Paper reading ntoes
J. R. Shewchuk, An Introduction to Conjugate Gradient Method without Agonizing Pain
T. M. Low, F. D. Igual, T. M. Smith, and E. S. Quintana-Orti, "Analytical Modeling Is Enough for High-Performance BLIS," ACM Trans. Math. Softw., vol. 43, no. 2, pp. 1-18, Aug. 2016.
This paper presents a high performance GEMM implementation with detailed performance model and analytical parameter derivation on multiple architectures. This work seems to show that high performance GEMM, which usually is provided by architecture experts or through hand-coded kernels and empirical parameter tuning, can be modeled quite accurately. The model also leads to parameters that compete with hand-tuned or empirical search. This paper sheds light on the obscure implementation of high performance GEMM on modern architecture.
Ming, J., Zhang, H., & Gao, D. (2011). Towards ground truthing observations in gray-box anomaly detection. In 2011 5th International Conference on Network and System Security (pp. 25–32). IEEE. http://doi.org/10.1109/ICNSS.2011.6059956
Anomaly detectors construct model of normal behavior of a program and detect deviations of program execution from such a model. Anomaly detector thus can detect zero-day exploitations without the knowledge of the exploit. Gray-box approach measn neither the source code nor the binary is analyzed. The sole information for the model comes from the execution (control flow and data flow) of the program. This work focuses on the data flow by monitoring arguments and return values of system calls. It then generates rules based on coinciding values of parameters and return values. A violation of the rule indicate abnormal execution. However the rule does not distinguish between true dependent values and coincidence. This work uses taint analysis to improve the truthfulness of such rules. The QEMU based TEMU dynamic taint analysis tool is utilized to generate taint information of system call arguments and return values. Equality on the condition of identical taint tags is therefore considered valid rule. The taint analysis works as follows. Upon first encounter of parameters or return values of a system call a new tag is associates with each of them. The TEMU follows the execution of the program and propagates the taint tags.
Gill, P. E., Golub, G. H., Murray, W., & Saunders, M. a. (1974). Methods for modifying matrix factorizations. Mathematics of Computation, 28(126), 505–505. http://doi.org/10.1090/S0025-5718-1974-0343558-6
This paper discusses efficient modifications to rank-1 perturbed matrix factorizations.
Luk, F. T. (1986). Algorithm-based Fault Tolerance for Parallel Matrix Equation Solvers. In K. Bromley & W. J. Miceli (Eds.), Real-Time Signal Processing (pp. 49–55). International Society for Optics and Photonics. http://doi.org/10.1117/12.949703
Luk, F. T., & Park, H. (1988). An Analysis of Algorithm-based Fault Tolerance Techniques. J. Parallel Distrib. Comput., 5(2), 172–184. http://doi.org/10.1016/0743-7315(88)90027-5
In this paper a very interesting viewpoint for ABFT is raised. Assuming a transient error strike during the matrix triangularization $A=ZU$. The interesting idea is that
Redinbo, G. R. (1998). Generalized algorithm-based fault tolerance: error correction via Kalman estimation. IEEE Transactions on Computers, 47(6), 639–655. http://doi.org/10.1109/12.689644
For a very illustrative tutorial on Kalman filter see this webpage
Introduction to Markov Modeling for Reliability, mathpages
An accessible introduction to Markov modeling for reliability under realistic repair conditions.