Filters








697 Hits in 1.7 sec

Continual learning with hypernetworks [article]

Johannes von Oswald and Christian Henning and Benjamin F. Grewe and João Sacramento
2022 arXiv   pre-print
Continual learning (CL) is less difficult for this class of models thanks to a simple key feature: instead of recalling the input-output relations of all previously seen data, task-conditioned hypernetworks  ...  We provide insight into the structure of low-dimensional task embedding spaces (the input space of the hypernetwork) and show that task-conditioned hypernetworks demonstrate transfer learning.  ...  Continual learning with hypernetwork output regularization.  ... 
arXiv:1906.00695v4 fatcat:xtdourohoza4livlpio75iitka

Continual Model-Based Reinforcement Learning with Hypernetworks [article]

Yizhou Huang, Kevin Xie, Homanga Bharadhwaj, Florian Shkurti
2021 arXiv   pre-print
We argue that this is too slow for lifelong robot learning and propose HyperCRL, a method that continually learns the encountered dynamics in a sequence of tasks using task-conditional hypernetworks.  ...  the state transition experience; second, it uses fixed-capacity hypernetworks to represent non-stationary and task-aware dynamics; third, it outperforms existing continual learning alternatives that rely  ...  Algorithm 1: Continual Reinforcement Learning via Hypernetworks (HyperCRL) 1: Input: T tasks, each with its own dynamics S × A → S , reward r(s, a).  ... 
arXiv:2009.11997v2 fatcat:nwf3ximskrhytj3pbvxqz33xsa

Continual Learning in Recurrent Neural Networks [article]

Benjamin Ehret, Christian Henning, Maria R. Cervera, Alexander Meulemans, Johannes von Oswald, Benjamin F. Grewe
2021 arXiv   pre-print
While a diverse collection of continual learning (CL) methods has been proposed to prevent catastrophic forgetting, a thorough investigation of their effectiveness for processing sequential data with recurrent  ...  Our study shows that established CL methods can be successfully ported to the recurrent case, and that a recent regularization approach based on hypernetworks outperforms weight-importance methods, thus  ...  RELATED WORK Continual learning with sequential data. As in Parisi et al.  ... 
arXiv:2006.12109v3 fatcat:dy7xrnm2nvfr7iwre6hqihatpa

Continual learning with hypernetworks

Johannes Von Oswald, Christian Henning, João Sacramento, Benjamin F Grewe
2020
Continual learning (CL) is less difficult for this class of models thanks to a simple key feature: instead of recalling the input-output relations of all previously seen data, task-conditioned hypernetworks  ...  Continual learning (CL) isless difficult for this class of models thanks to a simple key feature: instead ofrecalling the input-output relations of all previously seen data, task-conditionedhypernetworks  ...  Nc) c (3) e (t) Θ trgt (t, 1) Θ trgt (t, 2) Θ trgt (t, 3) Θ trgt (t, Nc) f h f h f h f h Continual learning with hypernetwork output regularization.  ... 
doi:10.5167/uzh-200390 fatcat:nchjgzvcs5hxfgo6do7bxrblmq

Continual Learning from Demonstration of Robotic Skills [article]

Sayantan Auddy, Jakob Hollenstein, Matteo Saveriano, Antonio Rodríguez-Sánchez, Justus Piater
2022 arXiv   pre-print
Our results show that hypernetworks outperform other state-of-the-art regularization-based continual learning approaches for learning from demonstration.  ...  To this end, we propose an approach for continual learning from demonstration using hypernetworks and neural ordinary differential equation solvers.  ...  With this, we compute the continual learning metrics shown in Tab. III.  ... 
arXiv:2202.06843v2 fatcat:hblan6epvrax3atomvei3lx3uu

Hypernetworks: A Molecular Evolutionary Architecture for Cognitive Learning and Memory

Byoung-Tak Zhang
2008 IEEE Computational Intelligence Magazine  
Here we review the principles underlying human learning and memory, and identify three of them, i.e., continuity, glocality, and compositionality, as the most fundamental to human-level machine learning  ...  The chemically-based massive interaction for information organization and processing in the molecular hypernetworks, referred to as hyperinteractionism, is contrasted with the symbolist, connectionist,  ...  Cognitive Learning with Hypernetworks We are now in a better position to examine the properties of the hypernetwork model.  ... 
doi:10.1109/mci.2008.926615 fatcat:dr23kuxnsnb2hayxjhf4r3hxbe

Hypernetworks for Continual Semi-Supervised Learning [article]

Dhanajit Brahma, Vinay Kumar Verma, Piyush Rai
2021 arXiv   pre-print
Learning from data sequentially arriving, possibly in a non i.i.d. way, with changing task distribution over time is called continual learning.  ...  We consolidate the knowledge of sequential tasks in the hypernetwork, and the base network learns the semi-supervised learning task.  ...  MCSSL: Meta-Consolidation for Continual Semi-Supervised Learning This section starts with the problem set-up of Continual Semi-Supervised Learning.  ... 
arXiv:2110.01856v1 fatcat:ckic6sbibnbqzl5c6fbhm2cgia

Evolving hypernetwork models of binary time series for forecasting price movements on stock markets

Elena Bautu, Sun Kim, Andrei Bautu, Henri Luchian, Byoung-Tak Zhang
2009 2009 IEEE Congress on Evolutionary Computation  
In particular, the hypernetwork approach outperforms other machine learning methods such as support vector machines, naive Bayes, multilayer perceptrons, and k-nearest neighbors.  ...  Applied to the Dow Jones Industrial Average Index and the Korea Composite Stock Price Index data, the experimental results show that the proposed method effectively learns and predicts the time series  ...  The quality of the hypernetwork assessed with feedback connection provides good insight about the quality of the patterns learned by the hypernetwork from the dataset.  ... 
doi:10.1109/cec.2009.4982944 dblp:conf/cec/BautuKBLZ09 fatcat:ba2orail55f4vltqcnf34imqay

Stochastic Hyperparameter Optimization through Hypernetworks [article]

Jonathan Lorraine, David Duvenaud
2018 arXiv   pre-print
Machine learning models are often tuned by nesting optimization of model weights inside the optimization of hyperparameters.  ...  We show that our technique converges to locally optimal weights and hyperparameters for sufficiently large hypernetworks.  ...  Sufficiently powerful hypernetworks can learn continuous best-response functions, which minimizes the expected loss for all hyperparameter distributions with convex support.  ... 
arXiv:1802.09419v2 fatcat:sntuz2kwcnfpxhpxwibeeha3ri

Evolutionary hypernetworks for learning to generate music from examples

Hyun-Woo Kim, Byoung-Hee Kim, Byoung-Tak Zhang
2009 2009 IEEE International Conference on Fuzzy Systems  
Evolutionary hypernetworks (EHNs) are recently introduced models for learning higher-order probabilistic relations of data by an evolutionary self-organizing process.  ...  Short-term and long-term sequential patterns can be extracted and combined to generate music with various styles by our method.  ...  Music completion as validation With the trained hypernetwork and given fragment of melody as a cue, we generate music by continuously predicting notes.  ... 
doi:10.1109/fuzzy.2009.5277047 dblp:conf/fuzzIEEE/KimKZ09 fatcat:g3nli2kr6jcubmi4xy6hzq247q

Evolutionary concept learning from cartoon videos by multimodal hypernetworks

Beom-Jin Lee, Jung-Wo Ha, Kyung-Min Kim, Byoung-Tak Zhang
2013 2013 IEEE Congress on Evolutionary Computation  
Two key ideas on evolutionary concept learning are representing concepts in a large collection (population) of hyperedges or a hypergraph and to incrementally learning from video streams based on an evolutionary  ...  Previous researches on concept learning have focused on unimodal data, usually on linguistic domains in a static environment.  ...  Multimodal hypernetwork incrementally learns higher-order concept relations from the visual and texual sets with subsampling-based evolutionary method (c).  ... 
doi:10.1109/cec.2013.6557700 dblp:conf/cec/LeeHKZ13 fatcat:q4trwieypnctzebnojfde7bfcq

Hypernetwork functional image representation [article]

Sylwester Klocek, Łukasz Maziarka, Maciej Wołczyk, Jacek Tabor, Jakub Nowak, Marek Śmieja
2019 arXiv   pre-print
We use a hypernetwork to automatically generate continuous functional representation of images at test time without any additional training.  ...  Since obtained representation is continuous, we can easily inspect the image at various resolutions.  ...  Hypernetwork. Hypernetwork is a convolutional neural network with some modifications, see Figure 5 . We created an eight layered network with one residual connection.  ... 
arXiv:1902.10404v2 fatcat:to2tsv6mnzb7viamqenot3cn5e

HyperInvariances: Amortizing Invariance Learning [article]

Ruchika Chavhan, Henry Gouk, Jan Stühmer, Timothy Hospedales
2022 arXiv   pre-print
Providing invariances in a given learning task conveys a key inductive bias that can lead to sample-efficient learning and good generalisation, if correctly specified.  ...  However, invariance learning is expensive and data intensive for popular neural architectures. We introduce the notion of amortizing invariance learning.  ...  A train i : HyperInvariance train accuracy with continuous invariance, Ai * : HyperInvariance test accuracy with continuous invariance. A train MTL : Multi-task baseline train accuracy.  ... 
arXiv:2207.08304v1 fatcat:pryxbpkzybbv3dri7ed3l5uk2a

Layered Hypernetwork Models for Cross-Modal Associative Text and Image Keyword Generation in Multimodal Information Retrieval [chapter]

Jung-Woo Ha, Byoung-Hee Kim, Bado Lee, Byoung-Tak Zhang
2010 Lecture Notes in Computer Science  
Here, we propose a novel text and image keyword generation method by cross-modal associative learning and inference with multimodal queries.  ...  We use a modified hypernetwork model, i.e. layered hypernetworks (LHNs) which consists of the first (lower) layer and the second (upper) layer which has more than two modality-dependent hypernetworks and  ...  As learning continues, the structure of a hypernetwork fits the distribution of given data more.  ... 
doi:10.1007/978-3-642-15246-7_10 fatcat:yvileqmhfnf6jhmvjwfuzzgmsm

Hypernetwork Dismantling via Deep Reinforcement Learning [article]

Dengcheng Yan, Wenxin Xie, Yiwen Zhang, Qiang He, Yun Yang
2022 arXiv   pre-print
In this work, we formulate the hypernetwork dismantling problem as a node sequence decision problem and propose a deep reinforcement learning (DRL)-based hypernetwork dismantling framework.  ...  Then trial-and-error dismantling tasks are conducted by an agent on these synthetic hypernetworks, and the dismantling strategy is continuously optimized.  ...  L = L Q + αL E (15) With repeatedly gathering experiences and learning from them, the agent updates its hypernetwork dismantling strategy continuously.  ... 
arXiv:2104.14332v2 fatcat:ob7txofxzffgvlbed6c3r3ekm4
« Previous Showing results 1 — 15 out of 697 results