ICML 2018 accepted papers on Architectures and efficiency
59 of the 623 papers accepted at ICML 2018 (9.5%) are on Architectures and efficiency, compared with 10.8% at ICML 2017. They fall into 5 topics.
See ICML 2018 in the interactive map
Submitting to ICML? Checklist →
Working on something in this area? Describe your idea in scime atlas to see the closest papers by meaning and how crowded the spot has become.
Topics and papers
RNNs and normalization · 43 papers
- GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks (450 citations)
- Adafactor: Adaptive Learning Rates with Sublinear Memory Cost (162 citations)
- Beyond Finite Layer Neural Networks: Bridging Deep Architectures and Numerical Differential Equations (153 citations)
- On the Power of Over-parametrization in Neural Networks with Quadratic Activation (152 citations)
- Spurious Local Minima are Common in Two-Layer ReLU Neural Networks (143 citations)
- Extracting Automata from Recurrent Neural Networks Using Queries and Counterexamples (124 citations)
- Rapid Adaptation with Conditionally Shifted Neurons (121 citations)
- Dynamic Evaluation of Neural Sequence Models (98 citations)
- Coded Sparse Matrix Multiplication (97 citations)
- Learning Longer-term Dependencies in RNNs with Auxiliary Losses (87 citations)
- Learning Memory Access Patterns (78 citations)
- Dynamical Isometry and a Mean Field Theory of RNNs: Gating Enables Signal Propagation in Recurrent Neural Networks (72 citations)
and 31 more in the interactive map.
CNN design · 10 papers
- Deep k-Means: Re-Training and Parameter Sharing with Harder Cluster Assignments for Compressing Deep Convolutions (91 citations)
- A Theoretical Explanation for Perplexing Behaviors of Backpropagation-based Visualizations (54 citations)
- Optimization Landscape and Expressivity of Deep CNNs (51 citations)
- Learning One Convolutional Layer with Overlapping Patches (43 citations)
- High Performance Zero-Memory Overhead Direct Convolutions (42 citations)
- DCFNet: Deep Neural Network with Decomposed Convolutional Filters (36 citations)
- Dynamical Isometry and a Mean Field Theory of CNNs: How to Train 10,000-Layer Vanilla Convolutional Neural Networks (36 citations)
- Geodesic Convolutional Shape Optimization (34 citations)
- Gradually Updated Neural Networks for Large-Scale Image Recognition (3 citations)
- WSNet: Compact and Efficient Networks Through Weight Sampling (3 citations)
Neural architecture search · 3 papers
- Efficient Neural Architecture Search via Parameters Sharing (1,269 citations)
- Understanding and Simplifying One-Shot Architecture Search (528 citations)
- Path-Level Network Transformation for Efficient Architecture Search (150 citations)
