Hi! This repo represents my way of studying DeepLearning

My LinkedIn:)

---

Resources that I use:

• Technotrack 2021
• HSE DeepLearning
• MIPT DeepLearningSchool
• Tinkoff course
• Dive into Deep Learning
• Other guides

---

Plans:

• Finish technotrack's lectures
• Make DeTR-like decoder for BERT encoder (multilable classification task)
• Understand NVIDIA-Dali (not at all, mix of shader-code and python makes it's work unstable and confusing)
• Look throught the Tinkoff's lectures
• Refactor cloudiness estimation project
• Learn how does Transformers work
• Learn how does GAN work
• CV trick (Squeeze&Excitation, DepthwiseSeparateConv and so on)
• Autograd tricks

---

Building blocks:

• Knoledge distillation, pruning
• Transformers
• GAN
• Autoencoders
• Metric learning

---

Pet projects:

• BERToDeTR for multilable classification (BERT as backbone)
• Cloudiness estimation (almost done, some minimal tests left)
• FaceGeneration (there are some basic results)
• DeTR for detection and segmentation
• Judging trampoline acrobatics (commertial one, there is video as a proof)
• Face recognition

---

Technologies used:

• PyTorch-Lightning
• Wandb
• Connectome
• NVIDIA-Dale
• Composer
• Torch-Pruning (a bit familiar)
• transformers, datasets (Huggingface)