Filters








375,235 Hits in 6.5 sec

Learning values across many orders of magnitude [article]

Hado van Hasselt and Arthur Guez and Matteo Hessel and Volodymyr Mnih and David Silver
2016 arXiv   pre-print
This is useful in value-based reinforcement learning, where the magnitude of appropriate value approximations can change over time when we update the policy of behavior.  ...  This clipping facilitates learning across many different games with a single learning algorithm, but a clipped reward function can result in qualitatively different behavior.  ...  Across games, 95% of median norms range over less than two orders of magnitude (roughly between 1 and 20), compared to almost four orders of magnitude for clipped Double DQN, and more than six orders of  ... 
arXiv:1602.07714v2 fatcat:62fejsgkijfolklhrar32vbos4

The discovery and comparison of symbolic magnitudes

Dawn Chen, Hongjing Lu, Keith J. Holyoak
2014 Cognitive Psychology  
BARTlet operates on distributions of magnitude variables created by applying dimension-specific weights (learned with the aid of empirical priors derived from pre-categorical comparisons) to more primitive  ...  Although numerous models of magnitude comparison have been proposed, the basic question of how symbolic magnitudes (e.g., size or intelligence of animals) are derived and represented in memory has received  ...  MATLAB code for the simulations reported here is available from the Web site of the UCLA Computational Vision and Learning Lab (http://cvl.psych.ucla.edu).  ... 
doi:10.1016/j.cogpsych.2014.01.002 pmid:24531498 fatcat:twactkcmsrehlhci6ynmdyf3uq

How fine can fine-tuning be? Learning efficient language models [article]

Evani Radiya-Dixit, Xin Wang
2020 arXiv   pre-print
Given a language model pre-trained on massive unlabeled text corpora, only very light supervised fine-tuning is needed to learn a task: the number of fine-tuning steps is typically five orders of magnitude  ...  Further, we find that there are surprisingly many good solutions in the set of sparsified versions of the pre-trained model.  ...  Our intuition comes from the observation that the amount of fine-tuning necessary to learn each task is very small (five orders of magnitude smaller than the dimensionality of the parameter space, Table  ... 
arXiv:2004.14129v1 fatcat:qdp5iu4nbraghgpiu5yclbcici

Investigating Numeracy Learning Ability of a Text-to-Text Transfer Model [article]

Kuntal Kumar Pal, Chitta Baral
2021 arXiv   pre-print
We consider four numeracy tasks: numeration, magnitude order prediction, finding minimum and maximum in a series, and sorting.  ...  Here we investigate the ability of text-to-text transfer learning model (T5), which has outperformed its predecessors in the conventional NLP tasks, to learn numeracy.  ...  For Magnitude Order Prediction task we use publicly available dataset, Numeracy600K. We synthetically create the data for rest of the tasks.  ... 
arXiv:2109.04672v1 fatcat:drfrnnj2nnb7neph6uikxnxruq

How Data Drive Early Word Learning: A Cross-Linguistic Waiting Time Analysis

Francis Mollica, Steven T. Piantadosi
2017 Open Mind  
With high statistical certainty, words require on the order of ∼ 10 learning instances, which occur on average once every two months.  ...  information across multiple situations.  ...  ACKNOWLEDGMENTS The authors thank Dick Aslin, Elika Bergelson, Celeste Kidd, and anonymous reviewers for comments on early drafts of this article.  ... 
doi:10.1162/opmi_a_00006 fatcat:4muqqke2nrcu5fosn3adeyutoi

The Development of Arabic Digit Knowledge in 4- to 7-Year-Old Children

Birgit Knudsen, Martin H. Fischer, Anne Henning, Gisa Aschersleben
2015 Journal of Numerical Cognition  
Performance across tasks revealed a clear developmental trajectory: children traverse from first knowing the cardinal values of number words to recognizing Arabic digits to knowing their cardinal values  ...  We document the developmental trajectory of 4- to 7-year-olds' proficiency in accessing magnitude information from Arabic digits in five tasks differing in magnitude manipulation requirements.  ...  We also would like to thank Marion Klein, Alexander Kirmβe and all students of the bachelor course 'Empirisches Praktikum 2012/2013' for their engagement and their contribution to the implementation of  ... 
doi:10.5964/jnc.v1i1.4 fatcat:azvzx6cfu5bplpxgk7bunqxana

Graph Frequency Analysis of Brain Signals

Weiyu Huang, Leah Goldsberry, Nicholas F. Wymbs, Scott T. Grafton, Danielle S. Bassett, Alejandro Ribeiro
2016 IEEE Journal on Selected Topics in Signal Processing  
brain graph frequencies associated with different levels of spatial smoothness across the brain regions.  ...  We observe that brain signals corresponding to different graph frequencies exhibit different levels of adaptability throughout learning.  ...  the learning rates across subjects.  ... 
doi:10.1109/jstsp.2016.2600859 pmid:28439325 pmcid:PMC5400112 fatcat:beibmmqjlfbutkflfpe5bjcb74

Effects of fictive reward on rat's choice behavior

Ko-Un Kim, Namjung Huh, Yunsil Jang, Daeyeol Lee, Min Whan Jung
2015 Scientific Reports  
, consistent with incremental learning of action values.  ...  the incremental value learning system.  ...  than actual reward in incremental value learning.  ... 
doi:10.1038/srep08040 pmid:25623929 pmcid:PMC4894400 fatcat:sneeghthsjcvvmycflnejkyk5e

RandAugment: Practical automated data augmentation with a reduced search space [article]

Ekin D. Cubuk and Barret Zoph and Jonathon Shlens and Quoc V. Le
2019 arXiv   pre-print
Recent work has shown that data augmentation has the potential to significantly improve the generalization of deep learning models.  ...  While these strategies were optimized for improving validation accuracy, they also led to state-of-the-art results in semi-supervised learning and improved robustness to common corruptions of images.  ...  Acknowledgements We thank Samy Bengio, Daniel Ho, Ildoo Kim, Jaehoon Lee, Zhaoqi Leng, Hanxiao Liu, Raphael Gontijo Lopes, Ruoming Pang, Ben Poole, Mingxing Tan, and the rest of the Brain team for their  ... 
arXiv:1909.13719v2 fatcat:ihay5kvgcvbzjge2eq3wtwr6dy

Machine learning, waveform preprocessing and feature extraction methods for classification of acoustic startle waveforms

Timothy J. Fawcett, Chad S. Cooper, Ryan J. Longenecker, Joseph P. Walton
2020 MethodsX  
Machine learning models utilizing methods from different families of algorithms were individually trained and then ensembled together, resulting in an extremely robust model.  ...  •ASR waveforms were normalized using the mean and standard deviation computed before the startle elicitor was presented•9 machine learning algorithms from 4 different families of algorithms were individually  ...  The authors would like to acknowledge the use of the services provided by Research Computing at the University of South Florida .  ... 
doi:10.1016/j.mex.2020.101166 pmid:33354518 pmcid:PMC7744771 fatcat:nyco4zhidbhktp557mnzvhqqsy

Reward, Value, and Salience [chapter]

T. Kahnt, P.N. Tobler
2017 Decision Neuroscience  
Value increases with the magnitude and probability of reward but decreases with the magnitude and probability of punishment, whereas salience increases with the magnitude and probability of both reward  ...  Value increases with the magnitude and probability of reward but decreases with the magnitude and probability of punishment, whereas salience increases with the magnitude and probability of both reward  ...  Moreover, in many cases, salience is confounded by value, magnitude, or probability [80] .  ... 
doi:10.1016/b978-0-12-805308-9.00009-9 fatcat:66hpdcggivg3hlrex4t74t26de

1 < 2 and 2 < 3: Non-Linguistic Appreciations of Numerical Order

Ursula S. Anderson, Sara Cordes
2013 Frontiers in Psychology  
Further, we suggest that patterns in the way that infants and non-human animals process numerical order can be accounted for by changes across development, the conditions under which representations are  ...  order.  ...  In sum, evidence strongly suggests that responding across these tasks was not the result of arbitrary sequence learning, but a function of the numerical values presented.  ... 
doi:10.3389/fpsyg.2013.00005 pmid:23355830 pmcid:PMC3554834 fatcat:26o6d46iljak5jq7iuqhkrnham

On the Adequacy of Untuned Warmup for Adaptive Optimization

Jerry Ma, Denis Yarats
2021 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
In this work, we refute this analysis and provide an alternative explanation for the necessity of warmup based on the magnitude of the update term, which is of greater relevance to training stability.  ...  Adaptive optimization algorithms such as Adam (Kingma and Ba, 2014) are widely used in deep learning. The stability of such algorithms is often improved with a warmup schedule for the learning rate.  ...  Stochastic gradient descent and its various first-order cousins (Polyak 1964; Nesterov 1983 ) have enabled numerous advances in deep learning across domains (Krizhevsky, Sutskever, and Hinton 2012; He  ... 
doi:10.1609/aaai.v35i10.17069 fatcat:uztrbydudbhzbmxvltsudvpcce

Simplified learning in complex situations: Knowledge partitioning in function learning

Stephan Lewandowsky, Michael Kalish, S. K. Ngang
2002 Journal of experimental psychology. General  
Because context did not predict function values, it is suggested that people use context to gate separate learning of simpler partial functions.  ...  In 4 experiments, using a function learning paradigm, a binary context variable was paired with the continuous stimulus variable of a to-be-learned function.  ...  (b) People gate access to parcels of knowledge on the basis of context, (c) regardless of whether context normatively predicts response magnitudes.  ... 
doi:10.1037/0096-3445.131.2.163 fatcat:57z44v4cine2pfd6wzyddlzura

Simplified learning in complex situations: Knowledge partitioning in function learning

Stephan Lewandowsky, Michael Kalish, S. K. Ngang
2002 Journal of experimental psychology. General  
Because context did not predict function values, it is suggested that people use context to gate separate learning of simpler partial functions.  ...  In 4 experiments, using a function learning paradigm, a binary context variable was paired with the continuous stimulus variable of a to-be-learned function.  ...  (b) People gate access to parcels of knowledge on the basis of context, (c) regardless of whether context normatively predicts response magnitudes.  ... 
doi:10.1037//0096-3445.131.2.163 pmid:12049238 fatcat:6akdaxj6svcgnddbp6xp2g4pqm
« Previous Showing results 1 — 15 out of 375,235 results