No abstract

Generative Adversarial Networks(GANs) are powerful generative models on numerous tasks and datasets but are also known for their training instability and mode collapse. The latter is because the optimal transportation map is discontinuous, but DNNs can only approximate continuous ones. One way to solve the problem is to introduce multiple discriminators or generators. However, their impacts are limited because the cost function of each component is the same. That is, they are homogeneous. In contrast, multiple discriminators with different cost functions can yield various gradients for the generator, which indicates we can use them to search for more transportation maps in the latent space. Inspired by this, we have proposed a framework to combat the mode collapse problem, containing multiple discriminators with different cost functions, named CES-GAN. Unfortunately, it may also lead to the generator being hard to train because the performance between discriminators is unbalanced, according to the Cannikin Law. Thus, a gradient selecting mechanism is also proposed to pick up proper gradients. We provide mathematical statements to prove our assumptions and conduct extensive experiments to verify the performance. The results show that CES-GAN is lightweight and more effective for fighting against the mode collapse problem than similar works.

In Vehicular Ad Hoc Networks (VANETs), content distribution directly relies on the fleeting and dynamic contacts between moving vehicles, which often leads to prolonged downloading delay and terrible user experience. Deploying Wifibased Access Points (APs) could relieve this problem, but it often requires a large amount of investment, especially at the city scale. In this paper, we propose the idea of ParkCast, which doesn't need investment, but leverages roadside parking to distribute contents in urban VANETs. With wireless device and rechargable battery, parked vehicles can communicate with any vehicles driving through them. Owing to the extensive parking in cities, available resources and contact opportunities for sharing are largely increased. To each road, parked vehicles at roadside are grouped into a line cluster as far as possible, which is locally coordinated for node selection and data transmission. Such a collaborative design paradigm exploits the sequential contacts between moving vehicles and parked ones, implements sequential file transfer, reduces unnecessary messages and collisions, and then expedites content distribution greatly. We investigate ParkCast through theoretic analysis and realistic survey and simulation. The results prove that our scheme achieve high performance in distribution of contents with different sizes, especially in sparse traffic conditions.

No abstract