Filters








33,506 Hits in 3.7 sec

Correcting Momentum in Temporal Difference Learning [article]

Emmanuel Bengio, Joelle Pineau, Doina Precup
2021 arXiv   pre-print
We argue that, unlike in supervised learning, momentum in Temporal Difference (TD) learning accumulates gradients that become doubly stale: not only does the gradient of the loss change due to parameter  ...  A common optimization tool used in deep reinforcement learning is momentum, which consists in accumulating and discounting past gradients, reapplying them at each iteration.  ...  TEMPORAL DIFFERENCE LEARNING We now test our hypotheses, that there is optimisation bias and that we can correct it, on RL problems.  ... 
arXiv:2106.03955v1 fatcat:u2isltjazjainpnqapbv4maxva

Controlled hierarchical filtering: Model of neocortical sensory processing [article]

Andras Lorincz
2003 arXiv   pre-print
Implicit memory phenomena; priming and prototype learning are emerging features of the model. Mathematical theorems ensure stability and attractive learning properties of the architecture.  ...  Connections to reinforcement learning are also established: both the control network, and the network with a hidden model converge to (near) optimal policy under suitable conditions.  ...  (In continuous time, adding up 'corrections' is equivalent to the temporal integration of the error.)  ... 
arXiv:cs/0308025v1 fatcat:zzy5t5qas5hxnkmkbswlt7s3ui

Weight Changes for Learning Mechanisms in Two-Term Back-Propagation Network [chapter]

Siti Mariyam, Ashraf Osman, Citra Ramadhe
2013 Artificial Neural Networks - Architectures and Applications  
In [ ] gave a brief definition of online learning and the difference with batch learning.  ...  Different ranges of effect values correspond to different learning models.  ...  In addition, this study also studies the weight change sign with respect to the temporal behaviour of gradient to study the learning behaviour of the network and also to measure the performance of the  ... 
doi:10.5772/51776 fatcat:nresm2cs6bhqrhsblj3fznfgiq

Momentum-like effects and the dynamics of perception, cognition, and action

Timothy L. Hubbard
2019 Attention, Perception & Psychophysics  
In a momentum-like effect, the likely future state of a current action or process is extrapolated.  ...  " and "extrapolation," perceptual inference of subjective or objective consequences, importance of time scale and temporal information, importance of the computational theory level, momentum-like effects  ...  , the temporal relationship between a response and a reinforcer is critical in learning).  ... 
doi:10.3758/s13414-019-01770-z pmid:31140136 fatcat:duepj7fegvhujlsqszjks5ului

SeCo: Exploring Sequence Supervision for Unsupervised Representation Learning [article]

Ting Yao and Yiheng Zhang and Zhaofan Qiu and Yingwei Pan and Tao Mei
2021 arXiv   pre-print
We materialize the supervisory signals through determining whether a pair of samples is from one frame or from one video, and whether a triplet of samples is in the correct temporal order.  ...  A steady momentum of innovations and breakthroughs has convincingly pushed the limits of unsupervised image representation learning. Compared to static 2D images, video has one more dimension (time).  ...  For sequential supervision, we develop a task of temporal order validation (Figure 1(c) ) and verify whether a series of frame patches are in the correct temporal order.  ... 
arXiv:2008.00975v2 fatcat:eb3hnhbuqfhjvhquax6oro4kei

Temporal Calibrated Regularization for Robust Noisy Label Learning [article]

Dongxian Wu, Yisen Wang, Zhuobin Zheng, Shu-tao Xia
2020 arXiv   pre-print
In this paper, we propose a Temporal Calibrated Regularization (TCR), in which we utilize the original labels and the predictions in the previous epoch together to make DNN inherit the simple pattern it  ...  has learned with little overhead.  ...  According to (8) , the resistant gradient is related to the difference in temporal dimension.  ... 
arXiv:2007.00240v1 fatcat:xvg7lyty6jfepkwb3je4yyisay

VideoMoCo: Contrastive Video Representation Learning with Temporally Adversarial Examples [article]

Tian Pan, Yibing Song, Tianyu Yang, Wenhao Jiang, Wei Liu
2021 arXiv   pre-print
By adaptively dropping out different frames during training iterations of adversarial learning, we augment this input sample to train a temporally robust encoder.  ...  MoCo is effective for unsupervised image representation learning. In this paper, we propose VideoMoCo for unsupervised video representation learning.  ...  The temporally robust feature representations empower VideoMoCo to make correct predictions. 3.4.  ... 
arXiv:2103.05905v2 fatcat:sf5xbs6n7zgubgdku2yi3j7oae

Video Contrastive Learning with Global Context [article]

Haofei Kuang, Yi Zhu, Zhi Zhang, Xinyu Li, Joseph Tighe, Sören Schwertfeger, Cyrill Stachniss, Mu Li
2021 arXiv   pre-print
Our formulation is able to capture global context in a video, thus robust to temporal content change.  ...  In this paper, we propose a new video-level contrastive learning method based on segments to formulate positive pairs.  ...  This will lead to a balanced 4-way classification problem: both tuples are in correct temporal order, t a correct and t + shuffled, t a shuffled and t + correct, and both shuffled.  ... 
arXiv:2108.02722v1 fatcat:k4kzxjzuwjaoxo6iw52536oipm

Temporal Ensembling for Semi-Supervised Learning [article]

Samuli Laine, Timo Aila
2017 arXiv   pre-print
We introduce self-ensembling, where we form a consensus prediction of the unknown labels using the outputs of the network-in-training on different epochs, and most importantly, under different regularization  ...  Using our method, we set new records for two standard semi-supervised learning benchmarks, reducing the (non-augmented) classification error rate from 18.44% to 7.05% in SVHN with 500 labels and from 18.63%  ...  In this test the difference between Π-model and temporal ensembling was quite significant at 1.5 percentage points.  ... 
arXiv:1610.02242v3 fatcat:x4urhkedibd7za6yqlp2v7hj2a

Vocal gymnastics and the bird brain

Franz Goller
1998 Nature  
Such a level of sophistication may not be necessary, however, if birds can achieve the correct acoustic effect by learning pressure adjustment through trial and error.  ...  But we now learn that intrinsic mechanical properties of the syrinx can contribute to temporal and acoustic song patterns.  ... 
doi:10.1038/25589 fatcat:gia2m3z7pfaupdgjr5pavtul3e

Adaptive scene-based nonuniformity correction method for infrared-focal plane arrays

Sergio N. Torres, Esteban M. Vera, Rodrigo A. Reeves, Sergio K. Sobarzo, Gerald C. Holst
2003 Infrared Imaging Systems: Design, Analysis, Modeling, and Testing XIV  
: momentum, regularization, and adaptive learning rate.  ...  In this paper we present an enhanced adaptive scene-based non-uniformity correction (NUC) technique.  ...  learning rate plus momentum and regularization.  ... 
doi:10.1117/12.487217 fatcat:izlktjs5hbbajer5yugjbzmzj4

Quicksilver: Fast Predictive Image Registration - a Deep Learning Approach [article]

Xiao Yang, Roland Kwitt, Martin Styner, Marc Niethammer
2017 arXiv   pre-print
Finally, we introduce a new correction network which greatly increases the prediction accuracy of an already existing prediction network.  ...  We also provide a probabilistic version of our prediction network which can be sampled during the testing time to calculate uncertainties in the predicted deformations.  ...  The correction network learns to predict the difference between the ground truth momentum and the predicted momentum from the prediction network.  ... 
arXiv:1703.10908v4 fatcat:cp7qz2gupzf7dgblp4qhywxav4

Hierarchically Decoupled Spatial-Temporal Contrast for Self-supervised Video Representation Learning [article]

Zehua Zhang, David Crandall
2021 arXiv   pre-print
Motivated by their effectiveness in supervised learning, we first introduce spatial-temporal feature learning decoupling and hierarchical learning to the context of unsupervised video learning.  ...  We show by experiments that augmentations can be manipulated as regularization to guide the network to learn desired semantics in contrastive learning, and we propose a way for the model to separately  ...  , these methods differ in how they obtain variant embeddings of the same instance, e.g., using augmentations [3, 25, 73, 78] , future representations [46] , or momentum features [19] .  ... 
arXiv:2011.11261v2 fatcat:aztia46gm5bzzln4i5sjfm4why

Actor-Critic Reinforcement Learning with Neural Networks in Continuous Games

Gabriel Leuenberger, Marco A. Wiering
2018 Proceedings of the 10th International Conference on Agents and Artificial Intelligence  
Actor-Critic Reinforcement Learning with Neural Networks in Continuous Games.  ...  It relies on a third multilayer perceptron to estimate the absolute error of the critic which is used to correct the learning rule of the Actor.  ...  Like other temporal difference learning algorithms CACLA uses r t+1 + γ V t (s t+1 ) in every time step as a target to be approximated by the Critic.  ... 
doi:10.5220/0006556500530060 dblp:conf/icaart/LeuenbergerW18 fatcat:pz4czbphpvbndlhzdzd4ra3s2m

Quicksilver: Fast predictive image registration – A deep learning approach

Xiao Yang, Roland Kwitt, Martin Styner, Marc Niethammer
2017 NeuroImage  
The difference is used as a correction to the predicted momentum to increase prediction accuracy.  ...  Given a trained prediction network, the correction network predicts the difference between the ground truth momentum and the predicted result.  ...  The correction network learns to predict the difference between the ground truth momentum and the predicted momentum from the prediction network.  ... 
doi:10.1016/j.neuroimage.2017.07.008 pmid:28705497 pmcid:PMC6036629 fatcat:hxcewqqseve7xiui64gqhio7ou
« Previous Showing results 1 — 15 out of 33,506 results