arXiv subjects

Florian Schmidt

Publications and source records attributed to Florian Schmidt.

23 records · Page 2Linked to original sources

arXiv2018

BrainSlug: Transparent Acceleration of Deep Learning Through Depth-First Parallelism

Neural network frameworks such as PyTorch and TensorFlow are the workhorses of numerous machine learning applications ranging from object recognition to machine translation. While these frameworks are versatile and straightforward to use, the training of and inference in deep neural networks is resource (energy, compute, and memory) intensive. In contrast to recent works focusing on algorithmic enhancements, we introduce BrainSlug, a framework that transparently accelerates neural network workloads by changing the default layer-by-layer processing to a depth-first approach, reducing the amount of data required by the computations and thus improving the performance of the available hardware caches. BrainSlug achieves performance improvements of up to 41.1% on CPUs and 35.7% on GPUs. These optimizations come at zero cost to the user as they do not require hardware changes and only need tiny adjustments to the software.

cs.DC↗

arXiv2018

Representation Learning for Resource Usage Prediction

Creating a model of a computer system that can be used for tasks such as predicting future resource usage and detecting anomalies is a challenging problem. Most current systems rely on heuristics and overly simplistic assumptions about the workloads and system statistics. These heuristics are typically a one-size-fits-all solution so as to be applicable in a wide range of applications and systems environments. With this paper, we present our ongoing work of integrating systems telemetry ranging from standard resource usage statistics to kernel and library calls of applications into a machine learning model. Intuitively, such a ML model approximates, at any point in time, the state of a system and allows us to solve tasks such as resource usage prediction and anomaly detection. To achieve this goal, we leverage readily-available information that does not require any changes to the applications run on the system. We train recurrent neural networks to learn a model of the system under consideration. As a proof of concept, we train models specifically to predict future resource usage of running applications.

cs.DC↗

arXiv2016

Neural Document Embeddings for Intensive Care Patient Mortality Prediction

We present an automatic mortality prediction scheme based on the unstructured textual content of clinical notes. Proposing a convolutional document embedding approach, our empirical investigation using the MIMIC-III intensive care database shows significant performance gains compared to previously employed methods such as latent topic distributions or generic doc2vec embeddings. These improvements are especially pronounced for the difficult problem of post-discharge mortality prediction.

cs.CL↗

arXiv2014

HotBox: Testing Temperature Effects in Sensor Networks

Low-power wireless networks, especially in outside deployments, are exposed to a wide range of temperatures. The detrimental effect of high temperatures on communication quality is well known. To investigate these influences under controlled conditions, we present HotBox, a solution with the following properties: (1) It allows exposition of sensor motes to a wide range of temperatures with a high degree of accuracy. (2) It supports specifying exact spatial orientation of motes which, if not ensured, interferes with repeatable experiment setups. (3) It is reasonably easy to assemble by following the information (code, PCB schematics, hardware list and crafting instructions) available online, facilitating further use of the platforms by other researchers. After presenting HotBox, we will show its performance and prove its feasibility as evaluation platform by conducting several experiments. These experiments additionally provide further insight into the influence of temperature effects on communication performance in low-power wireless networks.

cs.NI↗

arXiv2013

Support for Error Tolerance in the Real-Time Transport Protocol

Streaming applications often tolerate bit errors in their received data well. This is contrasted by the enforcement of correctness of the packet headers and payload by network protocols. We investigate a solution for the Real-time Transport Protocol (RTP) that is tolerant to errors by accepting erroneous data. It passes potentially corrupted stream data payloads to the codecs. If errors occur in the header, our solution recovers from these by leveraging the known state and expected header values for each stream. The solution is fully receiver-based and incrementally deployable, and as such requires neither support from the sender nor changes to the RTP specification. Evaluations show that our header error recovery scheme can recover from almost all errors, with virtually no erroneous recoveries, up to bit error rates of about 10%.

cs.NI↗