Summary | Rationale | Images | Credits | Powerpoint Δ | Comments
| All Nuggets
Reliable computing in an unreliable world
Is it possible to perform reliable computations on noisy data in hostile computing environments?
Summary
Modern computational tasks have the property that the input data
are fundamentally noisy or that the computational environments in which
they are performed is unreliable. For example, in almost all
scientific domains, large amounts of data are routinely collected
using automated sensing modalities which often involve significant
measurement errors. Such data are subsequently analyzed using
various algorithmic tools in order to formulate or verify hypotheses
on the underlying system generating the data. Understanding the
sensitivity of existing algorithmic tools as the quality of the
input data degrades is necessary in order to guarantee the accuracy
of our conclusions. Such tasks become even more challenging as the
computational resources (computing devices, networks, etc.) become
unreliable. For example, while downloading large files in
peer-to-peer networks, many nodes leave the network prior to
download completion. Designing coding schemes and protocols that
reconcile partially downloaded files would permit better utilization
of the existing bandwidth.
Rationale
Put a slightly more detailed explanation here, which can be aimed at general computer scientists.
Contributors and Credits
Petros Drineas, Piotr Indyk, Sampath Kannan, James Lee, Luis Rademacher, Madhu Sudan
Image Ideas
List ideas for possible images. You can also upload images you've found using a command like this Δ.
Comments
- to give feedback on this nugget, just add another bullet to this list
- "hostile" in tagline sounds a bit extreme, and isn't really supported by the text in the summary. - Salil
- The transition from scientific computation in the beginning of the summary to coding near the end (via the sentence "Such tasks...") feels too abrupt, e.g. it's not clear what "Such tasks" refers to. - Salil
- Minor comment: "computational resources" -> "computational resources themselves" - Salil
- Add (local) error correction ? - Piotr
- I think, data measurement errors is a very minor aspect of the grand theme implied in the title. Our computing environment is rapidly evolving into a huge global network spanning scales from molecular to planetary and set to penetrate all aspects of life. The requirement of reliability of computing elements would be unrealistic on molecular scales. Requiring reliability of the environment and overall configuration would severely restrict global applications. It seems crucial to understand the implications of dropping such requirements. There are many tools that can assure self-stabilization, error-correction, synchronization, etc. under diverse conditions. However the greatly increased challenges require greater and deeper understanding of what can be assured on all scales from cellular automata levels to global environments including possibly hostile components. -Leonid Levin.
- It's a great nugget idea! I think it would be nice to give more examples (other than peer-to-peer file downloading), and mention what is already known (like error-correcting codes). -Valentine
