Pattern Exploiting Training for NLP

Pattern Exploiting Training for NLP
GPT-3 created quite a buzz when it was first launched a couple of months back for its ability to generate great good amount of text to create a more realistic feel, but it suffered from one particular shortcoming that its simply too big. There has been an increasing move towards creating more light weight implementations in NLP,CV and othe areas for machine learning and artificial intelligence, evident from the fact that most of the papers now along with accuracy also feature inference times for the respective model, a good sign of end use awareness in research. [Read More]

Recommender System made easier with TFRS

Recommender System made easier with TFRS
Creating scalable deep learning-based recommendation systems from scratch is a significantly time-consuming task, especially so that we need to spend significant time on the non-productive elements involved in setting up the model for experiments. It is a given that we will need to perform levels of iterations on the model architecture to get optimal hyperparameters for a particular use case. Google has now launched just the instrument required to reduce the non-productive component in that cycle. [Read More]