Ml learning documentation blog

• Stable diffusion experiments
  1. Textual inversion
  2. Dreambooth

---

• Huggingface Deep Reinforcement Learning course

---

• Andrej Karpathy makemore series
  1. Makemore - learning NN from one of the top researchers and engineers in the world

---

• fastai 2022 course
  1. lotr app A cute app that combines language modeling, language summarization and image generation using the stable diffusion model. Here’s a link to a post with more details
  2. fastai pets on huggingface 2022 edition
  3. This blog :)

---

• Kaggle challenges:
  1. Natural Language Processing with Disaster Tweets. A top 17% result with a relatively small model (and a very quick training) but some interesting semi-automatic ML-driven data cleaning
  2. Spaceship titanic challenge. My best submission got me to top 10%
  3. Titanic intro challenge

---

• The huggingface course
  1. Remarqify app on huggingface. This app uses the distilgpt2 model retrained on books by Erich Maria Remarques

---

• Kaggle courses:
  1. Intro
  2. Pandas
  3. Intermediate Ml
  4. Data Visualization
  5. sql
  6. Deep learning intro
  7. Computer vision
  8. Data Cleaning
  9. Geospatial
  10. Explainability

Probably the most interesting for me was the course about explainability since it gives you a basic toolkit to systematically improve upon the results. The data visualisation (and pandas) are also a great for pandas noobs like me.

---

• Coursera’s Andrew Ng’s machine learning specialization. It’s a great step-by-step course with an emphasis on ‘classical’ regression in the beginning.
  1. Introduction to machine learning

---

• Mit deep learning course. A free and quick course with colab notebooks as exercises. One of the exercises included a sequential RNN to generate music based on this course. Here’s what LSTM RNN can generate after a quick training on a few musical files and then fine tuning (transfer learning) to a genre:

  1. Transfer learning to the style of Joplin

  2. Transfer learning to the style of Chopin

Approaching this using transformers would probably yield much better result.

---

• Jeremy’s fast.ai 2019 edition. Followed the video lectures and the textbook. Binder doesn’t work too well, but I saw no reason to migrate the work to huggingface:
  1. Bear classifier example
  2. Multi category example
  3. Pets example

---

Posts

• Jan 30, 2023

  Experiments with stable diffusion - Dreambooth

• Jan 11, 2023

  Experiments with stable diffusion - textual inversion

• Oct 27, 2022

  Papers

• Jun 11, 2022

  Fastai 2022: lesson four exercise and the Lord of the rings app

subscribe via RSS