal policy for the current task. The drawback is that the representation limits the application of reinforcement learning to multiple-step tasks with relatively small state-spaces. There has been a little theoretical work that proves that convergence to optimal solutions can be obtained when using generalization structures, but the structures are quite simple. The theory says little about complex structures, such as multi-layer, feedforward artificial neural networks (Rumelhart & McClelland, 1986), but empirical results indicate that the use of reinforcement learning with such structures is promising. These empirical results make no theoretical claims, nor compare the policies produced to optimal policies. A goal of our work is to be able to make the comparison between an optimal policy and one stored in an artificial neural network. A difficulty of performing such a study is finding a multiple-step task that is small enough that one can find an optimal policy using table lookup, yet large enough that, for practical purposes, an artificial neural network is really required. We have identified a limited form of the game OTHELLO as satisfying these requirements. The work we report here is in the very preliminary stages of research, but this paper provides background for the problem being studied and a description of our initial approach to examining the problem. In the remainder of this paper, we first describe reinforcement learning in more detail. Next, we present the game OTHELLO. Finally we argue that a restricted form of the game meets the requirements of our study, and describe our preliminary approach to finding an optimal solution to the problem.

Clouse, Jeff↗

NASA NTRS1990

The real-time learning mechanism of the Scientific Research Associates Advanced Robotic System (SRAARS)

Scientific research associates advanced robotic system (SRAARS) is an intelligent robotic system which has autonomous learning capability in geometric reasoning. The system is equipped with one global intelligence center (GIC) and eight local intelligence centers (LICs). It controls mainly sixteen links with fourteen active joints, which constitute two articulated arms, an extensible lower body, a vision system with two CCD cameras and a mobile base. The on-board knowledge-based system supports the learning controller with model representations of both the robot and the working environment. By consecutive verifying and planning procedures, hypothesis-and-test routines and learning-by-analogy paradigm, the system would autonomously build up its own understanding of the relationship between itself (i.e., the robot) and the focused environment for the purposes of collision avoidance, motion analysis and object manipulation. The intelligence of SRAARS presents a valuable technical advantage to implement robotic systems for space exploration and space station operations.

Chen, Alexander Y.↗

NASA NTRSDate not supplied

Discovery and Analysis of Rare High-Impact Failure Modes using Adversarial RL-Informed Sampling

Adaptive learning agents have tremendous potential to handle critical tasks currently performed by humans. Unfortunately, due to their complexity, it can be difficult to verify that these learning agents do not have critical failure modes. Standard verification and validation methods often do not apply directly to learning agents and Monte Carlo methods have difficulty covering even a small fraction of the state space, especially in multiagent systems or over long time horizons. To overcome this difficulty, we demonstrate an adaptive stress-testing method based on reinforcement learning of correlations that raise the probability of failure. This approach has three key properties: (1) it is able to find rare failure modes with far greater sample efficiency than Monte Carlo methods, (2) it can estimate the true probability of a failure mode despite the inherent bias in the learning method, and (3) it is capable of learning and resampling compact representations of multimodal failure spaces. These properties are important in practice as we need to find disparate failure modes while accounting for their actual relevance. This is a significant advantage over traditional adaptive stress testing methods that give abstract likelihoods of particular failure instances, but cannot estimate the probability of a broader failure mode. We test our algorithm on a simple problem from the aviation domain where an autonomous aircraft lands in gusty wind conditions. The results suggest that we can find failure modes with far fewer samples than the Monte Carlo approach and simultaneously estimate the probability of failure.

reinforcement learning↗

NASA NTRSDate not supplied

Discovery and Analysis of Rare High-Impact Failure Modes using Adversarial RL-Informed Sampling

Adaptive learning agents have tremendous potential to handle critical tasks currently performed by humans. Unfortunately, due to their complexity, it can be difficult to verify that these learning agents do not have critical failure modes. Standard verification and validation methods often do not apply directly to learning agents and Monte Carlo methods have difficulty covering even a small fraction of the state space, especially in multiagent systems or over long time horizons. To overcome this difficulty, we demonstrate an adaptive stress-testing method based on reinforcement learning of correlations that raise the probability of failure. This approach has three key properties: (1) it is able to find rare failure modes with far greater sample efficiency than Monte Carlo methods, (2) it can estimate the true probability of a failure mode despite the inherent bias in the learning method, and (3) it is capable of learning and resampling compact representations of multimodal failure spaces. These properties are important in practice as we need to find disparate failure modes while accounting for their actual relevance. This is a significant advantage over traditional adaptive stress testing methods that give abstract likelihoods of particular failure instances, but cannot estimate the probability of a broader failure mode. We test our algorithm on a simple problem from the aviation domain where an autonomous aircraft lands in gusty wind conditions. The results suggest that we can find failure modes with far fewer samples than the Monte Carlo approach and simultaneously estimate the probability of failure.

Validation↗

NASA NTRSDate not supplied

Discovery and Analysis of Rare High-Impact Failure Modes using Adversarial RL-Informed Sampling

Adaptive learning agents have tremendous potential to handle critical tasks currently performed by humans. Unfortunately, due to their complexity, it can be difficult to verify that these learning agents do not have critical failure modes. Standard verification and validation methods often do not apply directly to learning agents and Monte Carlo methods have difficulty covering even a small fraction of the state space, especially in multiagent systems or over long time horizons. To overcome this difficulty, we demonstrate an adaptive stress-testing method based on reinforcement learning of correlations that raise the probability of failure. This approach has three key properties: (1) it is able to find rare failure modes with far greater sample efficiency than Monte Carlo methods, (2) it can estimate the true probability of a failure mode despite the inherent bias in the learning method, and (3) it is capable of learning and resampling compact representations of multimodal failure spaces. These properties are important in practice as we need to find disparate failure modes while accounting for their actual relevance. This is a significant advantage over traditional adaptive stress testing methods that give abstract likelihoods of particular failure instances, but cannot estimate the probability of a broader failure mode. We test our algorithm on a simple problem from the aviation domain where an autonomous aircraft lands in gusty wind conditions. The results suggest that we can find failure modes with far fewer samples than the Monte Carlo approach and simultaneously estimate the probability of failure.

Validation↗

NASA NTRS1993

A hierarchical structure for representing and learning fuzzy rules

Yager provides an example in which the flat representation of fuzzy if-then rules leads to unsatisfactory results. Consider a rule base consisting to two rules: if U is 12 the V is 29; if U is (10-15) the V is (25-30). If U = 12 we would get V is G where G = (25-30). The application of the defuzzification process leads to a selection of V = 27.5. Thus we see that the very specific instruction was not followed. The problem with the technique used is that the most specific information was swamped by the less specific information. In this paper we shall provide for a new structure for the representation of fuzzy if-then rules. The representational form introduced here is called a Hierarchical Prioritized Structure (HPS) representation. Most importantly in addition to overcoming the problem illustrated in the previous example this HPS representation has an inherent capability to emulate the learning of general rules and provides a reasonable accurate cognitive mapping of how human beings store information.

Yager, Ronald R.↗

NASA NTRS2015

Predicting Time Series Outputs and Time-to-Failure for an Aircraft Controller Using Bayesian Modeling

Safety of unmanned aerial systems (UAS) is paramount, but the large number of dynamically changing controller parameters makes it hard to determine if the system is currently stable, and the time before loss of control if not. We propose a hierarchical statistical model using Treed Gaussian Processes to predict (i) whether a flight will be stable (success) or become unstable (failure), (ii) the time-to-failure if unstable, and (iii) time series outputs for flight variables. We first classify the current flight input into success or failure types, and then use separate models for each class to predict the time-to-failure and time series outputs. As different inputs may cause failures at different times, we have to model variable length output curves. We use a basis representation for curves and learn the mappings from input to basis coefficients. We demonstrate the effectiveness of our prediction methods on a NASA neuro-adaptive flight control system.

Statistics↗

NASA NTRS2005

Learning from the Mars Rover Mission: Scientific Discovery, Learning and Memory

Purpose: Knowledge management for space exploration is part of a multi-generational effort. Each mission builds on knowledge from prior missions, and learning is the first step in knowledge production. This paper uses the Mars Exploration Rover mission as a site to explore this process. Approach: Observational study and analysis of the work of the MER science and engineering team during rover operations, to investigate how learning occurs, how it is recorded, and how these representations might be made available for subsequent missions. Findings: Learning occurred in many areas: planning science strategy, using instrumen?s within the constraints of the martian environment, the Deep Space Network, and the mission requirements; using software tools effectively; and running two teams on Mars time for three months. This learning is preserved in many ways. Primarily it resides in individual s memories. It is also encoded in stories, procedures, programming sequences, published reports, and lessons learned databases. Research implications: Shows the earliest stages of knowledge creation in a scientific mission, and demonstrates that knowledge management must begin with an understanding of knowledge creation. Practical implications: Shows that studying learning and knowledge creation suggests proactive ways to capture and use knowledge across multiple missions and generations. Value: This paper provides a unique analysis of the learning process of a scientific space mission, relevant for knowledge management researchers and designers, as well as demonstrating in detail how new learning occurs in a learning organization.

Linde, Charlotte↗

NASA NTRS2009

NASA Tech Briefs, August 2009

Topics covered include: Aligning a Receiving Antenna Array to Reduce Interference; Collecting Ground Samples for Balloon-Borne Instruments; Tethered Pyrotechnic Apparatus for Acquiring a Ground Sample; Enhanced Video-Oculography System; Joint Carrier-Phase Synchronization and LDPC Decoding; Dual-Polarization, Sideband-Separating, Balanced Receiver for 1.5 THz Modular Battery Charge Controller; Efficient Multiplexer FPGA Block Structures Based on G4FETs; VLSI Microsystem for Rapid Bioinformatic Pattern Recognition; Low-Noise Amplifier for 100 to 180 GHz; Improved Fabrication of Ceramic Matrix Composite/Foam Core Integrated Structures; Inert Welding/Brazing Gas Filters and Dryers; Fabricating Copper Nanotubes by Electrodeposition; Reducing Aerodynamic Drag on Empty Open Cargo Vehicles; Rotary Percussive Auto-Gopher for Deep Drilling and Sampling; More About Reconfigurable Exploratory Robotic Vehicles; Thermostatic Valves Containing Silicone-Oil Actuators; Improving Heat Flux Performance of Flat Surface in Spray-Cooling Systems; Treating Fibrous Insulation to Reduce Thermal Conductivity; Silica-Aerogel Composites Opacified with La(sub0.7)Sr(sub0.3)MnO3; Cyclic Oxidation Behavior of CuCrAl Cold-Sprayed Coatings for Reusable Launch Vehicles; Ceramic Fiber Structures for Cryogenic Load-Bearing Applications; Elastomer Reinforced with Carbon Nanotubes; Biologically Inspired Purification and Dispersion of SWCNTs; A Technique for Adjusting Eigenfrequencies of WGM Resonators; Low-Pressure, Field-Ionizing Mass Spectrometer; Modifying Operating Cycles to Increase Stability in a LITS; Chamber for Simulating Martian and Terrestrial Environments; Algorithm for Detecting a Bright Spot in an Image; Extreme Programming: Maestro Style; Adaptive Behavior for Mobile Robots; Protocol for Communication Networking for Formation Flying; Planning Complex Sequences Using Compressed Representations; and Self-Supervised Learning of Terrain Traversability from Proprioceptive Sensors.

Source record↗

NASA NTRS2002

Solving and Learning Soft Temporal Constraints: Experimental Setting and Results

Soft temporal constraints problems allow to describe in a natural way scenarios where events happen over time and preferences are associated to event distances and durations. However, sometimes such local preferences are difficult to set, and it may be easier instead to associate preferences to some complete solutions of the problem. Machine learning techniques can be useful in this respect. In this paper we describe two solvers (one more general and the other one more efficient) for tractable subclasses of soft temporal problems, and we show some experimental results. The random generator used to build the problems on which tests are performed is also described. We also compare the two solvers highlighting the tradeoff between performance and representational power. Finally, we present a learning module and we show its behavior on randomly-generated examples.

Rossi, F.↗

NASA NTRS2010

Algorithms for Learning Preferences for Sets of Objects

A method is being developed that provides for an artificial-intelligence system to learn a user's preferences for sets of objects and to thereafter automatically select subsets of objects according to those preferences. The method was originally intended to enable automated selection, from among large sets of images acquired by instruments aboard spacecraft, of image subsets considered to be scientifically valuable enough to justify use of limited communication resources for transmission to Earth. The method is also applicable to other sets of objects: examples of sets of objects considered in the development of the method include food menus, radio-station music playlists, and assortments of colored blocks for creating mosaics. The method does not require the user to perform the often-difficult task of quantitatively specifying preferences; instead, the user provides examples of preferred sets of objects. This method goes beyond related prior artificial-intelligence methods for learning which individual items are preferred by the user: this method supports a concept of setbased preferences, which include not only preferences for individual items but also preferences regarding types and degrees of diversity of items in a set. Consideration of diversity in this method involves recognition that members of a set may interact with each other in the sense that when considered together, they may be regarded as being complementary, redundant, or incompatible to various degrees. The effects of such interactions are loosely summarized in the term portfolio effect. The learning method relies on a preference representation language, denoted DD-PREF, to express set-based preferences. In DD-PREF, a preference is represented by a tuple that includes quality (depth) functions to estimate how desired a specific value is, weights for each feature preference, the desired diversity of feature values, and the relative importance of diversity versus depth. The system applies statistical concepts to estimate quantitative measures of the user s preferences from training examples (preferred subsets) specified by the user. Once preferences have been learned, the system uses those preferences to select preferred subsets from new sets. The method was found to be viable when tested in computational experiments on menus, music playlists, and rover images. Contemplated future development efforts include further tests on more diverse sets and development of a sub-method for (a) estimating the parameter that represents the relative importance of diversity versus depth, and (b) incorporating background knowledge about the nature of quality functions, which are special functions that specify depth preferences for features.

Wagstaff, Kiri L.↗

NASA NTRS1998

A Self-Tuning Kalman Filter for Autonomous Spacecraft Navigation

Most navigation systems currently operated by NASA are ground-based, and require extensive support to produce accurate results. Recently developed systems that use Kalman Filter and Global Positioning System (GPS) data for orbit determination greatly reduce dependency on ground support, and have potential to provide significant economies for NASA spacecraft navigation. Current techniques of Kalman filtering, however, still rely on manual tuning from analysts, and cannot help in optimizing autonomy without compromising accuracy and performance. This paper presents an approach to produce a high accuracy autonomous navigation system fully integrated with the flight system. The resulting system performs real-time state estimation by using an Extended Kalman Filter (EKF) implemented with high-fidelity state dynamics model, as does the GPS Enhanced Orbit Determination Experiment (GEODE) system developed by the NASA Goddard Space Flight Center. Augmented to the EKF is a sophisticated neural-fuzzy system, which combines the explicit knowledge representation of fuzzy logic with the learning power of neural networks. The fuzzy-neural system performs most of the self-tuning capability and helps the navigation system recover from estimation errors. The core requirement is a method of state estimation that handles uncertainties robustly, capable of identifying estimation problems, flexible enough to make decisions and adjustments to recover from these problems, and compact enough to run on flight hardware. The resulting system can be extended to support geosynchronous spacecraft and high-eccentricity orbits. Mathematical methodology, systems and operations concepts, and implementation of a system prototype are presented in this paper. Results from the use of the prototype to evaluate optimal control algorithms implemented are discussed. Test data and major control issues (e.g., how to define specific roles for fuzzy logic to support the self-learning capability) are also discussed. In addition, architecture of a complete end-to-end candidate flight system that provides navigation with highly autonomous control using data from GPS is presented.

Truong, Son H.↗

NASA NTRS2019

Reinforcement Learning Applied to Cognitive Space Communications

The future of space exploration depends on robust, reliable communication systems. As the number of such communication systems increase, automation is fast becoming a requirement to achieve this goal. A reinforcement learning solution can be employed as a possible automation method for such systems. The goal of this study is to build a reinforcement learning algorithm which optimizes data throughput of a single actor. A training environment was created to simulate a link within the NASA Space Communication and Navigation (SCaN) infrastructure, using state of the art simulation tools developed by the SCaN Center for Engineering, Networks, Integration, and Communications (SCENIC) laboratory at NASA Glenn Research Center to obtain the closest possible representation of the real operating environment. Reinforcement learning was then used to train an agent inside this environment to maximize data throughput. The simulation environment contained a single actor in low earth orbit capable of communicating with twenty-five ground stations that compose the Near-Earth Network (NEN). Initial experiments showed promising training results, so additional complexity was added by augmenting simulation data with link fading profiles obtained from real communication events with the International Space Station. A grid search was performed to find the optimal hyperparameters and model architecture for the agent. Using the results of the grid search, an agent was trained on the augmented training data. Testing shows that the agent performs well inside the training environment and can be used as a foundation for future studies with added complexity and eventually tested in the real space environment.

Schubert, Carson D.↗

NASA NTRS1989

Production system chunking in SOAR: Case studies in automated learning

A preliminary study of SOAR, a general intelligent architecture for automated problem solving and learning, is presented. The underlying principles of universal subgoaling and chunking were applied to a simple, yet representative, problem in artificial intelligence. A number of problem space representations were examined and compared. It is concluded that learning is an inherent and beneficial aspect of problem solving. Additional studies are suggested in domains relevant to mission planning and to SOAR itself.

Allen, Robert↗

NASA NTRS2001

Implementation of a General Real-Time Visual Anomaly Detection System Via Soft Computing

The intelligent visual system detects anomalies or defects in real time under normal lighting operating conditions. The application is basically a learning machine that integrates fuzzy logic (FL), artificial neural network (ANN), and generic algorithm (GA) schemes to process the image, run the learning process, and finally detect the anomalies or defects. The system acquires the image, performs segmentation to separate the object being tested from the background, preprocesses the image using fuzzy reasoning, performs the final segmentation using fuzzy reasoning techniques to retrieve regions with potential anomalies or defects, and finally retrieves them using a learning model built via ANN and GA techniques. FL provides a powerful framework for knowledge representation and overcomes uncertainty and vagueness typically found in image analysis. ANN provides learning capabilities, and GA leads to robust learning results. An application prototype currently runs on a regular PC under Windows NT, and preliminary work has been performed to build an embedded version with multiple image processors. The application prototype is being tested at the Kennedy Space Center (KSC), Florida, to visually detect anomalies along slide basket cables utilized by the astronauts to evacuate the NASA Shuttle launch pad in an emergency. The potential applications of this anomaly detection system in an open environment are quite wide. Another current, potentially viable application at NASA is in detecting anomalies of the NASA Space Shuttle Orbiter's radiator panels.

Dominguez, Jesus A.↗