Publications

• Srivastava et al. (2022) Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models.

  Description: The Beyond the Imitation Game Benchmark (BIG-bench) is a collaborative benchmark intended to probe large language models and extrapolate their future capabilities. The more than 200 tasks included in BIG-bench are summarized by keyword here, and by task name here.

  My contribution: In this task, the language model is asked to find the strongest answer to a question about the structure of a logical argument. For example, the question might ask for an example of a piece of evidence that would weaken the argument’s conclusions, or an assumption that is implicitly being made by the argument. Following the GRE format, questions are presented in a multiple-choice format; the task contains 32 examples. See Jascha Sohl-Dickstein mentioning about my task here

  For more details on the task alongwith it’s performance plots here is the link.

  Links: Paper , Anouncement

• Vakil, V., Pachchigar, S., Chavda, C., Soni, S.(2021) Explainable predictions of different machine learning algorithms used to predict Early Stage diabetes.

  Description: Machine Learning and Artificial Intelligence can be widely used to diagnose chronic diseases so that necessary precautionary treatment can be done in critical time. Diabetes Mellitus which is one of the major diseases can be easily diagnosed by several Machine Learning algorithms. Early stage diagnosis is crucial to prevent dangerous consequences. In this paper we have made a comparative analysis of several machine learning algorithms viz. Random Forest, Decision Tree, Artificial Neural Networks, K Nearest Neighbor, Support Vector Machine, and XGBoost along with feature attribution using SHAP to identify the most important feature in predicting the diabetes on a dataset collected from Sylhet Hospital. As per the experimental results obtained, the Random Forest algorithm has outperformed all the other algorithms with an accuracy of 99 percent on this particular dataset.

  Links: Paper