Article
Multidisciplinary Sciences
Noam Nitzan, Rachel Swanson, Dietmar Schmitz, Gyorgy Buzsaki
Summary: The study found that sharp wave ripples (SPW-Rs) coincided with a transient brain-wide increase in functional connectivity, and the diversity in SPW-R features was correlated with their intrahippocampal topography along the septotemporal axis. Furthermore, variations in SPW-R features were related to the timing, sign, and magnitude of downstream responses.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2022)
Article
Computer Science, Artificial Intelligence
Krishnakumar Santhakumar, Hamidreza Kasaei
Summary: This paper introduces a hybrid model architecture to simultaneously address object recognition and grasping issues using a dynamically growing dual-memory recurrent neural network and an autoencoder. The problem of catastrophic forgetting is addressed by intrinsic memory replay, and the model is evaluated in a lifelong learning setting.
Article
Computer Science, Artificial Intelligence
Animesh Renanse, Alok Sharma, Rohitash Chandra
Summary: This paper examines the memory capacity of matrix-based RNNs and compares it with conventional neural networks. The study finds that neural networks with matrix representations have better memory capacity and the performance is enhanced when external memory is introduced.
Article
Multidisciplinary Sciences
S. Kapl, F. Tichanek, F. Zitricky, K. Jezek
Summary: The hippocampus plays a crucial role in spatial and episodic memory formation and retrieval. This study reveals that in rats, CA3 neurons can spontaneously switch representations and activate familiar spatial patterns without external cues. This phenomenon occurs during active exploration and is influenced by sudden changes in the environment.
SCIENTIFIC REPORTS
(2022)
Article
Multidisciplinary Sciences
Elliott Wimmer, Yunzhe Liu, Daniel C. McNamee, Raymond J. Dolan
Summary: Theories of neural replay propose that it supports various functions, mainly planning and memory consolidation. This study tests the hypothesis that different signatures of replay in the same task are related to model-based decision-making and memory preservation. Using magnetoencephalography and multivariate analysis, the researchers identify temporally compressed sequential reactivation, or replay, before choice and following reward feedback. The results support key theoretical proposals regarding the functional role of replay and demonstrate the modulation of planning and memory-related signals by ongoing computational and task demands.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2023)
Article
Computer Science, Artificial Intelligence
Chang Yifan, Chen Yulu, Zhang Yadan, Li Wenbo
Summary: This paper proposes a strategy called Continual Learning Easy to Hard (CLeToh) to improve the training efficacy of the Memory Replay (MR) method. CLeToh arranges the data from easy to hard based on their difficulty and trains the model gradually. It increases convergence speed, stabilizes the training process, and overcomes the problem of catastrophic forgetting in MR.
APPLIED INTELLIGENCE
(2023)
Article
Neurosciences
Daniel B. Rubin, Tommy Hosman, Jessica N. Kelemen, Anastasia Kapitonava, Francis R. Willett, Brian F. Coughlin, Eric Halgren, Eyal Y. Kimchi, Ziv M. Williams, John D. Simeral, Leigh R. Hochberg, Sydney S. Cash
Summary: Replay of motor cortex neural activity may occur during sleep following motor learning in humans.
JOURNAL OF NEUROSCIENCE
(2022)
Article
Environmental Sciences
Yi Zhang, Junfu Fan, Mengzhen Zhang, Zongwen Shi, Rufei Liu, Bing Guo
Summary: This paper proposes an improved method called a recurrent adaptive network for road crack segmentation, aiming to solve the imbalance issue between positive and negative samples. By dynamically evaluating the imbalance, determining the sampling rates, and adjusting the loss weights, the method achieves a flow between precision and recall. Experimental results on a high-resolution road crack image dataset demonstrate the effectiveness of the approach, which achieves state-of-the-art performance.
Article
Neurosciences
Daniel N. Barry, Bradley C. Love
Summary: Replay helps consolidate memories and promote generalization and adaptation, with different layers of a neural network playing distinct roles in the replay and consolidation processes.
Review
Neurosciences
Emma L. Roscow, Raymond Chua, Rui Ponte Costa, Matt W. Jones, Nathan Lepora
Summary: This article reviews the functional roles of replay in reinforcement learning and discusses recent developments in the fields of neuroscience and artificial intelligence. It is found that replay is crucial for memory consolidation and stabilizing learning in deep neural networks, supporting generalization and continual learning processes.
TRENDS IN NEUROSCIENCES
(2021)
Article
Neurosciences
Weinan Sun, Madhu Advani, Nelson Spruston, Andrew Saxe, James E. Fitzgerald
Summary: Memorization and generalization are cognitive processes that promote adaptive behavior. Animals memorize safe routes to water sources and generalize from these memories to predict new ones. Systems consolidation mechanisms construct neocortical memory traces, but why it only applies to a subset of memories is unclear. This study proposes that memories only consolidate when it aids generalization, providing new insights into adaptive behavior.
NATURE NEUROSCIENCE
(2023)
Article
Psychology, Multidisciplinary
Xiangshuai Zeng, Nicolas Diekmann, Laurenz Wiskott, Sen Cheng
Summary: This article proposes two different modes, retrieval and replay, to explain how episodic memory drives future behavior. The study found that episodic memory benefits learning in certain conditions, but the performance difference is significant only when the task is complex and the learning trials are limited. Furthermore, the two modes of accessing episodic memory affect spatial learning differently. One-shot learning is typically faster, while replay learning may reach better asymptotic performance.
FRONTIERS IN PSYCHOLOGY
(2023)
Article
Psychology, Multidisciplinary
Sebastian Michelmann, Uri Hasson, Kenneth A. Norman
Summary: When recalling memories, our brain accesses and searches through information-rich continuous episodes. This process is guided by high-level structure called event boundaries, which allows for faster memory scanning by skipping to the next boundary upon reaching a decision threshold.
PSYCHOLOGICAL SCIENCE
(2023)
Article
Computer Science, Artificial Intelligence
Dongjing Shan, Yong Luo, Xiongwei Zhang, Chao Zhang
Summary: Recurrent neural networks (RNNs) are excellent in sequence learning tasks, but training for long sequences is difficult. We propose dynamic recurrent routing neural networks (DRRNets) to address this challenge by dynamically allocating recurrent routes and imposing low-rank constraints.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Biology
Roger Johansson, Marcus Nystroem, Richard Dewhurst, Mikael Johansson
Summary: When recalling visual memories, our eyes repeat the same sequential eye movements as during the original encounter. These eye movements serve as self-generated cues, helping us recreate visuospatial relations during episodic remembering. The fidelity of these eye movements predicts the quality of the recalled memory. These findings provide direct evidence for the role of eye movements in memory and suggest that different properties of eye movements contribute differently to goal-relevant memories.
PROCEEDINGS OF THE ROYAL SOCIETY B-BIOLOGICAL SCIENCES
(2022)
Article
Multidisciplinary Sciences
Richard Naud, Henning Sprekeler
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2018)
Article
Biology
Simon Nikolaus Weber, Henning Sprekeler
Article
Multidisciplinary Sciences
Anna Kutschireiter, Simone Carlo Surace, Henning Sprekeler, Jean-Pascal Pfister
SCIENTIFIC REPORTS
(2017)
Article
Biochemical Research Methods
Simon N. Weber, Henning Sprekeler
PLOS COMPUTATIONAL BIOLOGY
(2019)
Article
Biochemical Research Methods
Loreen Hertaeg, Henning Sprekeler
PLOS COMPUTATIONAL BIOLOGY
(2019)
Article
Biochemical Research Methods
Laura Bella Naumann, Henning Sprekeler
PLOS COMPUTATIONAL BIOLOGY
(2020)
Article
Biology
Loreen Hertaeg, Henning Sprekeler
Article
Biology
Owen Mackwood, Laura B. Naumann, Henning Sprekeler
Summary: Understanding the connectivity in the brain and how it is influenced by activity-dependent synaptic plasticity is a major challenge in neuroscience. Research in mice's V1 area suggests that connections between excitatory and inhibitory neurons contribute to a stimulus-specific competition among neurons, highlighting the importance of synaptic plasticity in shaping cortical computations.
Article
Biochemical Research Methods
Michiel W. H. J. Remme, Urs Bergmann, Denis J. Alevi, Susanne Schreiber, Henning J. Sprekeler, Richard Kempter
Summary: The process of systems memory consolidation involves the transfer of memories between brain regions and the transformation of memory content over time. This can include the conversion of declarative memories from the hippocampal formation to neocortical networks. The consolidation process is thought to rely on Hebbian plasticity in networks with parallel synaptic pathways, and this mechanism can explain changes in memory representations over time, as well as the power-law forgetting curves typically observed in humans.
PLOS COMPUTATIONAL BIOLOGY
(2021)
Article
Biochemical Research Methods
Filip Vercruysse, Richard Naud, Henning Sprekeler
Summary: Cortical pyramidal cells have a specialized dendritic mechanism for generating bursts, playing a special role in information processing. Burst activity is sensitive to dendritic input levels and can be stabilized by a homeostatic inhibitory plasticity rule in dendrite-targeting interneurons. This allows a network state where both spikes and bursts occur asynchronously and irregularly, increasing the information encoded in bursts.
PLOS COMPUTATIONAL BIOLOGY
(2021)
Article
Biochemical Research Methods
Joram J. Keijser, Henning Sprekeler
Summary: The brain's cortical circuits process information through recurrent interactions between excitatory neurons and inhibitory interneurons. This study investigates the specificity of inhibitory feedback in stabilizing the circuit by enabling separate feedback control loops for different synaptic input streams. Using an optimization approach, the researchers found that the resulting circuit can be seen as a neural decoder that reverses the nonlinear biophysical computations within pyramidal cells. This study provides a proof of concept for understanding the structure-function relationships in cortical circuits.
PLOS COMPUTATIONAL BIOLOGY
(2022)
Article
Biology
Laura B. Naumann, Joram Keijser, Henning Sprekeler
Summary: This study investigates the establishment of context-invariant representations through feedback processing. The results show that feedback-modulated feedforward neural networks can dynamically generate invariant sensory representations, rather than on the level of individual neurons. This invariance is achieved by dynamically reorienting the manifold of neural activity and maintaining an invariant neural subspace at the population level.
Article
Multidisciplinary Sciences
M. Belen Pardi, Johanna Vogenstahl, Tamas Dalmay, Teresa Spano, De-Lin Pu, Laura B. Naumann, Friedrich Kretschmer, Henning Sprekeler, Johannes J. Letzkus
Review
Neurosciences
Henning Sprekeler
CURRENT OPINION IN NEUROBIOLOGY
(2017)