ml

Some personal project, kaggle and research, put as is.

Relevant domains treated:

Computer vision

- Classification : Cifar 10 / 100, cassava-leaf-disease

- Few Shot Learning : miniimagenet, CUB 200

- Distillation : Cifar 100

- GANs : Bengali AI , Fashion Product

- Medical : osic-pulmonary

NLP

- Sentence / Pair Classification : transformers, NLP Disaster, SNLI Dataset

- Summarisation : Text summarisation

- Knowledge Extraction : Berts, tweet-sentiment-extraction, sentiment 140 (used for finetuning roberta on tweets)

- text generation : transformers (IFart), research-abstract (generation of paper abstract), IFart (small french GPT)

Audio Processing

- Classification : birdsong-recognition

- Generation : midi data (attempt at music generation through midi files)

Time series

- Knowledge tracing transformers : riid-test-answer

- Knowledge tracing gbdt : data-science-bowl-2019

- Sales predictions : m5-forecasting-accuracy

Tabular Data

- benchmark deep learning (tabnet) vs gbdt on several dataset : titanic

- Few shot learning : instant-gratification

- AutoML benchmark (autokeras, autosklearn, autogluon, tpot, optuna, h2o) : Jane street market

Other stuffs

- abstraction-and-reasoning-challenge : minimalist EDA of the kaggle competition set up by françois chollet

- Gym : some small attempts at reinforcement learning

- Boursorama : scrapping of financial data (stock + articles)