Machine Learning for Networking Applications:

Graph Attention Networks for Failure Evaluation and Robust Network Design:

Robust network design, which aims to guarantee network availability under various failure scenarios while optimizing performance/cost objectives, has received significant attention. Existing approaches often rely on model-based mixed-integer optimization that is hard to scale or employ deep learning to solve specific engineering problems yet with limited generalizability. In this direction, we show that failure evaluation provides a common kernel to improve the tractability and scalability of existing solutions. By providing a neural network function approximation of this common kernel using graph attention networks, we develop a unified learning-based framework, FERN, for scalable Failure Evaluation and Robust Network design. FERN represents rich problem inputs as a graph and captures both local and global views by attentively performing feature extraction from the graph. It enables a broad range of robust network design problems, including robust network validation, network upgrade optimization, and fault-tolerant traffic engineering, to be recasted with respect to the common kernel and thus computed efficiently using neural networks and over a small set of critical failure scenarios.


Reinforcement Learning for Network Resource Allocation:

Reinforcement Learning has been increasingly used for efficient resource allocation. However, there are multiple challenges in networking applications that make the reinforcement learning tools not applicable. Some examples include multiple agents where the fairness objective causes non-linearity, scaling to large networks leads to efficient solutions splitting the network, constraints in decision making, etc. In this theme, we propose novel algorithms to take these aspects into account for efficient algorithms for traffic engineering, power control, and caching decisions.


Applications of Exploiting Data Structure in Matrix/Tensor for Better Sampling, Fingerprinting, and Estimation:

vaneet_photo vaneet_photo

Exploiting structure in the data across multiple dimensions can lead to efficient techniques for efficient sampling and fingerprinting. The figure alongside shows different network service atributes dependence on time, spatial locations and data service quality measurements thus motivating tensor structure. Further, the map shows the received signal strength in indoor has spatial dependence and the signals from different APs are correlated and thus adaptive sampling can reduce fingerprints needed for localization.


Exploiting Network Structure for Better Scheduling Strategies:

vaneet_photo

Learning the network characteristics is crucial for coming up with next generation network designs. Getting the measurements of the received power from a tower leads to path loss which helps in design of efficient codes. Fingerprinting roads can help find efficient scheduling/handoff decisions for a service provider. Learning attributes at the network helps learn interaction between different layers to achieve better designs. The distribution of interference can lead to better interference management techniques. This understanding will be useful to achieve new and better system designs. The figure alongside depicts a mobile user trajectory along a road through 3 cellular sectors showing interactions with the cellular network and the measured values of channel quality (Ec/Io) during 3 different drives. This structure has been used to novel scheduling strategies with prediction.

Modeling Application QoE from Passive Network Measurements:

vaneet_photo

Video and web applications are the most prominent applications in today's world. It is important to understand how radio network characteristics (such as signal strength, handovers, load, etc.) influence users' Quality-of-Experience (QoE). Understanding the relationship between QoE and network characteristics is a pre-requisite for cellular network operators to detect when and where degraded network conditions actually impact QoE. Unfortunately, cellular network operators do not have access to detailed server-side or client-side logs to directly measure QoE metrics, such as abandonment rate, stalls, and video/session length. We devise a machine-learning-based mechanism to infer web QoE metrics from network traces accurately. The image alongside depicts partial web downloads as a function of network attributes.

Home