Filters








5,250 Hits in 3.1 sec

Meta-Reinforcement Learning via Buffering Graph Signatures for Live Video Streaming Events [article]

Stefanos Antaris, Dimitrios Rafailidis, Sarunas Girdzijauskas
2021 arXiv   pre-print
Our experiments demonstrate the effectiveness of our proposed model, with an average relative gain of 25% against state-of-the-art strategies.  ...  We evaluate the proposed model on the link weight prediction task on three real-world datasets of live video streaming events.  ...  Meta-Graph adopts the Model-Agnostic Meta-Learning framework [14] and graph signature functions [21] to perform link prediction on new graphs with limited edges [13] .  ... 
arXiv:2111.09412v1 fatcat:3mml2xxlrrconkowa4kkvncoee

Continual Learning on Noisy Data Streams via Self-Purified Replay [article]

Chris Dongjoo Kim, Jinseo Jeong, Sangwoo Moon, Gunhee Kim
2021 arXiv   pre-print
The empirical results on MNIST, CIFAR-10, CIFAR-100, and WebVision with real-world noise demonstrate that our framework can maintain a highly pure replay buffer amidst noisy streamed data while greatly  ...  Our solution is based on two observations; (i) forgetting can be mitigated even with noisy labels via self-supervised learning, and (ii) the purity of the replay buffer is crucial.  ...  Acknowledgement We express our gratitude for the helpful comments on the manuscript by Junsoo Ha, Soochan Lee and the anonymous reviewers for their thoughtful suggestions.  ... 
arXiv:2110.07735v1 fatcat:42qdovvra5ghzaev6ubkl3ende

Catastrophic Forgetting in Deep Graph Networks: an Introductory Benchmark for Graph Classification [article]

Antonio Carta, Andrea Cossu, Federico Errica, Davide Bacciu
2021 arXiv   pre-print
To do so, we experiment with a structure-agnostic model and a deep graph network in a robust and controlled environment on three different datasets.  ...  Finally, we provide researchers with a flexible software framework to reproduce our results and carry out further experiments.  ...  This work makes the first step in this direction by carrying out continual learning experiments on graph classification benchmarks in a robust and controlled framework.  ... 
arXiv:2103.11750v1 fatcat:rgzetw7advf53oitqtpgdwquhm

TrafficStream: A Streaming Traffic Flow Forecasting Framework Based on Graph Neural Networks and Continual Learning [article]

Xu Chen and Junshan Wang and Kunqing Xie
2021 arXiv   pre-print
Extensive experiments demonstrate its excellent potential to extract traffic patterns with high efficiency on long-term streaming network scene.  ...  To tackle this problem, we propose a Streaming Traffic Flow Forecasting Framework, TrafficStream, based on Graph Neural Networks (GNNs) and Continual Learning (CL), achieving accurate predictions and high  ...  ., 2020] and Feature Graph [Wang et al., 2020a] presents different experience replay strategies for continual graph representation learning.  ... 
arXiv:2106.06273v1 fatcat:wli6agzgcfc5xdzfjingqicegi

RLlib Flow: Distributed Reinforcement Learning is a Dataflow Problem [article]

Eric Liang, Zhanghao Wu, Michael Luo, Sven Mika, Joseph E. Gonzalez, Ion Stoica
2021 arXiv   pre-print
Researchers and practitioners in the field of reinforcement learning (RL) frequently leverage parallel computation, which has led to a plethora of new algorithms and systems in the last few years.  ...  Replay: On-policy algorithms (e.g., PPO [27] , A3C [22] ) collect new experiences from the current policy to learn.  ...  Reinforcement Learning vs Data Streaming The key observation behind RLlib Flow is that the dataflow graph of RL algorithms are quite similar to those of data streaming applications.  ... 
arXiv:2011.12719v4 fatcat:o7euvwohgrgtrazko3niasln4e

Mitigating Forgetting in Online Continual Learning via Instance-Aware Parameterization

Hung-Jen Chen, An-Chieh Cheng, Da-Cheng Juan, Wei Wei, Min Sun
2020 Neural Information Processing Systems  
Online continual learning is a challenging scenario where a model needs to learn from a continuous stream of data without revisiting any previously encountered data instances.  ...  On the other hand, it also encourages fine-tuning the path if the incoming instance shares the similarity with previous instances.  ...  On the other hand, to our best knowledge, most reinforcement learning relies on experience replay to handle streaming data, in which the buffer size is often set to 1000 or bigger.  ... 
dblp:conf/nips/ChenCJ0S20 fatcat:qszxk7q2afb27kz65b4dk3ugjy

Selective Replay Enhances Learning in Online Continual Analogical Reasoning [article]

Tyler L. Hayes, Christopher Kanan
2021 arXiv   pre-print
We employ experience replay to mitigate catastrophic forgetting.  ...  In continual learning, a system learns from non-stationary data streams or batches without catastrophic forgetting.  ...  Specifically, streaming learning requires an agent to learn new samples one at a time (N t = 1) with only a single epoch through the entire dataset.  ... 
arXiv:2103.03987v2 fatcat:x44qwvx4szcelpecp4zxbukir4

CLEF 2017 NewsREEL Overview: Offline and Online Evaluation of Stream-based News Recommender Systems

Benjamin Kille, Andreas Lommatzsch, Frank Hopfgartner, Martha A. Larson, Torben Brodt
2017 Conference and Labs of the Evaluation Forum  
Compared with the previous year NewsREEL challenged participants with a higher volume of messages and new news portals.  ...  In addition, the main results of Living Lab and the Replay task are explained.  ...  The graph is managed in a Neo4j graph database. Recommendations are computed based on a database query.  ... 
dblp:conf/clef/KilleLHLB17 fatcat:k5aelrasrrdr7cve3u5hszgsfq

Distilling Causal Effect of Data in Class-Incremental Learning [article]

Xinting Hu, Kaihua Tang, Chunyan Miao, Xian-Sheng Hua, Hanwang Zhang
2021 arXiv   pre-print
Based on the framework, we find that although the feature/label distillation is storage-efficient, its causal effect is not coherent with the end-to-end feature learning merit, which is however preserved  ...  Extensive experiments on three CIL benchmarks: CIFAR-100, ImageNet-Sub&Full, show that the proposed causal effect distillation can improve various state-of-the-art CIL methods by a large margin (0.72%-  ...  Experiments are conducted on CIFAR-100 with 5 incremental steps.  ... 
arXiv:2103.01737v3 fatcat:m2hwe5podnfy5gtz3d4c7dmjmu

Deep Reinforcement Learning-Based Channel Allocation for Wireless LANs with Graph Convolutional Networks

Kota Nakashima, Shotaro Kamiya, Kazuki Ohtsu, Koji Yamamoto, Takayuki Nishio, Masahiro Morikura
2019 2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall)  
INDEX TERMS Wireless LAN, channel allocation, deep reinforcement learning, graph convolutional networks, replay buffer.  ...  First, we adopt graph convolutional layers in the model to extract the features of the channel vectors with topology information, which is the adjacency matrix of the graph dependent on the carrier sensing  ...  PRIORITIZED EXPERIENCE REPLAY In this paper, we sample the training data from the replay buffer D according to the prioritized experience replay [35] , which allocates priority to all samples based on  ... 
doi:10.1109/vtcfall.2019.8891178 dblp:conf/vtc/NakashimaKOYNM19 fatcat:dca3rcifbbbndgetet7i3qxxfm

Deep Reinforcement Learning-Based Channel Allocation for Wireless LANs with Graph Convolutional Networks

Kota Nakashima, Shotaro Kamiya, Kazuki Ohtsu, Koji Yamamoto, Takayuki Nishio, Masahiro Morikura
2020 IEEE Access  
INDEX TERMS Wireless LAN, channel allocation, deep reinforcement learning, graph convolutional networks, replay buffer.  ...  First, we adopt graph convolutional layers in the model to extract the features of the channel vectors with topology information, which is the adjacency matrix of the graph dependent on the carrier sensing  ...  PRIORITIZED EXPERIENCE REPLAY In this paper, we sample the training data from the replay buffer D according to the prioritized experience replay [35] , which allocates priority to all samples based on  ... 
doi:10.1109/access.2020.2973140 fatcat:fuskfwj6sjeablgw2lga2zjeny

Overcoming Catastrophic Forgetting in Graph Neural Networks with Experience Replay [article]

Fan Zhou, Chengtai Cao
2021 arXiv   pre-print
Extensive experiments on three benchmark datasets demonstrate the effectiveness of our ER-GNN and shed light on the incremental graph (non-Euclidean) structure learning.  ...  Towards that, we explore the Continual Graph Learning (CGL) paradigm and present the Experience Replay based framework ER-GNN for CGL to alleviate the catastrophic forgetting problem in existing GNNs.  ...  As part of our future work, we are planning to extend the continual learning to different graph-related tasks, such as alignments and cascades/diffusion prediction under sequences of evolving data.  ... 
arXiv:2003.09908v2 fatcat:yby3urpyxncdbfmwag4ffbublm

New Insights on Reducing Abrupt Representation Change in Online Continual Learning [article]

Lucas Caccia, Rahaf Aljundi, Nader Asadi, Tinne Tuytelaars, Joelle Pineau, Eugene Belilovsky
2022 arXiv   pre-print
Experience Replay (ER), where a small subset of past data is stored and replayed alongside new data, has emerged as a simple and effective learning strategy.  ...  We shed new light on this question by showing that applying ER causes the newly added classes' representations to overlap significantly with the previous classes, leading to highly disruptive parameter  ...  We combine the loss functions on the incoming and replay data L(X in ∪ X bf ) = γL 1 (X in ) + L 2 (X bf ) (3) We refer to this approach as Experience Replay with Asymmetric Metric Learning (ER-AML).  ... 
arXiv:2104.05025v3 fatcat:5xdh6quz4rct5o5nbrwdpkdmja

New Insights on Reducing Abrupt Representation Change in Online Continual Learning [article]

Lucas Caccia, Rahaf Aljundi, Nader Asadi, Tinne Tuytelaars, Joelle Pineau, Eugene Belilovsky
2022 arXiv   pre-print
Experience Replay (ER), where a small subset of past data is stored and replayed alongside new data, has emerged as a simple and effective learning strategy.  ...  We shed new light on this question by showing that applying ER causes the newly added classes' representations to overlap significantly with the previous classes, leading to highly disruptive parameter  ...  We combine the loss functions on the incoming and replay data L(X in ∪ X bf ) = γL 1 (X in ) + L 2 (X bf ) (3) We refer to this approach as Experience Replay with Asymmetric Metric Learning (ER-AML).  ... 
arXiv:2203.03798v3 fatcat:flmqus4tnjdcpjs4774ckg5wxq

Deep Reinforcement Learning-Based Channel Allocation for Wireless LANs with Graph Convolutional Networks [article]

Kota Nakashima, Shotaro Kamiya, Kazuki Ohtsu, Koji Yamamoto, Takayuki Nishio, Masahiro Morikura
2019 arXiv   pre-print
The present study proposes a deep reinforcement learning-based channel allocation scheme using graph convolutional networks (GCNs).  ...  Last year, IEEE 802.11 Extremely High Throughput Study Group (EHT Study Group) was established to initiate discussions on new IEEE 802.11 features.  ...  The main factors of the DQN include experience replay and fixed target Q-network. Generally, Q-learning with function approximation may not converge [14] .  ... 
arXiv:1905.07144v1 fatcat:mm2q7a3zprcgtf6tntrj4nluiy
« Previous Showing results 1 — 15 out of 5,250 results