ИИ Что?

#08 Summary. Masked Autoencoder is all you need for any modality. #deeplearning#ml ⚡️Briefly To solve complicated tasks machine learning algorithm should understand data and extract good features from it. Usually training generalizing models requires a lot of annotated data. However it is expensive and in some cases impossible. Masked Autoencoder technique allows to train model on unlabeled data and obtain surprisingly good feature representation for all common modalities. 🔎 Contents: - Explanation of MAE approach. - Recipe for all domains. - Crazy experimental results for all type of data. 👉Summary [ link ] Papers: BERT : text MAE : image M3MAE : image + text MAE that listen : audio spectrograms VideoMAE : video Looking forward to your comments and suggestions! Next time - Mind blowing paper from Meta AI about speech reconstruction from noninvasive brain signals. 🔥🔥🔥