Article
Physics, Fluids & Plasmas
Jing Liu, Sujie Li, Jiang Zhang, Pan Zhang
Summary: Modeling the joint distribution of high-dimensional data has been a central task in unsupervised machine learning. In this letter, a new tensor network model, autoregressive matrix product states (AMPS), is proposed, combining matrix product states and autoregressive modeling. Extensive numerical experiments demonstrate that AMPS significantly outperforms existing tensor network models, restricted Boltzmann machines, and is competitive with state-of-the-art neural network models in generative modeling and reinforcement learning.
Article
Computer Science, Artificial Intelligence
Qiang Zhou, Wen'an Zhou, Shirui Wang, Ying Xing
Summary: This paper proposes a novel Multiple Adversarial Networks (MAN) and its improved version (iMAN) for unsupervised domain adaptation, which reduce domain shift between source and target domains with adversarial learning, while considering class discrepancy, achieving significant effectiveness in experiments.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Alberto Barbado, Oscar Corcho, Richard Benjamins
Summary: This paper focuses on the black box problem in unsupervised learning, evaluating rule extraction techniques from OneClass SVM models and proposing algorithms for computing XAI-related metrics. The research evaluates the proposals with different data sets, including real-world data, aiming to extend XAI techniques to unsupervised machine learning models.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Materials Science, Multidisciplinary
Yuan-Heng Tseng, Fu-Jiun Jiang
Summary: This study investigates the impact of using various training sets on the performance of an unsupervised neural network (NN) for learning the phases of a two-dimensional ferromagnetic Potts model, specifically a deep learning autoencoder (AE). The results show that data below and near the transition temperature T-c are crucial in successfully training the AE. Additionally, the commonly used training procedures for unsupervised NNs are found to be inefficient, and the findings from this study can serve as useful guidelines for setting up effective trainings for unsupervised NNs.
RESULTS IN PHYSICS
(2022)
Article
Engineering, Electrical & Electronic
Fernando Gama, Nicolas Zilberstein, Martin Sevilla, Richard G. Baraniuk, Santiago Segarra
Summary: Accurate estimation of states is crucial for the design, synthesis, and analysis of nonlinear dynamical systems. Particle filters are computational estimators that simulate trajectories from a sampling distribution and average them based on importance weight. This work proposes learning sampling distributions using four methods, three of which are parametric and one is nonparametric. Computational experiments show that learned sampling distributions outperform designed ones.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Ronghang Zhu, Xiaodong Jiang, Jiasen Lu, Sheng Li
Summary: In this article, a novel transferable feature learning approach on graphs (TFLG) is proposed for unsupervised adversarial domain adaptation (DA). The approach incorporates sample- and class-level structure information across two domains by constructing graphs, designing a cross-domain graph convolutional operation, and utilizing a memory bank. Extensive experiments on benchmark datasets demonstrate the effectiveness of the proposed approach compared to the state-of-the-art UDA methods.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Alper Ahmetoglu, M. Yunus Seker, Justus Piater, Erhan Oztop, Emre Ugur
Summary: Symbolic planning and reasoning are powerful tools for robots tackling complex tasks. In this paper, a novel general method is proposed to find action-grounded object and effect categories and build probabilistic rules for non-trivial action planning. Experimental results demonstrate the effectiveness of this method in multi-step object manipulation tasks.
JOURNAL OF ARTIFICIAL INTELLIGENCE RESEARCH
(2022)
Article
Mathematics
Oleg Gorokhov, Mikhail Petrovskiy, Igor Mashechkin, Maria Kazachuk
Summary: In this paper, a new robust approach based on a convolutional autoencoder using fuzzy clustering is proposed to address the cybersecurity and reliability issues in computer systems. Compared to existing methods, this approach is more efficient in feature extraction and handling outliers.
Article
Engineering, Electrical & Electronic
Mohamed Labana, Walaa Hamouda
Summary: This paper addresses the power allocation problem in CRAN using an unsupervised deep learning approach, taking into account user association and proposing a new scheme to enhance the deep learning-based power allocation methods. The results show that this technique can achieve near optimal performance with negligible computational complexity.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY
(2021)
Article
Telecommunications
Rasoul Nikbakht, Anders Jonsson, Angel Lozano
Summary: This letter proposes the unsupervised training of a feedforward neural network to solve parametric optimization problems involving large numbers of parameters, which avoids the precomputation of labeled training data that supervised learning necessitates. The technique is applied to general constrained quadratic program and specialized wireless communication problems, showing satisfactory performance and superior scalability with problem dimensionality compared to convex solvers.
IEEE COMMUNICATIONS LETTERS
(2021)
Article
Computer Science, Artificial Intelligence
Yanan Li, Yifei Liu, Dingrun Zheng, Yuhan Huang, Yuling Tang
Summary: Unsupervised domain adaptation aims to alleviate data distribution mismatch and labeling consumption. In this paper, a method called Discriminable Feature Enhancement (DFE-DA) is proposed to identify cross-domain cotton boll maturity. Experimental results show that DFE-DA outperforms other methods with an average improvement of 12.8%, 10.3%, and 7.6% in different transfer tasks.
IMAGE AND VISION COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Wei Ju, Yiyang Gu, Xiao Luo, Yifan Wang, Haochen Yuan, Huasong Zhong, Ming Zhang
Summary: This paper proposes an unsupervised graph-level representation learning framework called Hierarchical Graph Contrastive Learning (HGCL), which addresses the issues of limited exploration of semantic information for graph representation and memory problems during optimization in graph domains. HGCL investigates the hierarchical structural semantics of a graph at both node and graph levels through contrastive learning. Experimental results demonstrate that HGCL outperforms a broad range of state-of-the-art baselines in graph classification and transfer learning tasks.
Article
Geochemistry & Geophysics
Wenqian Fang, Lihua Fu, Hongwei Li
Summary: This study introduces an unsupervised random-noise-suppression method for seismic data denoising, which can train a network directly on noisy target data without noise-free labels. The method is inspired by the idea of averaging multiple noisy observations for denoising and requires noise to satisfy zero-mean and independence of the signal assumptions. By using multiple observations as labels and rearranging data with self-similar blocks, unsupervised training is achieved effectively.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Shuai Fu, Jing Chen, Liang Lei
Summary: The domain shift between the training dataset and test dataset often affects the performance of deep learning models. In order to address this issue, we propose a Cooperative Attention Generative Adversarial Network (CAGAN) that generates target samples with given class labels and implements class-level transfer. The model integrates Coupled Generative Adversarial Networks (CoGAN) into a classification network and employs a semantic-consistent loss and attention layers for improved domain adaptation performance.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Tingting Zhao, Zifeng Wang, Aria Masoomi, Jennifer Dy
Summary: This paper discusses the concept of lifelong learning and focuses on unsupervised lifelong learning. By utilizing a Bayesian inference framework and a novel algorithm, the paper presents a method that can discover new clusters while retaining past knowledge, and demonstrates its effectiveness in both LL and batch settings.
Article
Multidisciplinary Sciences
Mickael Zbili, Sylvain Rama, Pierre Yger, Yanis Inglebert, Norah Boumedine-Guignon, Laure Fronzaroli-Moliniere, Romain Brette, Michael Russier, Dominique Debanne
Article
Biochemistry & Molecular Biology
Julien Fournier, Aman B. Saleem, E. Mika Diamanti, Miles J. Wells, Kenneth D. Harris, Matteo Carandini
Article
Neurosciences
Quentin Gaucher, Pierre Yger, Jean-Marc Edeline
JOURNAL OF PHYSIOLOGY-LONDON
(2020)
Article
Neurosciences
Jai Bhagat, Miles J. Wells, Kenneth D. Harris, Matteo Carandini, Christopher P. Burgess
Article
Multidisciplinary Sciences
Andrew J. Peters, Julie M. J. Fabre, Nicholas A. Steinmetz, Kenneth D. Harris, Matteo Carandini
Summary: The relationship between cortical and striatal activity is precise, topographic, causal, and invariant to behavior. Striatal activity follows a mediolateral gradient related to different behavioral correlates. This consistent and causal mapping of cortical activity in the striatum is not influenced by task engagement.
Article
Biology
E. Mika Diamanti, Charu Bai Reddy, Sylvia Schroder, Tomaso Muzzu, Kenneth D. Harris, Aman B. Saleem, Matteo Carandini
Summary: The study reveals that during navigation, neurons in the mouse primary visual cortex (V1) are influenced by the animal's spatial position, with this modulation also present in higher visual areas but not in the main thalamic pathway to V1. Similar to the hippocampus, spatial modulation in the visual cortex strengthens with experience and active behavior, suggesting that active navigation in a familiar environment enhances the spatial modulation of visual signals originating in the cortex.
Article
Neurosciences
Morgane M. Moss, Peter Zatka-Haas, Kenneth D. Harris, Matteo Carandini, Armin Lak
Summary: Research suggests that dopamine in the striatum plays a critical role in visual decision-making, encoding visual stimuli and rewarded actions in a lateralized fashion. Contrary to previous beliefs, dopamine signals in the DMS respond to contralateral stimuli and rewarded actions, facilitating associations between specific visual stimuli and actions.
JOURNAL OF NEUROSCIENCE
(2021)
Article
Biology
Peter Zatka-Haas, Nicholas A. Steinmetz, Matteo Carandini, Kenneth D. Harris
Summary: Correlates of sensory stimuli and motor actions are found in multiple cortical areas, but only the sensory signals localized in visual and frontal cortex play a causal role in task performance, while widespread dorsal cortical signals correlating with movement reflect processes that do not play a causal role.
Article
Neurosciences
Elizabeth A. Souter, Yen-Chu Chen, Vivien Zell, Valeria Lallai, Thomas Steinkellner, William S. Conrad, William Wisden, Kenneth D. Harris, Christie D. Fowler, Thomas S. Hnasko
Summary: This study demonstrates that glutamate corelease from cholinergic neurons in the medial habenula opposes nicotine self-administration, providing further support for targeting this synapse to develop potential treatments for nicotine addiction.
Article
Neurosciences
Anwar O. Nunez-Elizalde, Michael Krumin, Charu Bai Reddy, Gabriel Montaldo, Alan Urban, Kenneth D. Harris, Matteo Carandini
Summary: Functional ultrasound imaging (fUSI) is an effective method for measuring blood flow and inferring brain activity, showing a strong correlation with slow fluctuations in neural firing rate. The study found that fUSI signals are accurately predicted by the smoothed firing rate of local neurons, particularly inhibitory neurons, and match neural firing spatially across different brain regions.
Article
Multidisciplinary Sciences
Stephane Bugeon, Joshua Duffield, Mario Dipoppa, Anne Ritoux, Isabelle Prankerd, Dimitris Nicoloutsopoulos, David Orme, Maxwell Shinn, Han Peng, Hamish Forrest, Aiste Viduolyte, Charu Bai Reddy, Yoh Isogai, Matteo Carandini, Kenneth D. Harris
Summary: This study reveals that inhibitory subtypes in the primary visual cortex exhibit diverse correlates with brain state. These subtypes' activity patterns are organized by the main axis of transcriptomic variation. Different subtypes show significant differences in response to visual stimuli, as well as modulation by brain state. These findings highlight the importance of inhibitory neurons in cortical processing.
Editorial Material
Neurosciences
Sonja Gruen, Jennifer Li, Bruce McNaughton, Carl Petersen, David McCormick, Drew Robson, Gyorgy Buzsaki, Kenneth Harris, Terrence Sejnowski, Thomas Mrsic-Flogel, Henrik Linden, Per E. Roland
Summary: This article provides an overview of recent discoveries on the spatial interaction between neurons and networks of neurons, and explains the importance of these interactions in fundamental brain and brainstem mechanisms underlying detection, perception, learning, and behavior.
Correction
Multidisciplinary Sciences
Stephane Bugeon, Joshua Duffield, Mario Dipoppa, Anne Ritoux, Isabelle Prankerd, Dimitris Nicoloutsopoulos, David Orme, Maxwell Shinn, Han Peng, Hamish Forrest, Aiste Viduolyte, Charu Bai Reddy, Yoh Isogai, Matteo Carandini, Kenneth D. Harris
Article
Neurosciences
Celian Bimbard, Timothy P. H. Sit, Anna Lebedeva, Charu B. B. Reddy, Kenneth D. D. Harris, Matteo Carandini
Summary: Sensory cortices are considered to be multisensory, as they can be influenced by stimuli of multiple modalities. This study shows that the activity evoked by sounds in the primary visual cortex (V1) is stereotyped across neurons and mice, independent of projections from the auditory cortex. The low-dimensional nature of this activity contrasts with the high-dimensional code used by V1 for representing images. Furthermore, the sound-evoked activity can be accurately predicted by small body movements that are consistent across trials and mice, suggesting that apparent multisensory neural activity may arise from low-dimensional signals associated with internal state and behavior.
NATURE NEUROSCIENCE
(2023)
Article
Multidisciplinary Sciences
L. Federico Rossi, Kenneth D. Harris, Matteo Carandini