We consider the use of Bayesian topic models in the analysis of computer network traffic. Our approach utilizes latent Dirichlet allocation and time-varying dynamic latent Dirichlet allocation, with the goal of identifying significant co-occurrences of types of network traffic, these forming topics of user behavior. In our experiments, these topics of user behavior included: (i) web traffic, (ii) email client and instant messaging, (iii) Microsoft file access, (iv) email server, and (v) other miscellaneous traffic. Each identified behavior topic included a variety of different, but related, protocols without using any a priori knowledge of the purpose of the protocol. We believe that the techniques presented in this paper can be used to form more complex topics through the use of deep packet inspection, and that such topic models could prove useful in the identification of zero-day exploits or other network threats.

Sources of real-time traffic are generally highly unpredictable with respect to the instantaneous and average load which they create. Yet such sources will provide a significant portion of traffic in future networks, and will significantly affect the overall performance of and quality of service. Clearly high levels of compression are desirable as long as video quality remains satisfactory, and our research addresses this key issue with a novel learning-based approach. We propose the use of neural networks (NNs) as post-processors for any existing video compression scheme. The approach is to interpolate video sequences and compensate for frames which may have been lost or deliberately dropped. We show that deliberately dropping frames will significantly reduce the amount of offered traffic in the network, and hence the cell loss probability and network congestion, while the NN post-processor will preserve most of the desired video quality. Dropping frames at the sender or in the network is also a fast way to react to network overload and reduce congestion. Our interpolation techniques at the receiver, including neural network-based algorithms, provide output frame rates which are identical to (or possibly higher than) the original video sequence's frame rate. The resulting video quality is essentially equivalent to the sequence without frame drops, despite the loss of a significant fraction of the frames. Experimental evaluation using real video sequences is provided or interpolation with a connectionist NN using the backpropagation learning algorithm, the random NN (RNN) in a feed-forward configuration with its associated learning algorithm, and cubic spline interpolation. The experiments show that when more frames are being dropped or lost, the RNN performs generally better than the other techniques in terms of resulting video quality and overall performance. When the fraction of dropped frames is small, cubic splines offer better performance. Experimental data shows that this receiver-reconstructed subsampling technique significantly reduces the cell loss rates in an asynchronous transfer mode switch for different buffer sizes and service rates.

In this paper we describe a novel neural network technique for video compression, using a "point-process" type neural network model we have developed which is closer to biophysical reality and is mathematically much more tractable than standard models. Our algorithm uses an adaptive approach based upon the users' desired video quality Q, and achieves compression ratios of up to 500:1 for moving gray-scale images, based on a combination of motion detection, compression, and temporal subsampling of frames. This leads to a compression ratio of over 1000:1 for full-color video sequences with the addition of the standard 4:1:1 spatial subsampling ratios in the chrominance images. The signal-to-noise ratio ranges from 29 dB to over 34 dB. Compression is performed using a combination of motion detection, neural networks, and temporal subsampling of frames. A set of neural networks is used to adaptively select the desired compression of each picture block as a function of the reconstruction quality. The motion detection process separates out regions of the frame which need to be retransmitted. Temporal subsampling of frames, along with reconstruction techniques, lead to the high compression ratios.