Article
Computer Science, Artificial Intelligence
Silvia Bucci, Antonio D'Innocente, Yujun Liao, Fabio Maria Carlucci, Barbara Caputo, Tatiana Tommasi
Summary: Human adaptability relies on learning from both supervised and unsupervised tasks, and this approach can be applied to object recognition across domains. A multi-task method combining supervised and self-supervised knowledge provides competitive results.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Pierluigi Zama Ramirez, Adriano Cardace, Luca De Luigi, Alessio Tonioni, Samuele Salti, Luigi Di Stefano
Summary: The availability of labelled data is a major obstacle for using deep learning algorithms in computer vision tasks in new domains. However, by learning a mapping between task-specific deep features and implementing it using a neural network, we can share knowledge across tasks and generalize to unseen domains. Additionally, we propose strategies to constrain the learned feature spaces, improving the learning process and the generalization capability of the mapping network.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Ju Jia, Meng Luo, Jinshuo Liu, Weixiang Ren, Lina Wang
Summary: The study introduces a novel MPSA scheme that utilizes APL in the latent representation space to reduce nonlinear distribution discrepancy for cross-domain JPEG steganography detection.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Zhuosheng Zhang, Kehai Chen, Rui Wang, Masao Utiyama, Eiichiro Sumita, Zuchao Li, Hai Zhao
Summary: This work proposes new methods to use visual information as assistant signals in natural language processing tasks. It applies Transformer encoder and convolutional neural network to encode text and images, and fuses the two representations through an attention layer.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Le Thanh Nguyen-Meidine, Atif Belal, Madhu Kiran, Jose Dolz, Louis-Antoine Blais-Morin, Eric Granger
Summary: Beyond the complexity of CNNs that require training on large annotated datasets, the domain shift between design and operational data has limited the adoption of CNNs in many real-world applications. Additionally, state-of-the-art CNNs may not be suitable for such real-time applications given their computational requirements. Our proposed approach is compared against state-of-the-art methods for compression and STDA of CNNs on the Office31 and ImageClef-DA image classification datasets, and also against state-of-the-art methods for MTDA on Digits, Office31, and OfficeHome. In both settings, results indicate that our approach can achieve the highest level of accuracy across target domains, while requiring a comparable or lower CNN complexity.
IMAGE AND VISION COMPUTING
(2021)
Article
Computer Science, Artificial Intelligence
Lizi Liao, Ryuichi Takanobu, Yunshan Ma, Xun Yang, Minlie Huang, Tat-Seng Chua
Summary: Conversational systems have received significant attention recently. This paper introduces a Topic-guided Conversational Recommender (TCR) designed for the multi-domain setting, which achieves superior performance compared to a wide range of baselines on a large-scale task-oriented multi-domain dialogue dataset.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2022)
Article
Chemistry, Physical
Qiaohao Liang, Aldair E. Gongora, Zekun Ren, Armi Tiihonen, Zhe Liu, Shijing Sun, James R. Deneault, Daniil Bash, Flore Mekki-Berrada, Saif A. Khan, Kedar Hippalgaonkar, Benji Maruyama, Keith A. Brown, John Fisher Iii, Tonio Buonassisi
Summary: Bayesian optimization (BO) has shown effectiveness in guiding autonomous and high-throughput experiments in materials science. This study evaluated the efficiency of BO across five different experimental materials systems, finding that Gaussian Process (GP) with anisotropic kernels and Random Forest (RF) performed well in BO, outperforming commonly used GP with isotropic kernels. GP with anisotropic kernels demonstrated robustness, while RF is a close alternative with advantages such as lack of distribution assumptions, lower time complexity, and easier initial hyperparameter selection. The benefits of using GP with anisotropic kernels in future materials optimization campaigns were also highlighted.
NPJ COMPUTATIONAL MATERIALS
(2021)
Article
Engineering, Electrical & Electronic
Lei Zhang, Liyun Zuo, Baoyan Wang, Xin Li, Xiantong Zhen
Summary: In this paper, a variational hyperparameter inference method for few-shot learning across domains is proposed, which integrates meta learning and variational inference into the optimization of hyperparameters. By learning adaptive hyperparameters and modeling hyperparameters as distributions, the proposed method improves the generalization ability across domains.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Thomas Mensink, Jasper Uijlings, Alina Kuznetsova, Michael Gygli, Vittorio Ferrari
Summary: This paper conducts extensive experimental exploration of transfer learning across different image domains and task types. The study reveals that for most tasks, there exists a source task that outperforms ILSVRC'12 pre-training. The image domain is found to be the most important factor for achieving positive transfer, and the source dataset should include the image domain of the target dataset for best results. Additionally, transfer across task types can be beneficial, but its success depends on both the source and target task types.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Review
Immunology
Rodolfo A. Kolliker Frers, Matilde Otero-Losada, Tamara Kobiec, Lucas D. Udovin, Maria Laura Aon Bertolino, Maria Herrera, Francisco Capani
Summary: This review examines the use of light chain neurofilaments (NFLs) as peripheral MS biomarkers, aiming to improve our comprehension of the disease.
FRONTIERS IN IMMUNOLOGY
(2022)
Article
Behavioral Sciences
Kelsey A. Heslin, Jessica R. Purnell, Benjamin J. De Corte, Krystal L. Parker
Summary: The involvement of the cerebellum in suprasecond interval timing is controversial. Many previous studies have limitations and may have missed a critical window of cerebellar involvement. In this study, the rat lateral cerebellar nucleus was pharmacologically inactivated across three different peak interval timing tasks, and no strong support for cerebellar involvement in suprasecond interval timing was found.
BEHAVIORAL NEUROSCIENCE
(2022)
Article
Medicine, General & Internal
Varshini Varadaraj, Beatriz Munoz, Jennifer A. Deal, Yang An, Marilyn S. Albert, Susan M. Resnick, Luigi Ferrucci, Bonnielin K. Swenor
Summary: The study found that the association between vision and cognition varies between different measures of vision (visual acuity, contrast sensitivity, and stereo acuity). Patterns of cognitive decline may differ by type of vision impairment.
Article
Computer Science, Artificial Intelligence
Mingxin Jiang, Yuzhong Chen, Jiadong Yan, Zhenxiang Xiao, Wei Mao, Boyu Zhao, Shimin Yang, Zhongbo Zhao, Tuo Zhang, Lei Guo, Benjamin Becker, Dezhong Yao, Keith M. Kendrick, Xi Jiang
Summary: This study developed new anatomy-guided spatio-temporal graph convolutional networks (AG-STGCNs) to investigate the regularity and variability of functional connectivity differences between gyri and sulci across multiple task domains. Based on fMRI datasets from the Human Connectome Project, the study found significant differences in functional connectivity between gyral and sulcal regions within task domains compared to resting state. The study also found considerable variability in functional connectivity and information flow between gyri and sulci across different task domains, which are correlated with individual cognitive behaviors.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Interdisciplinary Applications
Ke Yan, Jinzheng Cai, Youjing Zheng, Adam P. Harrison, Dakai Jin, Youbao Tang, Yuxing Tang, Lingyun Huang, Jing Xiao, Le Lu
Summary: Large-scale datasets in medical imaging are often partially labeled or small, presenting challenges in training accurate lesion detection models. In this work, we introduce a framework named LENS that learns from multiple heterogeneous lesion datasets and improves performance through proposal fusion. By mining missing annotations from partially labeled datasets using clinical prior knowledge and cross-dataset knowledge transfer, we successfully address the issues of heterogeneous and partial labels.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2021)
Article
Rehabilitation
Brittany R. Lapin, Nicolas R. Thompson, Andrew Schuster, Irene L. Katzan
Summary: The study found that proxies completing patient-reported outcome measures for stroke patients reported worse functioning and more symptoms compared to patients, with lower agreement on more subjective domains. At the individual level, a large proportion of dyads reported meaningfully different scores across all domains, affecting the interpretability of proxy responses on outcome measures in a clinical setting.
ARCHIVES OF PHYSICAL MEDICINE AND REHABILITATION
(2021)
Article
Neurosciences
Giacomo Ariani, Young Han Kwon, Jorn Diedrichsen
JOURNAL OF NEUROPHYSIOLOGY
(2020)
Article
Neurosciences
Andrew S. Fox, Daniel Holley, Peter Christiaan Klink, Spencer A. Arbuckle, Carol A. Barnes, Joern Diedrichsen, Sze Chai Kwok, Colin Kyle, J. Andrew Pruszynski, Jakob Seidlitz, XuFeng Zhou, Russell A. Poldrack, Krzysztof J. Gorgolewski
Summary: Animal neuroimaging studies offer unique insights into brain structure and function, bridging the gap between animal and human neuroscience. However, nonhuman primate neuroimaging studies often lack statistical power, highlighting the need for increased data sharing to facilitate cross-species research. Efforts to share data have been limited by the lack of standardized tools and repositories, but advancements in Neurovault now allow for easy sharing of nonhuman primate neuroimaging results, promising to enable novel cross-species comparisons.
Article
Neurosciences
Giacomo Ariani, Neda Kordjazi, J. Andrew Pruszynski, Jorn Diedrichsen
Summary: The study shows that participants can use visual information to plan multiple future actions during ongoing movements, but the planning horizon is limited, and receiving information about more than three movements ahead does not result in faster sequence production. With practice, participants demonstrated larger performance improvements for larger viewing windows and an expansion of the planning horizon.
Article
Neurosciences
Naotoshi Abekawa, Hiroaki Gomi, Jorn Diedrichsen
Summary: The study examined how gaze control during reaching is modulated by task demands, finding that changes in reward contingencies can affect saccade latency and reach accuracy. Results showed that early saccades are costly for reaching, and the brain modulates inhibitory online coordination from the hand to the eye system depending on task requirements.
JOURNAL OF NEUROPHYSIOLOGY
(2021)
Article
Neurosciences
Eva Berlot, Nicola J. Popp, Scott T. Grafton, Jorn Diedrichsen
Summary: In the context of motor sequence learning, fMRI studies revealed differences in neuronal representations between premotor and parietal regions compared to the primary motor cortex (M1). While M1 showed specific representation of the first finger of each sequence, parietal areas represented the identity of the entire sequence and remained relatively stable during different executions. This suggests that the RS effect in M1 reflects a preparatory signal for movement initiation rather than a trained sequence representation.
JOURNAL OF NEUROSCIENCE
(2021)
Article
Neurosciences
Meret Branscheidt, Naveed Ejaz, Jing Xu, Mario Widmer, Michelle D. Harran, Juan Camilo Cortes, Tomoko Kitago, Pablo Celnik, Carlos Hernandez-Castillo, Jorn Diedrichsen, Andreas Luft, John W. Krakauer
Summary: Despite substantial motor recovery, longitudinal changes in functional connectivity after stroke were not detected between patients and controls, raising doubts about the relevance of changes in cortico-cortical connectivity for promoting motor recovery. The results suggest that the value of resting-state imaging in assessing poststroke cortical connectivity changes may need to be further examined.
JOURNAL OF NEUROPHYSIOLOGY
(2022)
Article
Neurosciences
Alexandra Reichenbach, Buse M. Urgen, Sotirios Apostolakis, Liora Michlin, Jorn Diedrichsen
Summary: In order to achieve fast feedback control of voluntary movements, the brain's motor control processes need to quickly recognize and analyze the visual consequences of motor commands. These processes are able to work well in complex visual environments and even when there are differences between physical actuator and visually perceived effect. Researchers found that the visuomotor system is highly sensitive to the spatial and temporal correlation between a cursor and hands, and is capable of implicitly learning the appropriate mapping within minutes. The spatial proximity between the end effector and visual consequence has an immediate but temporary effect on the assignment process.
JOURNAL OF NEUROPHYSIOLOGY
(2022)
Article
Neurosciences
Da Zhi, Maedbh King, Carlos R. Hernandez-Castillo, Jorn Diedrichsen
Summary: One important approach to human brain mapping is to define distinct regions linked to unique functions. However, comparing different parcellations based on brain data is challenging. To address this issue, this study proposes a new unbiased criterion, the distance-controlled boundary coefficient (DCBC), to evaluate discrete parcellations. The DCBC is used to evaluate existing parcellations of the human neocortex and to predict functional boundaries. The results show that anatomical parcellations do not perform better than chance, while those based on resting-state fMRI data perform well. In addition, multi-modal parcellations combining functional and anatomical criteria perform worse than those based on functional data alone.
HUMAN BRAIN MAPPING
(2022)
Article
Neurosciences
Nicola J. Popp, Carlos R. Hernandez-Castillo, Paul L. Gribble, Jorn Diedrichsen
Summary: Sensory feedback is crucial for fine control of hand movements, but little is known about its role in movement sequences. This study investigated the use of sensory feedback during the production of finger movement sequences. The researchers observed rapid adjustments of ongoing finger presses in response to feedback perturbations, with haptic feedback playing a key role. These adjustments reduced in size with practice but were still present at the end of training. Additionally, feedback perturbations resulted in a delayed onset of subsequent presses, suggesting a hierarchical organization of skilled movement sequences.
JOURNAL OF NEUROPHYSIOLOGY
(2022)
Article
Neurosciences
Luke Bashford, Dmitry Kobak, Jorn Diedrichsen, Carsten Mehring
Summary: We investigated the learning of motor skills using a path tracking task. We found that subjects' accuracy improved with practice, even when tracking unfamiliar paths. Subjects with higher tracking skills had lower movement variability and a longer planning horizon. The increase in performance in the expert group was partially attributed to the longer planning horizon.
JOURNAL OF NEUROPHYSIOLOGY
(2022)
Article
Neurosciences
Spencer A. Arbuckle, J. Andrew Pruszynski, Jorn Diedrichsen
Summary: The integration of somatosensory signals across fingers plays a crucial role in dexterous object manipulation, and this integration mainly occurs in the primary somatosensory cortex. Through stimulating different finger combinations and using fMRI technology, researchers have discovered unique nonlinear interactions between fingers. This integration contributes to the flexible mapping from finger sensory inputs to motor responses.
JOURNAL OF NEUROSCIENCE
(2022)
Article
Multidisciplinary Sciences
Daan B. Wesselink, Zeena-Britt Sanders, Laura R. Edmondson, Harriet Dempsey-Jones, Paulina Kieliba, Sanne Kikkert, Andreas C. Themistocleous, Uzay Emir, Jorn Diedrichsen, Hannes P. Saal, Tamar R. Makin
Summary: This study investigates remapping in the brain map after finger amputation using pharmacological and neuroimaging methods. The findings reveal persistent representation of missing fingers in the brain map even decades after amputation. The study also shows that the blocked finger representation remains despite input loss.
Editorial Material
Biology
Michael B. Eisen, Anna Akhmanova, Timothy E. Behrens, Jorn Diedrichsen, Diane M. Harper, Mihaela D. Iordanova, Detlef Weigel, Mone Zaidi
Summary: eLife is changing its editorial process to prioritize public reviews and assessments of preprints.
Article
Biology
Maedbh King, Ladan Shahshahani, Richard B. Ivry, Jorn Diedrichsen
Summary: While resting-state fMRI studies provide an overview of the connectivity between the human neocortex and cerebellum, it is unclear how cortical inputs converge onto cerebellar circuits. This study used task-based fMRI data to build different models of cortico-cerebellar connectivity, and found that models allowing for some degree of convergence provided the best predictions. The degree of convergence varied across the cerebellum, with higher convergence observed in areas related to language, working memory, and social cognition.
Article
Biology
Daniel Haenelt, Robert Trampel, Shahin Nasr, Jonathan R. Polimeni, Roger B. H. Tootell, Martin Sereno, Kerrin J. Pine, Luke J. Edwards, Saskia Helbling, Nikolaus Weiskopf
Summary: The characterization of cortical myelination is important for studying the relationship between brain structure and function, but current knowledge is based on post-mortem histology. This study used qMRI and ultra-high field strength fMRI to investigate myelination in the V2 cortex of living humans. The results demonstrate the feasibility of studying structure-function relationships in vivo using qMRI.