The repository contains 4 notebooks that correspond to each module nedded for this assignment from AIDL program:
-
NLP: AIDL_Q&A is BERT large model fine-tuend on SQUAD V1 dataset. The aim of this notebook is create a document Q&A intelligent system. The libraries used are Huggingface Transformers, Spacy for text similarity evaluation, gTTs in order to Q&A system speaks, and Apache Tika for structure the text. The framework of the code is Pytorch.
-
Computer Vision: U-net based architecture is sued in order to solve the problem of drawing restoration. The dataset used, was Da Vinci drawings. The framework was Keras on on top of Tensorflow, CV2 and PIL packages in order to deploy the images.
-
Speech enhacement: the notebook looks for explore or benchmark the performance of two tasks (noise cleaning and audio classification) considering two different input formats (1D raw .wav VS 2D melSpectorgram). The code is based on class lab using pytorch framework.
-
Reinforcement learning: Continuous control agent was developed, having on mind class lab only covers discrete control. The code is based on class lab using pytorch framework.