NLP for Time Series Forecasting - overview
Time series forecasting using machine learning (ML) algorithms have widespread applications across multiple domains. ML models for time series forecasting are usually based on numerical and categorical features that are strongly correlated to the target variable and these are engineered using historical data. However, while scoring these models, there are scenarios where the features of the model are inaccurately measured or prediction windows where the assumptions of the trained model breakdown. Performance of the ML models can be strongly affected in such scenarios and predicting these situations apriori remains a challenge.
In this project, we will look at Natural Language Processing (NLP) algorithms to predict when the performance of the ML models would be strongly impacted. The project will be application driven with emphasis on developing fundamental NLP models for time series forecasting. These models will be eventually deployed in a live system for real-time forecasting.
The end goal of the internship is a research prototype and submissions in top AI conferences (AAAI/IJCAI/ACL/EMNLP/NAACL).
Strong programming skills in Python.
Experience in ML algorithms for time series forecasting or NLP.
Good communication skills.