Rina Azoulay

Truthful and efficient mechanisms for Website dependent advertising auctions

Recently we have witnessed the phenomena of private Website owners who are willing to dedicate space on their Website for advertisements managed by leading search engines (e.g., Google, Bing and Yahoo). In most cases the choice of the advertisements displayed on a certain Website is made without taking into account the preferences of the Website hosting the advertisements. This causes the Website that dedicate space for the ads to be exposed to ads that are irrelevant to the content on the Website. In this paper, we focus on the design of auction protocols that take into account the Website owners preferences by associating him with a utility function reflecting the value for each ad shown in his space. In doing so we introduce the dispensation scheme into the second price sealed bid auction for both the single and the multi-slot cases. For the single-slot, we propose a mechanism that proved to be individual rational, truthful, and a free advertisement condition is defined. For the multi-slot case, we propose three truthful mechanisms. Simulation results show that the allocation efficiency of two of the proposed mechanisms maintain a near to optimal allocation efficiency compared to optimal allocation calculated using the Hungarian mechanism, while substantially reducing computational complexity from cubic to linear.

Map of nonprofit organization websites in Israel

In this study, we consider the structure and linking strategy of Hebrew websites of several nonprofit organizations. Because nonprofit organizations differ from commercial, educational, or governmental sectors, it is important to understand the ways they utilize the web. To the best of our knowledge, the linking structure of nonprofit organizations has not been previously studied. We surveyed websites of 54 nonprofit organizations in Israel; most of these sites have at least 100 volunteers. We compared their orientation and contents and we built their linking map. We divided the organizations into four main groups: economic aid and citizen rights organizations, health aid organizations, organizations supporting families and individuals with special needs, and organizations for women and children. We found that the number of links inside the special needs group is much higher than in the other groups. We tried to explain this behavior by considering the data obtained from the site-linking graph. The value of our results is in defining and testing a method to investigate a group of nonprofit organizations, using a case study of Israeli organizations.

Adaptation Schemes for Question's Level to be Proposed by Intelligent Tutoring Systems

The main challenge in developing a good Intelligent Tutoring System (ITS) is, surely, the question of how to adapt the hardness level of questions and tasks to the current students capabilities, assuming these are dynamically changing over the tutoring period. According to state of art, most ITS systems make use of the Q-learning algorithm for this adaptation task. Our paper presents innovative results that compare performances of several methods not applied before for ITS to handle the above challenge. In particular as far as we know this is the first attempt to apply the Bayesian inference algorithm for question level matching in ITS. Next, as this is a groundwork research done to identify the best adaptation scheme, we propose to use an artificial environment with simulated students for the evaluation phase. The results were benchmarked with the optimal performance of the system assuming the user model (abilities) is completely known to the ITS. The results show that the method that outperforms the other is based on a Bayesian Inference which achieves more than 90% of the optimal performance. Our conclusion is that it may be worthwhile to integrate Bayesian inference based algorithms to adapt to students level in the ITS. Future work is required in to confront these empirical results with those of real students.

Efficient bidding strategies for Cliff-Edge problems

In this paper, we propose an efficient agent for competing in Cliff-Edge (CE) and simultaneous Cliff-Edge (SCE) situations. In CE interactions, which include common interactions such as sealed-bid auctions, dynamic pricing and the ultimatum game (UG), the probability of success decreases monotonically as the reward for success increases. This trade-off exists also in SCE interactions, which include simultaneous auctions and various multi-player ultimatum games, where the agent has to decide about more than one offer or bid simultaneously. Our agent competes repeatedly in one-shot interactions, each time against different human opponents. The agent learns the general pattern of the population’s behavior, and its performance is evaluated based on all of the interactions in which it participates. We propose a generic approach which may help the agent compete against unknown opponents in different environments where CE and SCE interactions exist, where the agent has a relatively large number of alternatives and where its achievements in the first several dozen interactions are important. The underlying mechanism we propose for CE interactions is a new meta-algorithm, deviated virtual learning (DVL), which extends existing methods to efficiently cope with environments comprising a large number of alternative decisions at each decision point. Another competitive approach is the Bayesian approach, which learns the opponents’ statistical distribution, given prior knowledge about the type of distribution. For the SCE, we propose the simultaneous deviated virtual reinforcement learning algorithm (SDVRL), the segmentation meta-algorithm as a method for extending different basic algorithms, and a heuristic called fixed success probabilities (FSP). Experiments comparing the performance of the proposed algorithms with algorithms taken from the literature, as well as other intuitive meta-algorithms, reveal superiority of the proposed algorithms in average payoff and stability as well as in accuracy in converging to the optimal action, both in CE and SCE problems.

Machine learning methods for SIR prediction in cellular networks

Abstract Accurate assessment of the wireless coverage of a station is considered a key feature in 5G networks. Determining the reception coverage of transmitters becomes a complicated problem when there are interfering transmitters, and it becomes increasingly more complicated when the transmission powers of those transmitters are not uniform. In this paper, we compare different Machine Learning techniques that can be used to predict the wireless coverage maps. We consider the following Machine Learning methods: (1) Radial Basis Network; a type of Artificial Neural Network which typically uses Gaussian kernels, (2) an Artificial Neural Network which uses a sigmoid function as an activator,(3) A Multi-Layer Perceptron with two hidden layers, and (4) the K-Nearest-Neighbors technique. We show how it is possible to train the Neural Networks to generate coverage maps based on samples and we check the accuracy level of the learning process on a test set, using these four different learning techniques. The conclusion of our experiments is that if the sample points are randomly located, the Radial Basis Network and the Multi-Layer Perceptron perform better than the other methods. Thus, these models can be considered promising candidates for learning coverage maps, and can be used for efficient spectrum management within the framework of 5G cellular networks.