Abstract: Choice of an airline ticket depends not only on the attributes of the ticket, but also on other options available to the customer at the time of purchase. Traditional models in airline demand forecasting assume that the demand for different fareclasses is independent. These models predict the volume of demand for each fareclass of a flight in a booking horizon. Dependent demand forecasting models relax this assumption and predict the choice of a customer based on available options. We develop Joint forecasting models, which couple the prediction of demand volume and customer choice behaviour. Booking-curve and seasonality are considered to account for demand volume. Maximum-willingness-to-pay of the customer and the attributes of available products are used to model the customer choice behavior. The probability of a product being chosen is described by applying discrete-choice methodsWe also develop a method to estimate the parameters of the forecasting models. Parameters of demand volume are estimated using non-linear regression using Levenberg-Marquardt algorithm. Simulation is used to simultaneously estimate the parameters of maximum-willingess-to-pay and choice-logit parameters. This method combines the advantage of using closed-form methods wherever possible and applying simulation methods where closed-form methods are not available. This research is carried out jointly with SABRE Inc. Data generated by APOS∗ , using real airline data is used for estimation.
Abstract: Subsidy programs are offered by government and international agencies to improve affordability and accessibility of food and health products for socially deprived community. Although the donors in such programs allocate substantial resources to fund subsidies, a lot of it is wasted due to the inefficiency in the system arising out of product shrinkage and misplacement. This study analyzes the impact of private participation and advanced technology like RFID adoption on the donor subsidy under target consumption level. We formulate the problem as a donor funding the subsidy program through for-profit/not-for-profit newsvendor and compare the equivalent subsidy per consumption with and without RFID. We perform numerical analysis, collecting data from the public distribution system of India, and the results indicate that, unless the for-profit firm operates under a substantially reduced level of shrinkage and misplacement, the donor should always prefer a not-for-profit firm for program implementation. We also observe that among all the scenarios, a not-for-profit firm with advanced technology like RFID requires minimum donor subsidy to generate the target expected consumption.
Abstract: I am going to talk about how problem structures can be exploited to obtain tight reformulations and better linearizations in linearization-based algorithms for solving convex MINLPs. I am going to present automatic detection of these structures in an open source framework, MINOTAUR, for solving convex MINLP and its effectiveness in the performance of the LP/NLP based branch and bound method.
Abstract: With the data coming from various unreliable sources, researchers are presented with the problem of developing robust algorithms or modifying the working of existing algorithms so that they become label noise robust. If, in addition to the label noise, there is differential cost for misclassification, the problem becomes more difficult. We propose two solutions to the problem of cost sensitive binary classification in the presence of uniform label noise. Unlike existing work, our algorithms are better because they do not require the knowledge of noise rates.
Abstract - "Kidney exchange program (KEP) is developed to find optimal exchanges with a registry of incompatible donor-recipient pairs who can swap their donors to solve compatibility issues. International kidney exchange program is one such collaboration where countries could merge their registries to increase the possible number of transplants. There could be instances where a country might have a disadvantage in the merged registry. To avoid countries back out chances, we have proposed a credit based mechanism for international KEP. Simulation results suggest that if credits are fairly defined then all countries will benefit over long runs."
Abstract: Combinatorial game theory (CGT) is a branch of mathematics and theoretical computer science that typically studies sequential games with perfect information. We show the optimal play in a game called Gambling tic-tac-toe. We show various results based on the bid in the first and the second round. We show that the game ends in a draw with the best play. We also see the minimum difference between the budget of the players that the game has a result.