Immunizing exact analysis techniques against outlier contagion
Submitted by mernst on Wed, 2011-11-30 14:35
| Title | Immunizing exact analysis techniques against outlier contagion |
| Publication Type | Miscellaneous |
| Year of Publication | 2004 |
| Authors | Raz O, Ernst MD, Shaw M |
| Abstract | <p>Exact analysis techniques are often brittle because they make assumptions about input purity and cleanliness that are at odds with the real world. In particular, these analyses cannot handle noisy data, and they may not be sophisticated enough to handle either context-dependent behavior or complex application semantics. If the analyses could tolerate some noise and impurity, much of their power could extend to such real-world situations. </p> <p> The main result of this research is a procedure for adapting exact analyses to make them useful for analyzing impure data. We observe that the adaptation procedure yields useful results, though these results are understandably not perfect. We derive the adaptation procedure from experience with three different adaptations of a dynamic program analysis: (1) characterizing normal behavior in noisy data feeds, (2) detecting conditional program properties, and (3) handling modes in code. In addition, this research provides a characterization of alternative approaches to this type of adaptation and guidance for choosing an approach to match the impurity of concern. We justify the generalization not only by face validity but also by showing that an additional independent concrete adaptation follows the adaptation procedure. This additional adaptation extends the reports of static program analyses to handle false error reports.</p> |
| Citation Key | RazES2005 |
Last changed Mon, 2013-06-03 10:27

cs.