https://github.com/gaussalgo/l2l_mlprague23

Materials for the Learning to Learn workshop on Machine Learning Prague 2023.
https://github.com/gaussalgo/l2l_mlprague23

Last synced: 16 days ago
JSON representation

Materials for the Learning to Learn workshop on Machine Learning Prague 2023.

• Host: GitHub
• URL: https://github.com/gaussalgo/l2l_mlprague23
• Owner: gaussalgo
• License: mit
• Created: 2023-05-12T08:07:41.000Z (almost 2 years ago)
• Default Branch: main
• Last Pushed: 2023-06-08T12:46:33.000Z (almost 2 years ago)
• Last Synced: 2025-03-29T02:04:30.452Z (about 1 month ago)
• Language: Jupyter Notebook
• Size: 5.04 MB
• Stars: 6
• Watchers: 3
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Learning to Learn workshop at MLPrague 2023

This repo contains link to the materials for the Learning to Learn workshop on Machine Learning Prague 2023.

## Getting started

The primary playground env for the exercises below is [Google Colab](https://colab.research.google.com). 

The linked Colab notebooks contain the resolution of dependences, but if you'd like to run the exercises elsewhere, simply install the attached `requirements.txt` into any environment:

```shell

git clone https://github.com/gaussalgo/L2L_MLPrague23.git

pip install -r L2L_MLPrague23/requirements.txt

```

## Outline

### 1. Intro to Transformers

[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/gaussalgo/L2L_MLPrague23/blob/main/notebooks/transformers_intro.ipynb)

- Architectures

  - Difference to other arch's (attention layer)

  - Tasks (=objectives)

- Pre-training & Fine-tuning

- Inputs and outputs

  - Single token prediction

- Generation

  - Iterative prediction

  - Other generation strategies

  - [Hands-on] Constraining generated output (forcing & disabling)

### 2. In-context Learning and Few-shot Learning with Transformers

[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/gaussalgo/L2L_MLPrague23/blob/main/notebooks/ICL_intro.ipynb)

- Problem definition (usage)

- Contrast with Supervised ML

- Zero-shot vs few-shot

  - Examples

  - [Hands-on] comparison of zero-shot vs. few-shot performance (of some chosen ICL)

### 3. Methods for Improving ICL

#### Inference

- Demonstrations heterogeneity

- Prompt engineering

  - Promptsource - database of prompts?

  - [Hands-on] prompt engineering (inspired by the training data?)

#### Training Strategies

[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/gaussalgo/L2L_MLPrague23/blob/main/notebooks/existing_ICL_models.ipynb)

- Training strategies + existing models

  - training in explicit fewshot format (QA)

  - Instruction tuning

  - Multitask learning

  - Chain-of-Thought

  - Pre-training on a code

  - Fine-tuning with human feedback

#### Theory behind - why does ICL exist?

  - Data properties fostering ICL

  - Experiments

  - Explanations of the existing models?

### 4. Hands-on in Improving Few-shot ICL

[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/gaussalgo/L2L_MLPrague23/blob/main/notebooks/hands_on_improving_ICL.ipynb)

- [Hands-on] Customizing Few-shot ICL to specialized data

- Practical training pipeline

  - Overview of the training pipeline

  - Adaptor example

-------

### Models evaluation & competition [Optional]

If you trained your own great few-shot ICL model, it would be a pity not to test it on some unseen reasoning tasks.

See the [competition readme](competition) for how to evaluate the model and if it beats the baseline, how to spread the word!

-------