https://github.com/piyush-555/vcl-in-pytorch

PyTorch implementation of Variational Continual Learning
https://github.com/piyush-555/vcl-in-pytorch

bayesian-neural-networks continual-learning lifelong-learning pytorch variational-inference

Last synced: 30 days ago
JSON representation

PyTorch implementation of Variational Continual Learning

• Host: GitHub
• URL: https://github.com/piyush-555/vcl-in-pytorch
• Owner: Piyush-555
• Created: 2021-07-04T05:27:36.000Z (almost 4 years ago)
• Default Branch: main
• Last Pushed: 2021-07-17T07:17:07.000Z (almost 4 years ago)
• Last Synced: 2025-03-26T13:45:47.480Z (about 2 months ago)
• Topics: bayesian-neural-networks, continual-learning, lifelong-learning, pytorch, variational-inference
• Language: Python
• Homepage:
• Size: 819 KB
• Stars: 7
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
## Variational Continual Learning in PyTorch

Re-Implementation of [Variational Continual Learning](https://arxiv.org/abs/1710.10628) by  Nguyen, Li, Bui, and Turner (ICLR 2018).

### SplitMNIST: VCL v/s Baselines

![SplitMNIST Baselines](plots/splitmnist_baselines.png)

### PermutedMNIST: VCL v/s Baselines

![PermutedMNIST Baselines](plots/permutedmnist_baselines.png)

### Vary beta (Weight coefficient for KL term)

Varying beta while keeping memory fixed at 200 samples per task.

![Vary beta SplitMNIST](plots/splitmnist_fixed_memory_vary_beta.png)

![Vary beta PermutedMNIST](plots/permutedmnist_fixed_memory_vary_beta.png)

### Vary Memory

Varying memory while keeping beta fixed at 1.

![Vary Memory SplitMNIST](plots/splitmnist_fixed_beta_vary_memory.png)

![Vary Memory PermutedMNIST](plots/permutedmnist_fixed_beta_vary_memory.png)

### Vary Memory (Fixed prior)

Varying memory while keeping prior fixed to Gaussian distribution.

![Vary Memory SplitMNIST](plots/splitmnist_constant_KL_vary_memory.png)

![Vary Memory PermutedMNIST](plots/permutedmnist_constant_KL_vary_memory.png)