User data has been used by many companies to understand user behaviors and finding new business strategies. However, common techniques cannot be used when it comes to new products that have not yet been released due to the fact that there are no prior data available. In this work, we propose a framework for generating realistic user data on new products which can then be analyzed for insights. Our model uses Conditional Generative Adversarial Network (CGAN) with the Straight-Through Gumbel estimator which can also handle discrete-valued outputs. The CGAN is conditioned on product features learned using a recommendation system which can better capture the relationship between products. Experiments using a dataset consisting of view logs from a real estate listing website shows that our model outperforms other baselines on four performance metrics, and can effectively predict the finer characteristics of new products. INDEX TERMS Generative adversarial networks, deep learning, generative model, data generation, Gumbel-softmax trick, product embedding.
Abstract.A number system that is well-designed can affect the computational time and the hardware implementation. An interesting number system called Round-to-Nearest coding (RN-coding) was proposed to reduce a time consuming in a rounding process. Rounding to the nearest in RN-coding can be done using only truncation at any positions in a sequence of digits (representation). This concept can save a lot of time in a parallel or pipeline computation manner. However, an RN-coding does not support an on-line arithmetic computation. In this paper, we propose a rational digit number system which is composed of rational signed-digits in the digit set. This new system preserves a roundto-nearest property and is suitable for an on-line arithmetic computation. Performing online elementary arithmetic operations in our system can be done by an on-line digit set conversion algorithm. We show that our new algorithm, which is an improvement of an on-line addition algorithm in our previous work, can be demonstrated by an on-line finite automaton with a finite on-line delay .
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.