skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Continous Representation Learning via User Feedback

Abstract

Representation learning is a deep-learning based technique for extracting features from data for the purpose of machine learning. This requires a large amount of data, on order tens of thousands to millions of samples, to properly teach the deep neural network. This a system for continuous representation learning, where the system may be improved with a small number of additional samples (order 10-100). The unique characteristics of this invention include a human-computer feedback component, where assess the quality of the current representation and then provides a better representation to the system. The system then mixes the new data with old training examples to avoid overfitting and improve overall performance of the system. The model can be exported and shared with other users, and it may be applied to additional images the system hasn't seen before.

Publication Date:
Research Org.:
Pacific Northwest National Lab. (PNNL), Richland, WA (United States)
Sponsoring Org.:
USDOE
OSTI Identifier:
1307215
Report Number(s):
Continous Representation Learning via User Feedbac; 004877MLTPL00
Battelle IPID 30871-E
DOE Contract Number:
AC05-76RL01830
Resource Type:
Software
Software Revision:
00
Software Package Number:
004877
Software CPU:
MLTPL
Source Code Available:
No
Other Software Info:
Copyright software available through PNNL Technology Commercialization.
Country of Publication:
United States

Citation Formats

. Continous Representation Learning via User Feedback. Computer software. Vers. 00. USDOE. 8 Aug. 2016. Web.
. (2016, August 8). Continous Representation Learning via User Feedback (Version 00) [Computer software].
. Continous Representation Learning via User Feedback. Computer software. Version 00. August 8, 2016.
@misc{osti_1307215,
title = {Continous Representation Learning via User Feedback, Version 00},
author = {},
abstractNote = {Representation learning is a deep-learning based technique for extracting features from data for the purpose of machine learning. This requires a large amount of data, on order tens of thousands to millions of samples, to properly teach the deep neural network. This a system for continuous representation learning, where the system may be improved with a small number of additional samples (order 10-100). The unique characteristics of this invention include a human-computer feedback component, where assess the quality of the current representation and then provides a better representation to the system. The system then mixes the new data with old training examples to avoid overfitting and improve overall performance of the system. The model can be exported and shared with other users, and it may be applied to additional images the system hasn't seen before.},
doi = {},
year = 2016,
month = 8,
note =
}

Software:
To order this software, request consultation services, or receive further information, please fill out the following request.

Save / Share:
  • Cloud feedback—the change in top-of-atmosphere radiative flux resulting from the cloud response to warming—constitutes by far the largest source of uncertainty in the climate response to CO2 forcing simulated by global climate models (GCMs). We review the main mechanisms for cloud feedbacks, and discuss their representation in climate models and the sources of intermodel spread. Global-mean cloud feedback in GCMs results from three main effects: (1) rising free-tropospheric clouds (a positive longwave effect); (2) decreasing tropical low cloud amount (a positive shortwave [SW] effect); (3) increasing high-latitude low cloud optical depth (a negative SW effect). These cloud responses simulated bymore » GCMs are qualitatively supported by theory, high-resolution modeling, and observations. Rising high clouds are consistent with the fixed anvil temperature (FAT) hypothesis, whereby enhanced upper-tropospheric radiative cooling causes anvil cloud tops to remain at a nearly fixed temperature as the atmosphere warms. Tropical low cloud amount decreases are driven by a delicate balance between the effects of vertical turbulent fluxes, radiative cooling, large-scale subsidence, and lower-tropospheric stability on the boundary-layer moisture budget. High-latitude low cloud optical depth increases are dominated by phase changes in mixed-phase clouds. The causes of intermodel spread in cloud feedback are discussed, focusing particularly on the role of unresolved parameterized processes such as cloud microphysics, turbulence, and convection.« less
    Cited by 2
  • This report provides a step-by-step guide, or user manual, for personnel responsible for the planning and execution of the verification and validation (V&V), and developmental testing, of expert systems, conventional software systems, and various other types of artificial intelligence systems. While the guide was developed primarily for applications in the utility industry, it applies well to all industries. The user manual has three sections. In Section 1 the user assesses the stringency of V&V needed for the system under consideration, identifies the development stage the system is in, and identifies the component(s) of the system to be tested next. Thesemore » three pieces of information determine which Guideline Package of V&V methods is most appropriate for those conditions. The V&V Guideline Packages are provided in Section 2. Each package consists of an ordered set of V&V techniques to be applied to the system, guides on choosing the review/evaluation team, measurement criteria, and references to a book or report which describes the application of the method. Section 3 presents details of 11 of the most important (or least well-explained in the literature) methods to assist the user in applying these techniques accurately.« less
  • Reliable software is required for nuclear power industry applications. Verification and validation techniques applied during the software development process can help eliminate errors that could inhibit the proper operation of digital systems and cause availability and safety problems. Most of the techniques described in this report are valid for conventional software systems as well as for expert systems. The project resulted in a set of 16 V&V guideline packages and 11 sets of procedures based on the class, development phase, and system component being tested. These guideline packages and procedures help a utility define the level of V&V, which involvesmore » evaluating the complexity and type of software component along with the consequences of failure. In all, the project identified 153 V&V techniques for conventional software systems and demonstrated their application to all aspects of expert systems except for the knowledge base, which requires specially developed tools. Each of these conventional techniques covers anywhere from 2-52 total types of conventional software defects, and each defect is covered by 21-50 V&V techniques. The project also identified automated tools to Support V&V activities.« less
  • There is a need for a spatial decision support application that allows users to create customized metrics for comparing proposed locations of a new solar installation. This document discusses how PVMapper was designed to overcome the customization problem through the development of loosely coupled spatial and decision components in a JavaScript plugin architecture. This allows the user to easily add functionality and data to the system. The paper also explains how PVMapper provides the user with a dynamic and customizable decision tool that enables them to visually modify the formulas that are used in the decision algorithms that convert datamore » to comparable metrics. The technologies that make up the presentation and calculation software stack are outlined. This document also explains the architecture that allows the tool to grow through custom plugins created by the software users. Some discussion is given on the difficulties encountered while designing the system.« less

To initiate an order for this software, request consultation services, or receive further information, fill out the request form below. You may also reach us by email at: .

OSTI staff will begin to process an order for scientific and technical software once the payment and signed site license agreement are received. If the forms are not in order, OSTI will contact you. No further action will be taken until all required information and/or payment is received. Orders are usually processed within three to five business days.

Software Request

(required)
(required)
(required)
(required)
(required)
(required)
(required)
(required)