AI and machine learning have pervaded all parts of the world as we know it. While AI and machine learning are being increasingly used for natural or non-deterministic systems like cognitive recognition, perception, vision etc., we have seen various success stories in man-made, deterministic realms.
AI for man-made systems: We have applied machine learning for long in design, security and reliability to excellent results. In doing so, we have developed unique insights, triggered by the domains.
Firstly, we have found that it is possible to infer causality in a deterministic system with specific behaviors, like in a manufacturing or design system. In our experiments, we combined training data with a minimal amount of guidance from an “oracle” that knows the systems. We also omitted training data and only used oracular guidance, like in the case of the older expert systems. We found that in each case, it was possible to quickly converge and infer causality within a few iterations, despite the large sizes of these systems.
We have also found that the training data for these systems can be fine-tuned and improved on the basis of algorithmic, systemic guidance. Starting from completely random training data, we have shown that it is possible to iteratively refine training data to reach a succinct set of high-coverage training data.
Machine learning for community detection: We have found an efficient algorithm to navigate a large scale, massive heterogeneous knowledge graph and identify similarities in nodes and edges. This is the MAPR tool, that works on KnowEng, a knowledge graph with genomic information from high throughput experiments encoded in it. We are extending this tool to solve the problem of community detection in general knowledge graphs.
- Signature Pattern Covering via Local Greedy Algorithm and Pattern Shrink. Kim, H.; Im, S.; Abdelzaher, T. F.; Han, J.; Sheridan, D.; and Vasudevan, S. In 11th IEEE International Conference on Data Mining, ICDM 2011, Vancouver, BC, Canada, December 11-14, 2011, pages 330–339, 2011.
- Shobha Vasudevan, Samuel Hertz and Lingyi Liu, A Comparative Study of Assertion Mining Algorithms in GoldMine. Book chapter. To appear in Machine Learning Methods in VLSI Computer-Aided Design, Springer. Editors Abe Elfadel, Dunae Boning and Xin Li
- Lingyi Liu, David Sheridan, William Tuohy and Shobha Vasudevan, A Technique for Test Coverage Closure Using GoldMine. IEEE Trans. on CAD of Integrated Circuits and Systems (IEEE TCAD)31(5): 790-803 (2012)
- Lingyi Liu, Chen Hsuan Lin and Shobha Vasudevan, Word Level Feature Discovery to Enhance Quality of Assertion Mining, International Conference on Computer Aided Design ( ICCAD)2012: 210-217
- David Sheridan, Lingyi Liu, Hyungsul Kim and Shobha Vasudevan , A Coverage Guided Mining Approach for Automatic Generation of Succinct Assertions. In Proceedings of International Conference on VLSI Design (VLSI Design) 2014 . (Best paper award)
- Lingyi Liu, Xuanyu Zhong, Xiaotao Chen and Shobha Vasudevan, Diagnosing Root Causes of System Level Performance Violations, International Conference on Computer Aided Design(ICCAD) 2013: 295-302
- Lingyi Liu, David Sheridan, William Tuohy, Shobha Vasudevan Towards coverage closure: Using GoldMine assertions for generating design validation stimulus, Design Automation and Test in Europe (DATE) 2011: 173-178