Seismic attributes provides the useful aid to extract the hidden information from the seismic data. Seismic interpretation using single or primary attribute have two major pitfalls; 1) it may not be successful in discerning the target of interest (e.g., salt body, fault network, gas chimney, reservoir etc.), and 2) it may only delineate partial information of the target. Therefore, it is essential to observe the target using multi-attributes, as it will provide the accurate spatial extent of the target body.

Deep neural network (DNN) has the ability to establish a nonlinear relationship between the input and output. Since, they learned relation is based on data and not encompass any assumptions or approximations, therefore it requires labelled data set to train the machine, which led to data-driven predictions. However, applying DNN in seismic pose a challenge of very limited number of labelled data set/ground truth points. Hence, I would be interested to address this challenge by employing more advanced learning schemes

Seismic inversion is an important process to convert the acoustic response into sub-surface elastic properties. Since, the seismic inversion is a nonlinear problem henceforth has a multi-modal objective function. However, global optimization schemes are needed to attain optimal solution. Although, the existing global optimization techniques are computationally expensive and doesn't support parallel computing. Hence, it would be curious to apply advanced global optimizes in seismic inversion and analyzing their performance.