Filters








116,511 Hits in 4.1 sec

Explaining the Learning Dynamics of Direct Feedback Alignment

Justin Gilmer, Colin Raffel, Samuel S. Schoenholz, Maithra Raghu, Jascha Sohl-Dickstein
2017 International Conference on Learning Representations  
Two recently developed methods, Feedback Alignment (FA) and Direct Feedback Alignment (DFA), have been shown to obtain surprising performance on vision tasks by replacing the traditional backpropagation  ...  With this new connection between DFA and LAFS we are able to explain why the "alignment" happens in DFA. * Work done as a member of the Google Brain Residency Program (g.co/brainresidency)  ...  Motivated by feedback-alignment, a variant called direct-feedback-alignment (DFA) was proposed in (Nøkland, 2016) .  ... 
dblp:conf/iclr/GilmerRSRS17 fatcat:bi4cyrovw5he3aw27bees2asee

Neural manifold under plasticity in a goal driven learning behaviour [article]

Barbara Feulner, Claudia Clopath
2020 bioRxiv   pre-print
Here, we show in a computational model that modification of recurrent weights, driven by a learned feedback signal, can account for the observed behavioural difference between within- and outside-manifold  ...  Supporting this idea, recent experiments have shown that monkeys can learn to adapt their neural activity in motor cortex on a timescale of minutes, given that the change lies within the original low-dimensional  ...  Oby for discussions and comments on the manuscript. This work was supported by BBSRC BB/N013956/1, BB/N019008/1, Wellcome Trust 200790/Z/16/Z, Simons Foundation 564408 and EPSRC EP/R035806/1.  ... 
doi:10.1101/2020.02.21.959163 fatcat:hnu27zptcnbivhnuxtcsqrb6he

Credit Assignment in Neural Networks through Deep Feedback Control [article]

Alexander Meulemans, Matilde Tristany Farinha, Javier García Ordóñez, Pau Vilimelis Aceituno, João Sacramento, Benjamin F. Grewe
2022 arXiv   pre-print
The resulting learning rule is fully local in space and time and approximates Gauss-Newton optimization for a wide range of feedback connectivity patterns.  ...  The success of deep learning sparked interest in whether the brain learns by using similar techniques for assigning credit to each synaptic weight for its contribution to the network output.  ...  Acknowledgments and Disclosure of Funding  ... 
arXiv:2106.07887v3 fatcat:sacn6ls4qfezlctxup6mhh664e

Motivation, Challenges, Support (MCS) Cycle Model for the Development of PBL Tutors

Costas S. Constantinou, Stella A. Nicolaou
2018 Qualitative Research in Education  
Problem-Based Learning (PBL) is well known for enhancing students' problem-solving skills and teamwork, while the role of PBL tutors is to facilitate discussion rather than teach.  ...  Based on this, we propose the "Motivation, Challenges, Support (MCS) Cycle Model" for the development of PBL tutors.  ...  Some tutors explained that personal issues outside of the medical course could potentially influence group dynamics.  ... 
doi:10.17583/qre.2018.3064 fatcat:t57wvdpftrhyxmjf5hcexqypgu

Simultaneous coordinate representations are influenced by visual feedback in a motor learning task

P. N. Parmar, F. C. Huang, J. L. Patton
2011 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society  
These findings suggest that the type of visual feedback biases the way subjects perform, and that learning results can be significantly influenced by feedback.  ...  We explored how visual feedback influenced the coordinate frame in which the CNS stores and recalls these memories of learned skills in a reaching-generalization task.  ...  Generalization has been studied quite extensively for deducing coordinate reference frames of neuromotor learning, whether it is new hand positions [1, 2] , tasks [16] , movement directions [3] , dynamics  ... 
doi:10.1109/iembs.2011.6091668 pmid:22255891 pmcid:PMC8734938 dblp:conf/embc/ParmarHP11 fatcat:mtue3tvzrzbdzalvgp7jjaln4m

Direct Feedback Alignment Provides Learning in Deep Neural Networks [article]

Arild Nøkland
2016 arXiv   pre-print
In this work, the feedback alignment principle is used for training hidden layers more independently from the rest of the network, and from a zero initial condition.  ...  In fact, random feedback weights work evenly well, because the network learns how to make the feedback useful.  ...  The mechanism explains how an asymmetric feedback path can provide learning by aligning the back-propagated and forward propagated gradients with it's own, under the assumption of constant update directions  ... 
arXiv:1609.01596v5 fatcat:xi4c6dsj7zhbdjy7raw3xy5gpq

Neural manifold under plasticity in a goal driven learning behaviour

Barbara Feulner, Claudia Clopath
2021 PLoS Computational Biology  
Here, we show in a computational model that modification of recurrent weights, driven by a learned feedback signal, can account for the observed behavioural difference between within- and outside-manifold  ...  Supporting this idea, recent experiments have shown that monkeys can learn to adapt their neural activity in motor cortex on a timescale of minutes, given that the change lies within the original low-dimensional  ...  (G) Alignment of internal manifold and readout determines feedback learning performance.  ... 
doi:10.1371/journal.pcbi.1008621 pmid:33544700 pmcid:PMC7864452 fatcat:dp2ng3gtirekldrfdhbphhop3i

Forward Signal Propagation Learning [article]

Adam Kohan, Edward A. Rietman, Hava T. Siegelmann
2022 arXiv   pre-print
To further explain the behavior of sigprop, we provide evidence that sigprop provides useful learning signals in context to backpropagation.  ...  In biology, this explains how neurons without feedback connections can still receive a global learning signal.  ...  Comparison of Learning Algorithms Relaxing Learning Constraints Under Backpropagation. a) the backpropagation algorithm b,c) the feedback alignment and direct feedback alignment algorithms.  ... 
arXiv:2204.01723v1 fatcat:r2mz75rgqraqre5ntkvs773ugu

Evidence of multiple coordinate representations during generalization of motor learning

Pritesh N. Parmar, Felix C. Huang, James L. Patton
2014 Experimental Brain Research  
We also investigated how feedback might affect the results and found in several cases that performance was better for visual displays that were aligned with the limb (in first person) versus non-aligned  ...  Several studies have suggested that the motor system takes advantage of a coordinate system when learning a novel sensorimotor environment.  ...  We thank Sandro Mussa-Ivaldi, Max Berniker, and the rest of the Robotics Lab at the Rehabilitation Institute of Chicago for support along the long path to publishing this paper.  ... 
doi:10.1007/s00221-014-4034-6 pmid:25248844 pmcid:PMC4289976 fatcat:63eaei7b4zbuhc7r5buf2gntyy

Random feedback weights support learning in deep neural networks [article]

Timothy P. Lillicrap, Daniel Cownden, Douglas B. Tweed, Colin J. Akerman
2014 arXiv   pre-print
We show that a network can learn to extract useful information from signals sent through these random feedback connections. In essence, the network learns to learn.  ...  The brain processes information through many layers of neurons.  ...  Proof #3: B acts like the pseudoinverse of W Here we will prove that, under fairly restrictive conditions, feedback alignment prescribes hidden unit updates which are in the same direction as those prescriped  ... 
arXiv:1411.0247v1 fatcat:giepvxpzrrhgxd2xdddsv5yw24

Action Learning in a Graduate Online Course

Brent Goertzen, Brittney Squire
2019 Journal of Leadership Education  
This application brief explains action learning and how the authors applied action learning principles in an online graduate course on leadership development.  ...  Practitioner and student reflections are offered as well as an analysis of student learning. Recommendations from these experiences offer ways to refine this pedagogical practice.  ...  The third component requires students to explain specific intervention strategies in alignment with the stated organizational needs and intended learning objectives.  ... 
doi:10.12806/v18/i2/a1 fatcat:mcvaofi325dmxia2b7mbrzlrfa

Principles of Organizational Co-Evolution of Business and IT: a Complexity Perspective

Mengmeng Zhang, Honghui Chen, Kalle Lyytinen
2019 European Conference on Information Systems  
The paper also shows the capability of a system dynamic method to capture some of the holistic behaviors that emerge from implementing the 10 principles.  ...  Second, we build a system dynamics model based on the 10 principles to apply these principles as a means to improve the BITC management.  ...  By focusing on continuous learning and adaptations between multiple system components (Ciborra, 2000) , the co-evolutionary theory seeks to explain the multi-level, multi-directional, non-linear, and  ... 
dblp:conf/ecis/ZhangCL19 fatcat:mll44icornearepkkjfkfarcp4

Single-phase deep learning in cortico-cortical networks [article]

Will Greedy, Heng Wei Zhu, Joseph Pemberton, Jack Mellor, Rui Ponte Costa
2022 arXiv   pre-print
However, these models are either unable to effectively backpropagate error signals across multiple layers or require a multi-phase learning process, neither of which are reminiscent of learning in the  ...  Finally, we demonstrate that our model is capable of learning complex image classification tasks (MNIST and CIFAR-10).  ...  We would also like to thank Callum Wright and the rest of the High Performance Computing team at the University of Bristol for constant and quick help with BP.  ... 
arXiv:2206.11769v1 fatcat:uaq6a44xtfev3dyoe43vbvhlhq

RE-CONCEPTUALIZING L2 WRITING REVISION STRATEGIES FROM AN ECOLOGICAL PERSPECTIVE: AN INTERVIEW BASED INQUIRY

Masakazu Mishima
2019 English Review: Journal of English Education  
attempt to establish sociocognitive alignment with the instructor�s expectations delivered through feedback.  ...  The study aimed to explore how sociomaterial conditions present in the pedagogical context lead to the writer�s use of revision strategies in responding to instructor feedback.  ...  Thereby, it offers an extended interpretation of the results in an attempt to explain how one L2 writer's socio-materially afforded revision strategies emerged as a result of dynamic interplay of his agency  ... 
doi:10.25134/erjee.v7i2.1771 fatcat:y22hx3lncrbsvkuampbf3p57r4

FROM SKILL-ACQUISITION TO DYNAMIC LEARNING: LEARNING-ORIENTED ASSESSMENT IN THE SIGHT TRANSLATION CLASSROOM

Xueni Zhang, Wan Hu, Maike Oergel
2021 Current Trends in Translation Teaching and Learning E  
Assessment, previously associated with the evaluation of learning outcomes, has recently been recognised as an integral part of the learning process.  ...  In order to evaluate the feasibility and accountability of the model, the three tenets of LOA were aligned with module design and discussed alongside students' perceptions.  ...  From skillacquisition to dynamic learning: Learning-oriented assessment in the sight translation classroom. Current Trends in Translation Teaching and Learning E.  ... 
doi:10.51287/cttle202110 fatcat:2xgzbtk5bvcpldleeu3mkbbija
« Previous Showing results 1 — 15 out of 116,511 results