A Learning Algorithm for Continually Running Fully Recurrent Neural Networks

1989 ◽  
Vol 1 (2) ◽  
pp. 270-280 ◽  
Author(s):  
Ronald J. Williams ◽  
David Zipser
Keyword(s):  
Recurrent Neural Networks ◽  
Learning Algorithm ◽  
Recurrent Networks ◽  
Exact Form ◽  
Practical Algorithms ◽  
Learning Tasks ◽  
Training Interval ◽  
Computationally Expensive ◽  
Time Periods ◽  
Over Time

The exact form of a gradient-following learning algorithm for completely recurrent networks running in continually sampled time is derived and used as the basis for practical algorithms for temporal supervised learning tasks. These algorithms have (1) the advantage that they do not require a precisely defined training interval, operating while the network runs; and (2) the disadvantage that they require nonlocal communication in the network being trained and are computationally expensive. These algorithms allow networks having recurrent connections to learn complex tasks that require the retention of information over time periods having either fixed or indefinite length.

Sufficient Conditions for Error Backflow Convergence in Dynamical Recurrent Neural Networks

2002 ◽  
Vol 14 (8) ◽  
pp. 1907-1927 ◽  
Author(s):  
Alex Aussem
Keyword(s):  
Neural Networks ◽  
Recurrent Neural Networks ◽  
Error Propagation ◽  
Finite Impulse Response ◽  
Learning Algorithm ◽  
Previous Analysis ◽  
Sufficient Conditions ◽  
Recurrent Networks ◽  
Computational Overhead ◽  
Propagation Network

This article extends previous analysis of the gradient decay to a class of discrete-time fully recurrent networks, called dynamical recurrent neural networks, obtained by modeling synapses as finite impulse response (FIR) filters instead of multiplicative scalars. Using elementary matrix manipulations, we provide an upper bound on the norm of the weight matrix, ensuring that the gradient vector, when propagated in a reverse manner in time through the error-propagation network, decays exponentially to zero. This bound applies to all recurrent FIR architecture proposals, as well as fixed-point recurrent networks, regardless of delay and connectivity. In addition, we show that the computational overhead of the learning algorithm can be reduced drastically by taking advantage of the exponential decay of the gradient.

scholarly journals Structured and Systematic Team and Procedure Training in Severe Trauma: Going from ‘Zero to Hero’ for a Time-Critical, Low-Volume Emergency Procedure Over Three Time Periods

2021 ◽  
Vol 45 (5) ◽  
pp. 1340-1348
Author(s):  
Maryam Meshkinfamfard ◽  
Jon Kristian Narvestad ◽  
Johannes Wiik Larsen ◽  
Arezo Kanani ◽  
Jørgen Vennesland ◽  
...  
Keyword(s):  
Injury Severity ◽  
Trauma System ◽  
Severe Trauma ◽  
Emergency Procedure ◽  
Emergency Thoracotomy ◽  
Severity Scores ◽  
Time Periods ◽  
Low Volume ◽  
Time Critical ◽  
Over Time

Abstract Background Resuscitative emergency thoracotomy is a potential life-saving procedure but is rarely performed outside of busy trauma centers. Yet the intervention cannot be deferred nor centralized for critically injured patients presenting in extremis. Low-volume experience may be mitigated by structured training. The aim of this study was to describe concurrent development of training and simulation in a trauma system and associated effect on one time-critical emergency procedure on patient outcome. Methods An observational cohort study split into 3 arbitrary time-phases of trauma system development referred to as ‘early’, ‘developing’ and ‘mature’ time-periods. Core characteristics of the system is described for each phase and concurrent outcomes for all consecutive emergency thoracotomies described with focus on patient characteristics and outcome analyzed for trends in time. Results Over the study period, a total of 36 emergency thoracotomies were performed, of which 5 survived (13.9%). The “early” phase had no survivors (0/10), with 2 of 13 (15%) and 3 of 13 (23%) surviving in the development and mature phase, respectively. A decline in ‘elderly’ (>55 years) patients who had emergency thoracotomy occurred with each time period (from 50%, 31% to 7.7%, respectively). The gender distribution and the injury severity scores on admission remained unchanged, while the rate of patients with signs on life (SOL) increased over time. Conclusion The improvement over time in survival for one time-critical emergency procedure may be attributed to structured implementation of team and procedure training. The findings may be transferred to other low-volume regions for improved trauma care.

scholarly journals Selective network discovery via deep reinforcement learning on embedded spaces

2021 ◽  
Vol 6 (1) ◽  
Author(s):  
Peter Morales ◽  
Rajmonda Sulo Caceres ◽  
Tina Eliassi-Rad
Keyword(s):  
Reinforcement Learning ◽  
Learning Algorithm ◽  
Sequential Decision ◽  
Network Discovery ◽  
Learning Tasks ◽  
Partially Observed ◽  
Decision Making Problem ◽  
Resource Collection ◽  
Improved Performance ◽  
Discovery Algorithms

AbstractComplex networks are often either too large for full exploration, partially accessible, or partially observed. Downstream learning tasks on these incomplete networks can produce low quality results. In addition, reducing the incompleteness of the network can be costly and nontrivial. As a result, network discovery algorithms optimized for specific downstream learning tasks given resource collection constraints are of great interest. In this paper, we formulate the task-specific network discovery problem as a sequential decision-making problem. Our downstream task is selective harvesting, the optimal collection of vertices with a particular attribute. We propose a framework, called network actor critic (NAC), which learns a policy and notion of future reward in an offline setting via a deep reinforcement learning algorithm. The NAC paradigm utilizes a task-specific network embedding to reduce the state space complexity. A detailed comparative analysis of popular network embeddings is presented with respect to their role in supporting offline planning. Furthermore, a quantitative study is presented on various synthetic and real benchmarks using NAC and several baselines. We show that offline models of reward and network discovery policies lead to significantly improved performance when compared to competitive online discovery algorithms. Finally, we outline learning regimes where planning is critical in addressing sparse and changing reward signals.

scholarly journals Longitudinal assessment of the bovine ocular bacterial community dynamics in calves

2021 ◽  
Vol 3 (1) ◽  
Author(s):  
Alison C. Bartenslager ◽  
Nirosh D. Althuge ◽  
John Dustin Loy ◽  
Matthew M. Hille ◽  
Matthew L. Spangler ◽  
...  
Keyword(s):  
Bacterial Community ◽  
Community Composition ◽  
Community Dynamics ◽  
Bacterial Community Composition ◽  
Clinical Signs ◽  
Sampling Time ◽  
Experimental Population ◽  
Opportunistic Pathogens ◽  
Time Periods ◽  
Over Time

Abstract Background Infectious Bovine Keratoconjunctivitis (IBK), commonly known as pinkeye, is one of the most significant diseases of beef cattle. As such, IBK costs the US beef industry at least 150 million annually. However, strategies to prevent IBK are limited, with most cases resulting in treatment with antibiotics once the disease has developed. Longitudinal studies evaluating establishment of the ocular microbiota may identify critical risk periods for IBK outbreaks or changes in the microbiota that may predispose animals to IBK. Results In an attempt to characterize the establishment and colonization patterns of the bovine ocular microbiota, we conducted a longitudinal study consisting of 227 calves and evaluated the microbiota composition over time using amplicon sequence variants (ASVs) based on 16S rRNA sequencing data and culture-based approaches. Beef calves on trial consisted of both male (intact) and females. Breeds were composed of purebred Angus and composites with varying percentages of Simmental, Angus, and Red Angus breeds. Average age at the start of the trial was 65 days ±15.02 and all calves remained nursing on their dam until weaning (day 139 of the study). The trial consisted of 139 days with four sampling time points on day 0, 21, 41, and 139. The experimental population received three different vaccination treatments (autogenous, commercial (both inactivated bacteria), and adjuvant placebo), to assess the effectiveness of different vaccines for IBK prevention. A significant change in bacterial community composition was observed across time periods sampled compared to the baseline (p < 0.001). However, no treatment effect of vaccine was detected within the ocular bacterial community. The bacterial community composition with the greatest time span between sampling time periods (98d span) was most similar to the baseline sample collected, suggesting re-establishment of the ocular microbiota to baseline levels over time after perturbation. The effect of IgA levels on the microbial community was investigated in a subset of cattle within the study. However, no significant effect of IgA was observed. Significant changes in the ocular microbiota were identified when comparing communities pre- and post-clinical signs of IBK. Additionally, dynamic changes in opportunistic pathogens Moraxella spp. were observed and confirmed using culture based methods. Conclusions Our results indicate that the bovine ocular microbiota is well represented by opportunistic pathogens such as Moraxella and Mycoplasma. Furthermore, this study characterizes the diversity of the ocular microbiota in calves and demonstrates the plasticity of the ocular microbiota to change. Additionally, we demonstrate the ocular microbiome in calves is similar between the eyes and the perturbation of one eye results in similar changes in the other eye. We also demonstrate the bovine ocular microbiota is slow to recover post perturbation and as a result provide opportunistic pathogens a chance to establish within the eye leading to IBK and other diseases. Characterizing the dynamic nature of the ocular microbiota provides novel opportunities to develop potential probiotic intervention to reduce IBK outbreaks in cattle.

scholarly journals Evolutionary history of zoogeographical regions surrounding the Tibetan Plateau

2020 ◽  
Vol 3 (1) ◽  
Author(s):  
Jiekun He ◽  
Siliang Lin ◽  
Jiatang Li ◽  
Jiehua Yu ◽  
Haisheng Jiang
Keyword(s):  
Tibetan Plateau ◽  
Evolutionary History ◽  
The Tibetan Plateau ◽  
Change Over Time ◽  
Deep Time ◽  
Terrestrial Vertebrates ◽  
History Of ◽  
Striking Change ◽  
Time Periods ◽  
Over Time

AbstractThe Tibetan Plateau (TP) and surrounding regions have one of the most complex biotas on Earth. However, the evolutionary history of these regions in deep time is poorly understood. Here, we quantify the temporal changes in beta dissimilarities among zoogeographical regions during the Cenozoic using 4,966 extant terrestrial vertebrates and 1,278 extinct mammal genera. We identify ten present-day zoogeographical regions and find that they underwent a striking change over time. Specifically, the fauna on the TP was close to the Oriental realm in deep time but became more similar to the Palearctic realms more recently. The present-day zoogeographical regions generally emerged during the Miocene/Pliocene boundary (ca. 5 Ma). These results indicate that geological events such as the Indo-Asian Collision, the TP uplift, and the aridification of the Asian interior underpinned the evolutionary history of the zoogeographical regions surrounding the TP over different time periods.

scholarly journals Phagolysosome formation in normal and colchicine-treated macrophages.

1975 ◽  
Vol 142 (4) ◽  
pp. 903-913 ◽  
Author(s):  
E L Pesanti ◽  
S G Axline
Keyword(s):  
Acid Phosphatase ◽  
Lysosomal Enzyme ◽  
Peritoneal Macrophages ◽  
Cell Type ◽  
Inert Particles ◽  
Time Periods ◽  
Mouse Peritoneal Macrophages ◽  
Over Time

Intracellular lysosomal fusion has been evaluated in cultivated mouse peritoneal macrophages by measurement of transfer of acid phosphatase to polyvinyltoluene (PVT)-containing phagolysosomes. Enzyme transfer was found to be directly and significantly related to the uptake of PVT and to be independent of time allowed for phagolysosome formation over time periods of 15 min to 18 h. In addition, the extent of transfer of lysosomal enzyme to phagolysosomes was unaffected by treatment of the cells with 10(-6) M colchicine, a dose which eradicates morphologically identifiable microtubules in this cell type within 2 h. The data indicate that intracellular fusion of lysosomes with phagosomes in the macrophage does not require formed microtubules and suggest that fusion occurs promptly after interiorization of inert particles.

Precarity Unbound

2007 ◽  
Vol 32 (3) ◽  
pp. 319-340 ◽  
Author(s):  
Nancy Ettlinger
Keyword(s):  
Everyday Life ◽  
Final Section ◽  
Multiple Scales ◽  
Bush Administration ◽  
Radical Democracy ◽  
Top Down ◽  
Hardt And Negri ◽  
Global Events ◽  
Time Periods ◽  
Over Time

Departing from tendencies to bound precarity in particular time periods and world regions, this article develops an expansive view of precarity over time and across space. Beyond effects of specific global events and macroscale structures, precarity inhabits the microspaces of everyday life. However, people attempt to disengage the stress of precarious life by constructing the illusion of certainty. Reflexive denial of precarious life entails essentialist strategies that implicitly or explicitly classify and homogenize people and phenomena, legitimize the constructed boundaries, and in the process aim at eliminating difference and possibilities for negotiation; the tension between these goals and material realities helps explain misrepresentations that can be catastrophic at multiple scales, re-creating precarity. Reactions to 9/11 by the Bush administration represent a case in point of reflexive denial of precarity through strategies that created illusions of certainty with deleterious results. Normatively, the paradox of precarious life and reflexive denials prompts questions as to how urges for certainty in the context of precarity might be constructively channeled. the author approaches this challenge in the final section by drawing from a nexus of concerns about post-Habermasian radical democracy, individual thought and feeling, and network dynamics. Whereas Hardt and Negri reverse the direction of the Foucauldian concept of biopower from top-down to bottom-up, the author draws from Foucault's concept of governmentality in relation to resistance to imagine a cooperative politics operating within as well as across scales.

scholarly journals Visualizing Belief in Meritocracy, 1930–2010

2018 ◽  
Vol 4 ◽  
pp. 237802311881180 ◽  
Author(s):  
Jonathan J. B. Mijs
Keyword(s):  
Survey Data ◽  
Representative Sample ◽  
Western World ◽  
Birth Cohorts ◽  
Popular Belief ◽  
Time Period ◽  
Level Data ◽  
Two Measures ◽  
Time Periods ◽  
Over Time

In this figure I describe the long trend in popular belief in meritocracy across the Western world between 1930 and 2010. Studying trends in attitudes is limited by the paucity of survey data that can be compared across countries and over time. Here, I show how to complement survey waves with cohort-level data. Repeated surveys draw on a representative sample of the population to describe the typical beliefs held by citizens in a given country and period. Leveraging the fact that citizens surveyed in a given year were born in different time-periods allows for a comparison of beliefs across birth cohorts. The latter overlaps with the former, but considerably extends the time period covered by the data. Taken together, the two measures give a “triangulated” longitudinal record of popular belief in meritocracy. I find that in most countries, popular belief in meritocracy is (much) stronger for more recent periods and cohorts.

scholarly journals Change in Metaphorical Framing: Metaphors of TRade in 225 Years of State of the Union Addresses (1790–2014)

2018 ◽  
Vol 41 (2) ◽  
pp. 260-279 ◽  
Author(s):  
Christian Burgers ◽  
Kathleen Ahrens
Keyword(s):  
Relative Frequency ◽  
Conceptual Metaphor ◽  
Conceptual Metaphor Theory ◽  
Source Domain ◽  
State Of The Union ◽  
Metaphor Theory ◽  
Time Periods ◽  
Diachronic Analysis ◽  
Types Of Change ◽  
Over Time

AbstractThe literature provides diverging perspectives on the universality and stability of economic metaphors over time. This article contains a diachronic analysis of economic metaphors describing trade in a corpus of 225 years of US State of the Union addresses (1790–2014). We focused on two types of change: (i) replacement of a source domain by another domain and (ii) change in mapping within a source domain. In our corpus, five source domains of trade were predominant: (i) PhysicalObject, (ii) Building, (iii) Container, (iv) Journey, and (v) LivingBeing. Only the relative frequency of the Container source domain was related to time. Additionally, mappings between source and target domains were mostly stable. Nevertheless, our analyses suggest that the Trade metaphors in our corpus are related to concreteness in a more nuanced way as typically assumed in conceptual metaphor theory: metaphors high in the concreteness dimension of physicality and low in the concreteness dimension of specificity are likeliest to be used over longer time periods, by providing communicators with freedom to adjust the metaphor to changing societal circumstances.

scholarly journals Transfer Learning with Deep Recurrent Neural Networks for Remaining Useful Life Estimation

2018 ◽  
Vol 8 (12) ◽  
pp. 2416 ◽  
Author(s):  
Ansi Zhang ◽  
Honglei Wang ◽  
Shaobo Li ◽  
Yuxin Cui ◽  
Zhonghao Liu ◽  
...  
Keyword(s):  
Neural Networks ◽  
Transfer Learning ◽  
Recurrent Neural Networks ◽  
Prediction Models ◽  
Learning Algorithm ◽  
Remaining Useful Life ◽  
Operating Conditions ◽  
Training Data ◽  
Good Prediction ◽  
Useful Life

Prognostics, such as remaining useful life (RUL) prediction, is a crucial task in condition-based maintenance. A major challenge in data-driven prognostics is the difficulty of obtaining a sufficient number of samples of failure progression. However, for traditional machine learning methods and deep neural networks, enough training data is a prerequisite to train good prediction models. In this work, we proposed a transfer learning algorithm based on Bi-directional Long Short-Term Memory (BLSTM) recurrent neural networks for RUL estimation, in which the models can be first trained on different but related datasets and then fine-tuned by the target dataset. Extensive experimental results show that transfer learning can in general improve the prediction models on the dataset with a small number of samples. There is one exception that when transferring from multi-type operating conditions to single operating conditions, transfer learning led to a worse result.

Sign in / Sign up

Export Citation Format

Share Document