andy
3606944190
Affected files: .obsidian/graph.json .obsidian/workspace-mobile.json .obsidian/workspace.json Food/Meal Plan.md Lab/Linux/Alpine.md Lab/Linux/KDE.md Lab/Scratch Domain.md Lab/Windows/Active Directory.md Languages/Spanish/README.md Languages/Spanish/Spanish.md Money/Accounts.md Money/Monthly/23-04.md Money/Monthly/23-05.md Money/Monthly/23-06.md Projects/Mixonomer.md Projects/NoteCrawler.md Projects/Projects.md Projects/README.md Projects/Selector.md Projects/To Do App.md Projects/img/selector-arch.png STEM/AI/Classification/Supervised/SVM.md STEM/AI/Neural Networks/CNN/FCN/Super-Resolution.md STEM/AI/Neural Networks/CNN/GAN/GAN.md STEM/AI/Neural Networks/CNN/GAN/cGAN.md STEM/AI/Neural Networks/CNN/Interpretation.md STEM/AI/Neural Networks/Deep Learning.md STEM/AI/Neural Networks/MLP/MLP.md STEM/AI/Neural Networks/Properties+Capabilities.md STEM/AI/Neural Networks/RNN/Representation Learning.md STEM/AI/Pattern Matching/Markov/Markov.md STEM/AI/Searching/Informed.md STEM/AI/Searching/README.md STEM/AI/Searching/Searching.md STEM/AI/Searching/Uninformed.md STEM/CS/Languages/Javascript.md STEM/CS/Languages/Python.md STEM/CS/Languages/dotNet.md STEM/CS/Resources.md STEM/IOT/Cyber-Physical Systems.md STEM/IOT/Networking/Networking.md STEM/IOT/Networking/README.md STEM/IOT/Software Services.md STEM/img/cyberphysical-social-data.png STEM/img/cyberphysical-system-types.png STEM/img/cyberphysical-systems.png STEM/img/depth-first-cons.png STEM/img/depth-first.png STEM/img/iot-mesh-network.png STEM/img/iot-network-radar.png STEM/img/iot-network-types 1.png STEM/img/iot-network-types.png STEM/img/markov-start-end-matrix.png STEM/img/markov-start-end-probs.png STEM/img/markov-start-end.png STEM/img/markov-state-duration.png STEM/img/markov-state.png STEM/img/markov-weather.png STEM/img/search-bidirectional.png STEM/img/search-breadth-first.png STEM/img/search-lim-goal.png STEM/img/search-lim1.png STEM/img/search-lim2.png STEM/img/search-lim3-2.png STEM/img/search-lim3.png STEM/img/search-lim4.png STEM/img/searching-graph-tree.png STEM/img/searching-graph.png Work/Freelancing.md |
||
---|---|---|
.. | ||
FCN | ||
GAN | ||
CNN.md | ||
Convolutional Layer.md | ||
Examples.md | ||
Inception Layer.md | ||
Interpretation.md | ||
Max Pooling.md | ||
Normalisation.md | ||
README.md | ||
UpConv.md |
Before 2010s
- Data hungry
- Need lots of training data
- Processing power
- Niche
- No-one cared/knew about CNNs
After
- ImageNet
- 16m images, 1000 classes
- GPUs
- General processing GPUs
- CUDA
- NIPS/ECCV 2012
- Double digit % gain on ImageNet accuracy
Full Connected
- Move from convolutional operations towards vector output
- Stochastic drop-out
- Sub-sample channels and only connect some to dense layers
As a Descriptor
- Most powerful as a deeply learned feature extractor
- Dense classifier at the end isn't fantastic
- Use SVM to classify prior to penultimate layer
Finetuning
- Observations
- Most CNNs have similar weights in conv1
- Most useful CNNs have several conv layers
- Many weights
- Lots of training data
- Training data is hard to get
- Labelling
- Reuse weights from other network
- Freeze weights in first 3-5 conv layers
- Learning rate = 0
- Randomly initialise remaining layers
- Continue with existing weights