Cloud Classification with Unsupervised Deep Learning
Authors:
Takuya Kurihana,
Ian Foster,
Rebecca Willett,
Sydney Jenkins,
Kathryn Koenig,
Ruby Werman,
Ricardo Barros Lourenco,
Casper Neo,
Elisabeth Moyer
Abstract:
We present a framework for cloud characterization that leverages modern unsupervised deep learning technologies. While previous neural network-based cloud classification models have used supervised learning methods, unsupervised learning allows us to avoid restricting the model to artificial categories based on historical cloud classification schemes and enables the discovery of novel, more detail…
▽ More
We present a framework for cloud characterization that leverages modern unsupervised deep learning technologies. While previous neural network-based cloud classification models have used supervised learning methods, unsupervised learning allows us to avoid restricting the model to artificial categories based on historical cloud classification schemes and enables the discovery of novel, more detailed classifications. Our framework learns cloud features directly from radiance data produced by NASA's Moderate Resolution Imaging Spectroradiometer (MODIS) satellite instrument, deriving cloud characteristics from millions of images without relying on pre-defined cloud types during the training process. We present preliminary results showing that our method extracts physically relevant information from radiance data and produces meaningful cloud classes.
△ Less
Submitted 30 September, 2022;
originally announced September 2022.
Running the Network Harder: Connection Provisioning under Resource Crunch
Authors:
Rafael B. R. Lourenco,
Massimo Tornatore,
Charles U. Martel,
Biswanath Mukherjee
Abstract:
Traditionally, networks operate at a small fraction of their capacities; however, recent technologies, such as Software-Defined Networking, may let operators run their networks harder (i.e., at higher utilization levels). Higher utilization can increase the network operator's revenue, but this gain comes at a cost: daily traffic fluctuations and failures might occasionally overload the network. We…
▽ More
Traditionally, networks operate at a small fraction of their capacities; however, recent technologies, such as Software-Defined Networking, may let operators run their networks harder (i.e., at higher utilization levels). Higher utilization can increase the network operator's revenue, but this gain comes at a cost: daily traffic fluctuations and failures might occasionally overload the network. We call such situations Resource Crunch. Dealing with Resource Crunch requires certain types of flexibility in the system. We focus on scenarios with flexible bandwidth requirements, e.g., some connections can tolerate lower bandwidth allocation. This may free capacity to provision new requests that would otherwise be blocked. For that, the network operator needs to make an informed decision, since reducing the bandwidth of a high-paying connection to allocate a low-value connection is not sensible. We propose a strategy to decide whether or not to provision a request (and which other connections to degrade) focusing on maximizing profits during Resource Crunch. To address this problem, we use an abstraction of the network state, called a Connection Adjacency Graph (CAG). We propose PROVISIONER, which integrates our CAG solution with an efficient Linear Program (LP). We compare our method to existing greedy approaches and to LP-only solutions, and show that our method outperforms them during Resource Crunch.
△ Less
Submitted 25 April, 2018; v1 submitted 29 September, 2017;
originally announced October 2017.
