1,211 Hits in 4.9 sec

Stochastic Control with Stale Information--Part I: Fully Observable Systems [article]

Touraj Soleymani, John S. Baras, Karl H. Johansson
2018 arXiv   pre-print
In this study, we adopt age of information as a measure of the staleness of information, and take initial steps towards analyzing the control performance of stochastic systems with stale information.  ...  These solutions are in fact a control policy, which specifies the control inputs of the plant, and a queuing policy, which specifies the staleness of information at the controller.  ...  However, timeliness has not yet studied systematically from the control perspective. B. Contributions and Outline In Part I of our study, we concentrate on fully observable systems.  ... 
arXiv:1810.10983v1 fatcat:lzjmzuxlqzcunghuzamc2ctmi4

More Effective Distributed ML via a Stale Synchronous Parallel Parameter Server

Qirong Ho, James Cipar, Henggang Cui, Jin Kyu Kim, Seunghak Lee, Phillip B Gibbons, Garth A Gibson, Gregory R Ganger, Eric P Xing
2013 Advances in Neural Information Processing Systems  
We propose a parameter server system for distributed ML, which follows a Stale Synchronous Parallel (SSP) model of computation that maximizes the time computational workers spend doing useful work on ML  ...  We provide a proof of correctness under SSP, as well as empirical results demonstrating that the SSP model achieves faster algorithm convergence on several different ML problems, compared to fully-synchronous  ...  This work is supported in part by NIH 1R01GM087694 and 1R01GM093156, DARPA FA87501220324, and NSF IIS1111142 to Eric P. Xing.  ... 
pmid:25400488 pmcid:PMC4230489 fatcat:7zsk6nl6ibhwfe3ukmsuipy2xe

MindTheStep-AsyncPSGD: Adaptive Asynchronous Parallel Stochastic Gradient Descent [article]

Karl Bäckström, Marina Papatriantafilou, Philippas Tsigas
2019 arXiv   pre-print
Stochastic Gradient Descent (SGD) is very useful in optimization problems with high-dimensional non-convex target functions, and hence constitutes an important component of several Machine Learning and  ...  Asynchronous, parallel SGD (AsyncPSGD) has received particular attention, due to observed performance benefits.  ...  ACKNOWLEDGEMENTS This work was partially supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP), Knut and Alice Wallenberg Foundation, the SSF proj. "FiC" nr.  ... 
arXiv:1911.03444v1 fatcat:mcjtmmz6qjd35e3ntdlnn45apm

Model-Based Value Estimation for Efficient Model-Free Reinforcement Learning [article]

Vladimir Feinberg, Alvin Wan, Ion Stoica, Michael I. Jordan, Joseph E. Gonzalez, Sergey Levine
2018 arXiv   pre-print
Such methods hold the promise of incorporating imagined data coupled with a notion of model uncertainty to accelerate the learning of continuous control tasks.  ...  We present model-based value expansion, which controls for uncertainty in the model by only allowing imagination to fixed depth.  ...  This research is supported in part by DHS Award HSHQDC-16-3-00083, NSF CISE Expeditions Award CCF-1139158, and gifts from Alibaba, Amazon Web Services, Ant Financial, CapitalOne, Ericsson, GE, Google,  ... 
arXiv:1803.00101v1 fatcat:ri2egvapd5dk7mh27pfm64txca

Managed communication and consistency for fast data-parallel iterative analytics

Jinliang Wei, Wei Dai, Aurick Qiao, Qirong Ho, Henggang Cui, Gregory R. Ganger, Phillip B. Gibbons, Garth A. Gibson, Eric P. Xing
2015 Proceedings of the Sixth ACM Symposium on Cloud Computing - SoCC '15  
Additionally, the performance of the widely used stochastic gradient descent (SGD) algorithm is sensitive to step size.  ...  when updating shared model parameters to maximize parallelism, the accumulated error may seriously impact the quality of refinements and thus delay completion time, a problem that usually gets worse with  ...  We thank Mu Li, Jin Kyu Kim, Aaron Harlap, Xun Zheng and Zhiting Hu for their suggestions and help with setting up other third-party systems for comparison.  ... 
doi:10.1145/2806777.2806778 dblp:conf/cloud/WeiDQHCGGGX15 fatcat:mgqx7iwlare3tciivbyszgk5oq

Staleness-aware Async-SGD for Distributed Deep Learning [article]

Wei Zhang, Suyog Gupta, Xiangru Lian, Ji Liu
2016 arXiv   pre-print
Stochastic Gradient Descent (SGD) is the preferred optimization algorithm for training these networks and asynchronous SGD (ASGD) has been widely adopted for accelerating the training of large-scale deep  ...  In this paper, we propose a variant of the ASGD algorithm in which the learning rate is modulated according to the gradient staleness and provide theoretical guarantees for convergence of this algorithm  ...  Clearly, the n-softsync protocol provides an effective mechanism for controlling the staleness of the gradients in the system.  ... 
arXiv:1511.05950v5 fatcat:q6bc342ysbdetdgwtnzni3kjtu

Communication, convergence, and stochastic stability in self-assembly

Michael J. Fox, Jeff S. Shamma
2010 49th IEEE Conference on Decision and Control (CDC)  
We also suggest how the presented process can be augmented with communications to provide stability. 49th IEEE Conference on Decision and Control  ...  We propose a stochastic decision policy for the agents that provides a performance guarantee in the form of stochastic stability for any finite number of agents and any acyclic target graph.  ...  In other words, the stochastically stable states of P ǫ are all of the absorbing, or fully built states with a specific number of assemblies.  ... 
doi:10.1109/cdc.2010.5717190 dblp:conf/cdc/FoxS10 fatcat:cvr7ot6qj5c25bvjheninmmk3a

Game of Threads

Jose Rodrigo Sanchez Vicarte, Benjamin Schreiber, Riccardo Paccagnella, Christopher W. Fletcher
2020 Proceedings of the Twenty-Fifth International Conference on Architectural Support for Programming Languages and Operating Systems  
More importantly, we view a large part of our contribution to be showing how thread scheduling can impact a highly stochastic algorithm like A-SGD in a controllable way.  ...  ,B−1 ▽loss(θ, mb i .x, mb i .y) θ = θ − α * д /* update parameters */ end end Algorithm 1: (Minibatch) stochastic gradient descent. For simplicity, assume B divides |T |.  ...  As the attack proceeds, the validation accuracy will change in observable ways.  ... 
doi:10.1145/3373376.3378462 dblp:conf/asplos/VicarteSPF20 fatcat:v7nveklj2bh5le7pie7yvs3zuq

Probabilistic Robot Navigation in Partially Observable Environments

Reid G. Simmons, Sven Koenig
1995 International Joint Conference on Artificial Intelligence  
with approximate metric information We demon stcate Itw robustness of this appiorch «\ controlling an actuaJ indoor mobile robot navigating corridors 1 10B0 LEARNING  ...  Autonomous mobile robots need very reliable navigation capabilities in order to operate unattended for long periods of time This paper reports on first results of a research program that uses par tially observable  ...  stochastic domains In Proceedings of the IJCAI 1995 [Simmons 1994] R Simmons Becoming increasingly reliable In Proceedings of the International Conference on Artificial Intelli genc e Planning Systems  ... 
dblp:conf/ijcai/SimmonsK95 fatcat:zbsq54g7nbdmljfbz4e6ewlqzm

ASYNC: A Cloud Engine with Asynchrony and History for Distributed Machine Learning [article]

Saeed Soori, Bugra Can, Mert Gurbuzbalaba, Maryam Mehri Dehnavi
2020 arXiv   pre-print
With introducing three main modules and bookkeeping system-specific and application parameters, ASYNC provides practitioners with a framework to implement asynchronous machine learning methods.  ...  To demonstrate ease-of-implementation in ASYNC, the synchronous and asynchronous variants of two well-known optimization methods, stochastic gradient descent and SAGA, are demonstrated in ASYNC.  ...  However, many CCMs take information from the current state of the system as input and couple this information with the barrier control strategy to dynamically build the computation graph.  ... 
arXiv:1907.08526v4 fatcat:uhwcdnl67bcczdncxgrv6hod7a

Dynamic Stale Synchronous Parallel Distributed Training for Deep Learning [article]

Xing Zhao and Aijun An and Junfeng Liu and Bao Xin Chen
2019 arXiv   pre-print
paradigm by dynamically determining the staleness threshold at the run time.  ...  In this paper, we present a distributed paradigm on the parameter server framework called Dynamic Stale Synchronous Parallel (DSSP) which improves the state-of-the-art Stale Synchronous Parallel (SSP)  ...  To answer the first question, we observe the difference between the two types of DNNs (with or without fully connected layers): x A fully connected layer requires more parameters than a convolutional layer  ... 
arXiv:1908.11848v1 fatcat:ta3pop7phjcb5esipaxf754pfq

Susceptibility of Age of Gossip to Timestomping [article]

Priyanka Kaswan, Sennur Ulukus
2022 arXiv   pre-print
These show that fully connected nature of a network can be both a benefit and a detriment for information freshness; full connectivity, while enabling fast dissemination of information, also enables fast  ...  We consider a fully connected network consisting of a source that maintains the current version of a file, n nodes that use asynchronous gossip mechanisms to disseminate fresh information in the network  ...  We studied the effects of timestomping attacks on the age of gossip in a large fully connected network.  ... 
arXiv:2205.08510v1 fatcat:gxyy7p2dhvgepkotaqamzjgnnm

Energy Minimization for Federated Asynchronous Learning on Battery-Powered Mobile Devices via Application Co-running [article]

Cong Wang, Bin Hu, Hongyi Wu
2022 arXiv   pre-print
From a series of experiments, we find that co-running the training process in the background with foreground applications gives the system a deep energy discount with negligible performance slowdown.  ...  Then we propose an online algorithm using the Lyapunov framework to explore the solution space via the energy-staleness trade-off.  ...  Estimate g i (t, t + τ i ) with Eq. ( 4 ). 6 α i (t) ← arg min Pi,bi,gi V P i (t) − Q(t)b i (t) + H(t)g i (t, t + τ i ). 7 Inform control decision α i (t) to server. 8 Server: Update Q(t), H(t) according  ... 
arXiv:2204.13878v1 fatcat:q2ocgx6htbavtiwj5rpmgm2dde


Dean Učkar, Juraj Dobrila University of Pula, Danijel Petrović
2022 Ekonomska Misao i Praksa  
Finally, the average efficiency of insurance companies improved in the observed period, while the gap between large, medium and small insurers keeps widening.  ...  Due to financial consolidation that overtook the Croatian financial market, the number of insurance companies dicreased from 24 in 2015, at the start of the observed period, to 15 in 2020.  ...  Croatia osiguranje d.d. is fully efficient in the observed period.  ... 
doi:10.17818/emip/2022/1.3 fatcat:wtwlmyr7gza3pd4bdp4zxnphoi

Asynchronous Complex Analytics in a Distributed Dataflow Architecture [article]

Joseph E. Gonzalez, Peter Bailis, Michael I. Jordan, Michael J. Franklin, Joseph M. Hellerstein, Ali Ghodsi, Ion Stoica
2015 arXiv   pre-print
information passing.  ...  Scalable distributed dataflow systems have recently experienced widespread adoption, with commodity dataflow engines such as Hadoop and Spark, and even commodity SQL engines routinely supporting increasingly  ...  Acknowledgments This research was supported in part by NSF CISE Expeditions Award CCF-1139158, LBNL Award 7076018, DARPA XData Award FA8750-12-2-0331, the NSF Graduate Research Fellowship (grant DGE-1106400  ... 
arXiv:1510.07092v1 fatcat:32qafevpyjdfffjl64fvpkwnyi
« Previous Showing results 1 — 15 out of 1,211 results