Resnet
PulseAugur coverage of Resnet — every cluster mentioning Resnet across labs, papers, and developer communities, ranked by signal.
-
New Model Fusion Technique Improves Zero-Shot Performance
Researchers have developed a new neuron-centric approach to model fusion, addressing challenges posed by representational divergence in independently trained neural networks. This method frames fusion as a representatio…
-
Branch scaling improves ResNet generalization via depth-wise decay
Researchers have theoretically demonstrated that the generalization capabilities of wide residual networks (ResNets) can be improved by scaling factors that decay rapidly with depth. This approach, when combined with ea…
-
Lattice theory provides algebraic framework for deep convolutional networks
Researchers have developed a new algebraic framework for deep convolutional neural networks using lattice theory and mathematical morphology. This approach systematically analyzes standard network layers, revealing that…
-
New Math Framework Explains Transformer Training Dynamics
A new paper introduces a mathematical framework for understanding how Transformers train, particularly in the mean-field regime where both depth and width approach infinity. Unlike ResNets which can be modeled by ODEs, …
-
Neural Feature Dynamics framework offers new insights into deep network training
Researchers have developed a new framework called Neural Feature Dynamics (NFD) to better understand how features evolve during the training of deep neural networks, particularly in the infinite-depth limit. The study f…
-
Layerwise LQR framework optimizes deep networks using geometry-aware control
Researchers have developed Layerwise LQR (LLQR), a new optimization framework for deep learning models. LLQR reformulates second-order optimization methods, like Newton's method, as a linear quadratic regulator problem.…
-
New research reveals implicit bias drives neural scaling laws in deep learning
Researchers have identified two new dynamical scaling laws that describe how neural network performance changes with complexity measures throughout training. These laws, observed across various architectures like CNNs a…
-
Progressive Approximation in Deep Residual Networks: Theory and Validation
Researchers have introduced Layer-wise Progressive Approximation (LPA), a new training principle for deep residual networks. This method reframes residual networks as a layer-by-layer approximation process, demonstratin…