Article
Biology
Mattia Chini, Thomas Pfeffer, Ileana Hanganu-Opatz, Liset M. de la Prida
Summary: This study reveals that changes in the ratio of excitation to inhibition control the decorrelation of neural activity during brain development, which may contribute to the pathogenesis of neurodevelopmental disorders.
Article
Computer Science, Artificial Intelligence
Kamalakant Laxman Bawankule, Rupesh Kumar Dewang, Anil Kumar Singh
Summary: This article proposes an early straggler detection method using a recurrent neural network and an autoregressive integrated moving average model, which can significantly reduce job execution time.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Hardware & Architecture
Gauthaman Murali, Xiaoyu Sun, Shimeng Yu, Sung Kyu Lim
Summary: The article presents the first-ever heterogeneous mixed-signal monolithic 3-D IC designs of the RRAM CNN accelerator, which overcome the bottleneck caused by ADCs and offer significant improvement in energy efficiency compared to traditional 2-D IC designs.
IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Yifan Zhu, Qika Lin, Hao Lu, Kaize Shi, Ping Qiu, Zhendong Niu
Summary: A heterogeneous knowledge embedding-based attentive RNN model is proposed in this paper to recommend scientific paper citations, combining structural and textual representations focusing on the "author-text query scenario". By capturing scholars' previous writing and citing preferences through a limited inquiry based on a user's identity, the model aims to reduce recommendation errors. The experimental results on the DBLP dataset demonstrate the feasibility and effectiveness of the method, showing improvements in Mean Reciprocal Rank (MRR) and Normalized Discounted Cumulative Gain (NDCG) compared to existing models.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Cell Biology
Chunyue Geoffrey Lau, Huiqi Zhang, Venkatesh N. Murthy
Summary: The deletion of the BDNF receptor TrkB in PV interneurons in the mouse olfactory cortex impairs multiple aspects of PV neuronal function, leading to aberrant spiking patterns in principal neurons and a paradoxical decrease in overall excitability in cortical circuits. This study demonstrates the critical role of TrkB in shaping the evoked pattern of activity in a cortical network by modulating PV circuit plasticity and development.
JOURNAL OF CELLULAR PHYSIOLOGY
(2022)
Article
Engineering, Multidisciplinary
M. A. Maia, I. B. C. M. Rocha, P. Kerfriden, F. P. van der Meer
Summary: Driven by the need for faster numerical simulations, the use of machine learning techniques is rapidly growing in computational solid mechanics, especially in concurrent multiscale finite element analysis. Surrogate models are being used to approximate microscopic behavior and accelerate simulations, but challenges related to their data-driven nature compromise their reliability. This study introduces a neural network that incorporates classical constitutive models to introduce non-linearity and address these challenges. The network demonstrates the ability to predict unloading/reloading behavior without prior training, unlike popular data-hungry models such as RNNs.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2023)
Article
Multidisciplinary Sciences
Siyang Li, Panpan Zhu, Yaoting Xue, Lei Wang, Tuck-Whye Wong, Xuxu Yang, Haofei Zhou, Tiefeng Li, Wei Yang
Summary: Heterogeneous nucleation can be inhibited by utilizing hydrogel coatings, which isolate solid surfaces and water. These coatings raise the boiling temperature of water and reduce cavitation pressure on solid surfaces, making them promising for innovation in heat transfer and fluidic systems.
Article
Neurosciences
Biswadeep Chakraborty, Saibal Mukhopadhyay
Summary: This paper proposes a heterogeneous recurrent spiking neural network (HRSNN) with unsupervised learning for spatio-temporal classification of video activity recognition tasks. With the novel unsupervised HRSNN model, the accuracy for the KTH dataset reaches 94.32%, 79.58% and 77.53% for the UCF11 and UCF101 datasets respectively, and 96.54% for the event-based DVS Gesture dataset. The key innovation of HRSNN lies in its recurrent layer, which consists of heterogeneous neurons with varying firing/relaxation dynamics and is trained via heterogeneous spike-time-dependent-plasticity (STDP) with varying learning dynamics for each synapse. The experimental results demonstrate that this novel combination of heterogeneity in architecture and learning method outperforms current homogeneous spiking neural networks.
FRONTIERS IN NEUROSCIENCE
(2023)
Article
Neurosciences
Joseph D. Zak, Nathan E. Schoppa
Summary: The study reveals the presence of various cell types within the olfactory bulb, which play crucial roles in processing and modulating olfactory information through different chemical synaptic connections, highlighting the importance of local circuits in shaping olfactory function.
Article
Engineering, Electrical & Electronic
Amin Faraji, Mostafa Noohi, Sayed Alireza Sadrossadat, Ali Mirvakili, Weicong Na, Feng Feng
Summary: The article introduces the application of batch normalization to deep RNN for faster training and improved model accuracy. By modifying the distribution of internal nodes, training speed can be increased, internal covariance shift can be reduced, and the training of deep neural networks can be accelerated.
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES
(2022)
Article
Biology
Stefanie Engert, Gabriella R. Sterne, Davi D. Bock, Kristin Scott
Summary: This study reveals the structure and connectivity of gustatory sensory neurons in fruit flies, identifies distinct neuronal groups involved in recognizing different taste modalities, and uncovers the impact of synaptic connections on gustatory signal transmission.
Article
Biology
Aylesse Sordillo, Cornelia Bargmann
Summary: In this study, the role of RIM interneurons in C. elegans locomotion was uncovered, showing that RIM acts differently in different states, lengthening reversals when depolarized through glutamate and tyramine neurotransmitters and lengthening forward runs when hyperpolarized through its gap junctions. Additionally, the combined outputs of chemical synapses and gap junctions from RIM regulate forward-to-reversal transitions. Multiple classes of RIM synapses create behavioral inertia during spontaneous locomotion.
Article
Biochemical Research Methods
Max Buhlan, Dmitrij Ljaschenko, Nicole Scholz, Tobias Langenhan
Summary: The study of how mechanical forces affect biological events is crucial for understanding physiological and pathophysiological phenomena. However, there is a lack of knowledge about force parameters, inadequate experimental administration of force stimuli, and a lack of noninvasive means to record their effects. This study introduces a procedure to study the impact of force stimulation on adhesion G-protein-coupled receptor dissociation in mechanosensory neurons. The procedure utilizes the mechanical force spectrum during the natural flexion-extension cycle of the femorotibial joint of fruit flies, and allows for noninvasive observation of transgenic mechanosensitive reporters before, during, and after mechanical stimulation.
Article
Engineering, Electrical & Electronic
Amin Faraji, Sayed Alireza Sadrossadat, Weicong Na, Feng Feng, Qi-Jun Zhang
Summary: This paper introduces the use of deep gated recurrent unit (Deep GRU) as a novel macromodeling approach for nonlinear circuits. The GRU, similar to LSTM, has gating units that control information flow and address the vanishing gradient problem. The proposed method outperforms the conventional LSTM macromodeling method in terms of accuracy and parameter efficiency, and the application of Gaussian dropout on deep GRU further improves its performance and reduces overfitting. The models obtained from the proposed method are also significantly faster than transistor-level models.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2023)
Article
Materials Science, Multidisciplinary
M. Worthington, H. B. Chew
Summary: In this paper, the ability of artificial neural networks (ANNs) for machine learning to predict crack paths and provide insights into crack growth mechanics is assessed. The results show that ANN accurately captures the process zone size, crack growth sequence, and resulting crack patterns. Furthermore, ANN is capable of predicting stochastic crack growth and providing a quantified likelihood of each path.
JOURNAL OF THE MECHANICS AND PHYSICS OF SOLIDS
(2023)