Samantha-Syed Khairunnesa has successfully passed the PhD preliminary exam. The proposal abstract:

``Uncertainty is an integral part of ML programs, and, it is to be noted that reasoning about ML programs deterministically may not be able to capture the uncertainty that comes with some ML APIs. Besides having a mechanism to quantify uncertainty in ML APIs as required, studies show that a large number of ML bug fixes involve changing the API used, or, order of the APIs used and should be regarded as another area worth considering when it comes to reasoning about ML programs. These bugs makes DNN programs more prone to introducing additional bugs and root cause and behavior of these new bugs often differs from the original. Although such bugs can be avoided through enabling behavioral and/or temporal contracts, to capture the entirety of such contracts require encapsulating ML-related meta information such as label space, input layer etc. and then associate with respective contracts. As ML is becoming a goto technique in various safety and non-safety applications, correctness of ML programs has become a requirement. Presence of design, inference and benchmark of ML contracts can help in this regard to avoid ML specific bugs and lead to better API usage and understanding.’’