Deep Neural Network
Deep neural networks (DNNs) are complex computational models aiming to mimic the human brain's learning capabilities, primarily focusing on achieving high accuracy and efficiency in various tasks. Current research emphasizes understanding DNN training dynamics, including phenomena like neural collapse and the impact of architectural choices (e.g., convolutional, transformer, and operator networks) and training strategies (e.g., weight decay, knowledge distillation, active learning). This understanding is crucial for improving DNN performance, robustness (including against adversarial attacks and noisy data), and resource efficiency in diverse applications ranging from image recognition and natural language processing to scientific modeling and edge computing.
Papers
ProAct: Progressive Training for Hybrid Clipped Activation Function to Enhance Resilience of DNNs
Seyedhamidreza Mousavi, Mohammad Hasan Ahmadilivani, Jaan Raik, Maksim Jenihhin, Masoud Daneshtalab
Is Value Functions Estimation with Classification Plug-and-play for Offline Reinforcement Learning?
Denis Tarasov, Kirill Brilliantov, Dmitrii Kharlapenko
VS-PINN: A fast and efficient training of physics-informed neural networks using variable-scaling methods for solving PDEs with stiff behavior
Seungchan Ko, Sang Hyeon Park
W-Net: One-Shot Arbitrary-Style Chinese Character Generation with Deep Neural Networks
Haochuan Jiang, Guanyu Yang, Kaizhu Huang, Rui Zhang
Contextual fusion enhances robustness to image blurring
Shruti Joshi, Aiswarya Akumalla, Seth Haney, Maxim Bazhenov
Compositional Curvature Bounds for Deep Neural Networks
Taha Entesari, Sina Sharifi, Mahyar Fazlyab
PolyLUT-Add: FPGA-based LUT Inference with Wide Inputs
Binglei Lou, Richard Rademacher, David Boland, Philip H.W. Leong
Crafting Heavy-Tails in Weight Matrix Spectrum without Gradient Noise
Vignesh Kothapalli, Tianyu Pang, Shenyang Deng, Zongmin Liu, Yaoqing Yang
Error Bounds of Supervised Classification from Information-Theoretic Perspective
Binchuan Qi
From Feature Visualization to Visual Circuits: Effect of Adversarial Model Manipulation
Geraldin Nanfack, Michael Eickenberg, Eugene Belilovsky
Automatic Input Feature Relevance via Spectral Neural Networks
Lorenzo Chicchi, Lorenzo Buffoni, Diego Febbe, Lorenzo Giambagli, Raffaele Marino, Duccio Fanelli