https://github.com/vita-group/l2o-training-techniques

[NeurIPS 2020 Spotlight Oral] "Training Stronger Baselines for Learning to Optimize", Tianlong Chen*, Weiyi Zhang*, Jingyang Zhou, Shiyu Chang, Sijia Liu, Lisa Amini, Zhangyang Wang
https://github.com/vita-group/l2o-training-techniques

curriculum-learning imitation-learning learning-to-learn learning-to-optimize meta-learning self-improving training-tricks

Last synced: 2 months ago
JSON representation

[NeurIPS 2020 Spotlight Oral] "Training Stronger Baselines for Learning to Optimize", Tianlong Chen*, Weiyi Zhang*, Jingyang Zhou, Shiyu Chang, Sijia Liu, Lisa Amini, Zhangyang Wang

• Host: GitHub
• URL: https://github.com/vita-group/l2o-training-techniques
• Owner: VITA-Group
• License: mit
• Created: 2020-09-17T17:39:28.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2021-12-30T08:46:56.000Z (over 3 years ago)
• Last Synced: 2025-03-29T09:51:12.135Z (3 months ago)
• Topics: curriculum-learning, imitation-learning, learning-to-learn, learning-to-optimize, meta-learning, self-improving, training-tricks
• Language: Python
• Homepage: https://tianlong-chen.github.io/about/
• Size: 9.01 MB
• Stars: 26
• Watchers: 2
• Forks: 7
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Training Stronger Baselines for Learning to Optimize

[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)

Code for this paper [Training Stronger Baselines for Learning to Optimize]().

Tianlong Chen*, Weiyi Zhang*, Jingyang Zhou, Shiyu Chang, Sijia Liu, Lisa Amini, Zhangyang Wang

## Overview

With many efforts devoted to designing more sophisticated **L2O models**, we argue for another orthogonal, under-explored theme: the **training techniques** for those L2O models. We show that even **the simplest L2O model could have been trained much better**.

- **Curriculum Learning**

  We first present a progressive training scheme to gradually increase the optimizer unroll length, to mitigate a well-known L2O dilemma of truncation bias (shorter unrolling) versus gradient explosion (longer unrolling).

- **Imitation Learning**

  We further leverage off-policy imitation learning to guide the L2O learning , by taking reference to the behavior of analytical optimizers. 

Our improved training techniques are plugged into a variety of state-of-the-art L2O models, and immediately boost their performance, **without making any change to their model structures.** 

## Experiment Results

### Training the L2O-DM baseline to surpass the state-of-the-art

![](./Figs/L2O-DM.png)

### Training state-of-the-art L2O models to boost more performance

![](./Figs/L2O-Scale.png)

### Ablation study of our proposed techniques

![](./Figs/Ablation.png)

### Imitation Learning v.s. Self-Improving



## Reproduce Details

Experimental details on L2O-DM and RNNProp are refer to this [README](https://github.com/Tianlong-Chen/L2O-Training-Techniques/blob/master/L2O-DM%20%26%20RNNProp/README.md).

Experimental details on L2O-Scale are refer to this [README](https://github.com/Tianlong-Chen/L2O-Training-Techniques/blob/master/L2O-Scale/README.md).

## Citation

If you use this code for your research, please cite our paper:

```

```